143 86 2MB
English Pages 192 [190] Year 2015
Contemporary Morphological Theories
To Nina, our dearest friend and companion. Always.
Contemporary Morphological Theories A User’s Guide
Thomas W. Stewart
© Thomas W. Stewart, 2016 Edinburgh University Press Ltd The Tun – Holyrood Road 12(2f) Jackson’s Entry Edinburgh EH8 8PJ www.euppublishing.com Typeset in 10/12pt Adobe Garamond by Servis Filmsetting Ltd, Stockport, Cheshire, and printed and bound in Great Britain by CPI Group (UK) Ltd, Croydon CR0 4YY A CIP record for this book is available from the British Library ISBN 978 0 7486 9267 5 (hardback) ISBN 978 0 7486 9269 9 (webready PDF) ISBN 978 0 7486 9268 2 (paperback) ISBN 978 0 7486 9270 5 (epub) The right of Thomas W. Stewart to be identified as the author of this work has been asserted in accordance with the Copyright, Designs and Patents Act 1988, and the Copyright and Related Rights Regulations 2003 (SI No. 2498).
Grateful acknowledgement is made for permission to reproduce material previously published elsewhere. Every effort has been made to trace the copyright holders, but if any have been inadvertently overlooked, the publisher will be pleased to make the necessary arrangements at the first opportunity.
Contents
Acknowledgements Abbreviations Foreword
viii ix xi
1 Opening the discussion1 1.1 Morphology: what, and why? 1 1.2 Theory and metatheory 3 1.3 The continua: how to interpret a table 4 1.3.1 Morpheme-based versus word/lexeme-based 5 1.3.2 Formalist versus functionalist 5 1.3.3 In-grammar versus in-lexicon 6 1.3.4 Phonological versus syntactic formalism 7 1.3.5 Incremental versus realisational 7 1.4 The road ahead 8 2 Theory profiles10 2.0 A brief prologue 10 2.1 A-Morphous Morphology 10 2.2 Autolexical Syntax/Automodular Grammar 15 2.3 Categorial Morphology 22 2.4 Construction Morphology 26 2.5 Distributed Morphology 32 2.6 Lexeme–Morpheme Base Morphology 38 2.7 Lexical Morphology and Phonology/Stratal Optimality Theory 42 2.8 Minimalist Morphology 47 2.9 Natural Morphology 52 2.10 The Network Model 59 2.11 Network Morphology 63 2.12 Paradigm Function Morphology 67 2.13 Prosodic Morphology 72 2.14 Word-based Morphology 76
vi Contemporary Morphological Theories
2.15 Word Syntax 2.16 Overview
81 84
3 Time for a test drive: putting descriptive frameworks through their paces90 3.0 Purpose of this chapter 90 3.1 Scottish Gaelic nouns: initial consonant mutation 90 3.1.1 A-Morphous Morphology 91 3.1.2 Autolexical Syntax/Automodular Grammar 93 3.1.3 Categorial Morphology 94 3.1.4 Construction Morphology 95 3.1.5 Distributed Morphology 97 3.1.6 Lexeme–Morpheme Base Morphology 98 3.1.7 Lexical Morphology and Phonology/ Stratal Optimality Theory 99 3.1.8 Minimalist Morphology 100 3.1.9 Natural Morphology 101 3.1.10 The Network Model 102 3.1.11 Network Morphology 104 3.1.12 Paradigm Function Morphology 105 3.1.13 Prosodic Morphology 107 3.1.14 Word-based Morphology 109 3.1.15 Word Syntax 110 3.1.16 What is at stake here? 111 3.2 Georgian verbs: agreement marker disjunctivity 112 3.2.1 A-Morphous Morphology 113 3.2.2 Autolexical Syntax/Automodular Grammar 114 3.2.3 Categorial Morphology 114 3.2.4 Construction Morphology 115 3.2.5 Distributed Morphology 118 3.2.6 Lexeme–Morpheme Base Morphology 119 3.2.7 Lexical Morphology and Phonology/ Stratal Optimality Theory 120 3.2.8 Minimalist Morphology 120 3.2.9 Natural Morphology 121 3.2.10 The Network Model 122 3.2.11 Network Morphology 122 3.2.12 Paradigm Function Morphology 123 3.2.13 Prosodic Morphology 125 3.2.14 Word-based Morphology 125 3.2.15 Word Syntax 126 3.2.16 What is at stake here? 126 3.3 Sanskrit gerunds: prefix–suffix interaction 128 3.3.1 A-Morphous Morphology 129
Contents vii
4
3.3.2 Autolexical Syntax/Automodular Grammar 3.3.3 Categorial Morphology 3.3.4 Construction Morphology 3.3.5 Distributed Morphology 3.3.6 Lexeme-Morpheme Base Morphology 3.3.7 Lexical Morphology and Phonology/ Stratal Optimality Theory 3.3.8 Minimalist Morphology 3.3.9 Natural Morphology 3.3.10 The Network Model 3.3.11 Network Morphology 3.3.12 Paradigm Function Morphology 3.3.13 Prosodic Morphology 3.3.14 Word-based Morphology 3.3.15 Word Syntax 3.3.16 What is at stake here?
130 131 132 132 133 135 136 137 138 139 141 143 144 145 146
Broadening the discussion151 4.0 The search for definition 151 4.1 Typology 152 4.2 Productivity 155 4.3 Fair comparison 160 4.4 Outro 161
References Index
164 175
Acknowledgements
In bringing this book up and out, there are many along the way whose generous support I want to acknowledge. Linguists whose mentorship and morphological research I have truly benefitted from include Brian Joseph, Arnold Zwicky, Greg Stump, and Rich Janda. Without doubt, it has been my great fortune to be in conversation with each one of them. A timely discussion with Andrew Spencer also provided a truly helpful dose of meta- cognition on matters of theory-partisanship. Gillian Leslie, Laura Williamson, and Richard Strachan, at Edinburgh University Press, have all extended to me their patience, encouragement, trust, and creative collaboration, for which I am very grateful. Last yet foremost, I thank my partner in life, Alma Kuhlemann, for sparing no resources to share with me her grasp of texts, as she inhabited at one and the same time the positions of critical reader and sensitive translator of resistant expressions. The cooperation of a number of publishers and authors is truly appreciated in the granting of reproduction permission for graphic elements in their copyrighted materials. Tables 2.3 and 2.4 reproduced by kind permission of Elsevier Publishing and Michael Hammond, volume editor. Tables 2.15, 2.17, and 2.22, as well as Figures 2.1 and 4.1, reproduced by kind permission of Cambridge University Press. Table 2.20 reproduced by kind permission of Oxford University Press. Tables 2.24 and 2.25, as well as Figure 2.10, reproduced by kind permission of Springer Publishing. Figures 2.2, 2.3, and 2.8 reproduced by kind permission of the respective authors. Figure 2.5 reproduced by kind permission of MIT Press. Figures 2.6 and 2.9 reproduced by kind permission of Mouton de Gruyter. Figure 4.2 reproduced by kind permission of John Benjamins Publishing.
Abbreviations
acc. accusative case AMR allomorphic morphological rule (in NM) AVM attribute-value matrix CG Categorial Grammar dat. dative case DM Distributed Morphology DS D-structure (< ‘Deep Structure’) EC Elsewhere Condition EWP Extended Word and Paradigm fem. feminine gender GB Government and Binding theory gen. genitive case ger. gerund GPFM Generalised Paradigm Function Morphology GPSG Generalised Phrase Structure Grammar HPSG Head-driven Phrase Structure Grammar IA Item and Arrangement IFD Identify Function Default instr. instrumental case IP Item and Process LC left-cancellation LF Logical Form LFT lexeme-formation template LM&P Lexical Morphology and Phonology LMBM Lexeme–Morpheme Base Morphology loc. locative case masc. masculine gender MM Minimalist Morphology MPR morphonological rule (in NM) MR morphological (spell-out) rule (in NM) MS Morphological Structure (in DM) Morphological Spelling (in LMBM)
x Contemporary Morphological Theories
MSR NM nom. ntr. OCP OT PF PFM pl. PR PV RC RHR SDSP sg. SN SS voc. WFR WP
morphosyntactic representation Natural Morphology nominative case neuter gender Obligatory Contour Principle Optimality Theory Phonological Form (in DM) Paradigm Function (in PFM) Paradigm Function Morphology plural number phonological rule (in NM) preverb-derived verb (Sanskrit, see §3.3) right-cancellation Right-hand Head Rule System-defining Structural Principles (in NM) singular number subject nominalisation S-structure (< ‘Surface Structure’) vocative case Word-Formation Rule Word and Paradigm
Foreword
The impetus for what has become this book is an ongoing fascination with the so- called ‘initial consonant mutations’ in Scottish Gaelic, ever since I first encountered them at Aberdeen University in 1987. In the years that followed, I have continued my research, trying to account for them in theoretical terms from several angles (e.g. Stewart 2004). Whereas the linguistic literature that approaches the initial mutations as one or more phonological phenomena struggles to explain the special grammatical determination of their distribution, those studies that seek to couch them in syntactic terms are required to augment the repertoire of categories and node-types in what seem to me rather ad hoc ways. For my part, however, I take the perspective that synchronically these initial consonant mutations in Gaelic – and indeed throughout Celtic – are to a large degree morphological in function and significance, and ought to be approached as such. The most commonly circulated approaches to morphology are not obviously disposed toward viewing non- concatenative processes/relations/objects such as alternations as truly morphological, because these demonstrably do not behave like affixes. Rather than force the issue by using a purely or primarily concatenative model that might enjoy wider familiarity, I have found it necessary to open up the field and to try to discover which of the other existing theories of morphology (if indeed any) are better suited for the description of the phenomenon in question. This task consisted in a critical comparison of a range of morphological theories, taking each on its own terms, to the degree possible. The proximate intention was to select a descriptive framework, but without weakening its basic claims or grafting onto it implausible machinery. This standard made it possible to focus above all on whether a given framework could do what I wanted it to do in a (potentially) convincing manner. This perspective also allowed me to consider individual theories independent of disciplinary politics, on the one hand, and without the orthogonal constraint of needing the operation of morphology to interface smoothly with any particular overall model of the grammar, on the other. This book presents and discusses a number of theories that have been proposed to address the issue of how to think and talk about word structure, in order to bring out each theory’s main claims and to test its implications through use on actual data sets. Although there are available some truly helpful introductory texts in morphology,
xii Contemporary Morphological Theories
there is to date no single integrative resource to get a person ‘up and running’ with the discourse of morphological theories that meets the following criteria: 1. It must consider, at least provisionally, that morphology is actually a subdiscipline of linguistics not wholly explained by some other subdiscipline; 2. it must reckon with a range of data beyond the affix-morpheme and/or the automatic sound-alternation; and 3. it must acknowledge that the professional literature in morphology comes with its assumptions already ‘baked in’, as it were, leaving logically possible alternative choices often rendered in something less than their full dimensions. It is important for a practitioner to be aware that a range of approaches exists, but in order to make a profitable choice (even one which may ultimately entail some ingenuity and enterprise on the user’s part), it is at least as important to know why any given theory takes the shape it does. The user of morphological theories is called upon to navigate the distinct emphases in the theoretical literature, to keep apart technical definitions and formal conventions that differ in sometimes maddeningly subtle, sometimes surprisingly radical ways, and to develop the sensibilities needed to infer those implicit assumptions that are so ingrained in the views of the respective theorists that they may simply never think to articulate what something as basic as a morpheme, for example, is to them. It is my hope that this book may serve anyone who is looking for a linguistic meta-language, or a new-to-them linguistic meta-language, within which to record observations, to distil predictions, and/or to tag collections of morphological data. Now, about the elephant on the cover. A traditional story that seems to have originated in India and subsequently spread through Jain, Hindu, Buddhist, and Sufi traditions (and well beyond, to be sure) tells of a scene in which six or so individuals encounter an elephant for the first time. Owing to blindness (or to lack of light, depending on the telling), each person has a different sensory experience of the animal. When they compare their assessments of the elephant, it turns out that each has taken a singular, partial experience to represent what is fundamental about ‘elephant-ness’. The argument that ensues has one person claiming that an elephant is like a rope (having touched only the tail), another that it is like a wall (having touched the side), and so on. The debate is always resolved in the moral that the whole is more varied and complex than any one perspective can capture. Morphology, I suggest, is like an elephant. At least in this regard.
Chapter 1
Opening the discussion
1.1 M O R P H O L O G Y : W H A T , A N D W H Y ? Fundamental disagreement about the definition of morphology is to be found among linguists, and consequently the very enterprise of morphological theoretical discourse is susceptible to misapprehension. Does the morphology of a language consist of the set of ‘constructions in which bound forms appear among the constituents’ (Bloomfield 1933: 207)? Is it ‘the study of the word formation processes of language’ (Siegel 1979: 12), or is morphology better conceived as simply ‘the syntax of words’ (Selkirk 1982; cf. Toman 1983)? Is morphology ‘the component of grammar that builds words out of pieces’ (Pinker 1999: 293), ‘the complex process by which abstract morphosyntactic representations are realised morphophonologically’ (Aronoff 1994: 9), or rather ‘the study of morphemes and their arrangements in forming words’ (Nida 1949: 1)? These descriptions seem to be surveying similar terrain, but in their reckoning they orient on topographical features of distinct orders, and as a result, the representations they produce not only bring borders into dispute, but also take irreconcilable views on fundamental issues such as cause and effect in word-formation. Rather than appeal to any single authoritative pronouncement as to what morphology is, it is perhaps more fruitful to consider what people who say they are studying or describing linguistic morphology do. The way into morphology is through the investigation of words, of their structure, and of their interrelationships. There are internal relations among the parts of complex words such as hyperactive or powerhouse that suggest that the parts are not necessarily equal partners in the complex. Here, act and house are recognisable subunits, serving as the central member in the respective complexes, and this is true despite the fact that the former may be conventionally (yet not uncontroversially) shortened to hyper (and not *active, much less the root *act) in informal use, and the addition of power-leads to a figurative interpretation of house. Therefore, the core or head member is structurally important in the process of building new, related words, but this core does not remain unaffected when it enters into construction with ‘junior’ partners. Words also relate to other words with which they share formal elements, forming paradigms when the related words are grammatically contrastive forms of a single
2 Contemporary Morphological Theories
abstract lexeme (roughly, a ‘dictionary headword’; cf. Matthews 1972: 160, 1991: 24–7), such that the English verb lexeme love subtends a cluster of related verb-forms: love, loves, loved, loving. All of these verb-forms share the same root and meaning, but they have distinct grammatical conditions for appropriate use, i.e. they are not interchangeable.1 Nearly all verbs in English have corresponding clusters of paradigmatically related forms, and indeed it is diagnostic of the category [Verb] to possess a full set of such forms. A partially analogous notion for other inter-word relations is that of co-derivatives, such as bluish, boyish, sheepish, roguish, etc., which serve to suggest the existence and status of a formative, here specifically an affix, and even more specifically a suffix -ish in English that is used in forming adjective lexemes out of morphologically simpler adjectives meaning ‘somewhat X’, and out of nouns meaning ‘resembling X in some characteristic fashion’. These co-derivatives do not, however, form a paradigm in that there is no grammatically required number of such co-derivatives that a formative must have, nor is there a principled limit as to how many co-derivatives a single formative may be deployed in building. There is a tension evident concerning the ontological status of words, however. In a fairly general sense, words are constituent elements in phrases and clauses, and in this context they are most usefully viewed as wholes. As we have seen above, in contrast, the presence of the elements of the internal structure of particular words – most neutrally, although rather clumsily, called subword units – is difficult, if not impossible to ignore. Morphologists act within this space of tension, attending to construction of words and at the same time to construction with words. What are words? How do you know when you have one? Is there a one-size-fits- all criterion by which a word can be recognised in any given language? Is it even a safe bet to say that all languages operate with a level corresponding to words? Dixon and Aikhenvald (2002), for example, present a wide-ranging typological discussion of different notions of the term ‘word’, particularly in light of languages where word-sized units relevant for phonological analysis do not coincide with word-sized units that grammatical analysis identifies. Their survey highlights points of potential confusion and proposes a battery of criteria for determining word-status for a given string in a given language. Although particular languages more readily support one basic unit type over another, Dixon and Aikhenvald conclude that it seems ‘likely that every language will have both a phonological word . . . and a grammatical word’ (32). These questions are notoriously difficult to resolve to everyone’s satisfaction, yet what constitutes the syntactically independent units of a given language seems intuitively clear to both observers and native speakers. Although each might carve up the speech stream differently, they will do so systematically and with great confidence as to where the ‘joints’ are. The alternative segmentations are merely indicative of the hierarchical and structured nature of language – the trained linguist or tutored second-language learner may have a tendency to break speech into relatively smaller chunks, whereas the native speaker may, especially in the absence of literacy in the language, perceive larger substrings as ‘the’ meaningful units of their language, and indeed might resist claims of finer-grained segmentability. Thus a
Opening the discussion 3
linguist’s morphological analysis will tend to go as far as possible, while the language user’s analysis will go only as far as practically necessary. A psycholinguistic or computational model of lexical processing stands in an intermediary position between (1) the full range of patterns that may be recognised, all else being equal, and (2) those distinctions that seem to matter in the successful (or typical) morphological operation of the user, human or otherwise. The question of what a particular theory of morphology is seeking to represent is likewise a source as well as a product of assumptions and hypotheses about the patterns within and among words, in dialogue with empirical evidence grounded in language behaviour. Therefore, the way we approach morphology is crucially informed by what we believe the relevant universe of discourse contains, the patterns into which its constituent entities are (or may plausibly be) arranged, and how said entities may interact with one another. Why morphology, then? Morphology inherently sits between sound and content, and so the territorial demarcation with adjacent subdisciplines in linguistic theory is not surprising. Nor has it been resolved. Spencer and Zwicky (1998) point out the trajectory of morphology’s status from standard component alongside phonology in American Structuralist grammars to a position clearly in the background in Generative linguistics, ‘largely eclipsed by phonology and syntax’ (1). In this latter environment, the sound-composition of word structure is attended to under a suitably enriched model of phonology, and the assembly of words out of morphemes is more similar to the operation of phrase-and clause-building. The resulting extensions of the respective domains towards a macro-phonology and a micro-syntax might collaborate in rendering morphology itself otiose, but only if the arbitrary linking of sound and meaning that words hold together can be shown to inhere in the grammar itself. A simplification in the architecture of grammar – namely, the deletion of a component – would represent a theoretical gain, all else being equal, but is linguistic theory better off with fewer, more powerful components in it? The present book illustrates some of the attempts to expand, compact, or otherwise delimit the work of representing word-formation and word-relation, attempts that reach beyond a programme of one or two leading ideas to achieve the breadth of a descriptive framework (Zwicky 1992: 328–9), suitable for use and testing with a range of empirical data. 1.2 T H E O R Y A N D M E T A T H E O R Y As with the acquisition of any abstract concept, e.g. ‘language’, one first must become aware that there exist other entities that could plausibly fall under that same heading. Without coming to realise the possibility of contrasting form with functional (near-)equivalence, one could remain locked in an initial state where, in the case of language, other ways of speaking are interpreted as versions of one’s own way, but erroneous to one degree or another. In the context of this book, then, conceptualising ‘morphological theory’ presupposes that there are at least two distinct, coherent, and usable systems available for the characterisation of words.
4 Contemporary Morphological Theories
Once one starts exploring, there are, in fact, multiple dimensions along which, and degrees to which, morphological theories may vary, such that ‘[l]aunching and navigating a theoretical ship involves far more than simply choosing a framework- type and nailing its flag to the mast’ (Pullum 2013: 516). When reading the respective literatures of existing morphological approaches, one may identify those portions of word-structure and word-relatedness that a given framework chooses to highlight. These focal phenomena are of course no accident, as an analyst will pursue and promote a description that he or she finds convincing, not deficient or self-defeating. In this connection, it might be surprising that a not insignificant amount of morphological theorising is in fact carried out by linguists who hold that there is no distinct morphological component in grammar, or that there is no lexicon per se, or that paradigms of inflected word-forms, for example, are epiphenomenal – suitable for second-language instruction or reference-grammar design, but not reflective of our knowledge or processing of language in any real way. There are, without doubt, counter-arguments for each of these positions (cf. Zwicky 1992). One of the guiding purposes in this book is to lay out for consideration some of the major metatheoretical options relevant to morphology. Individual theories may then be seen as realising particular constellations of theoretical choices, and the analyses and predictions that these theories produce are the ramifications of said choices. In what follows, the rationale for the presentation mode in Chapter 2 is described and placed in context. 1.3 T H E C O N T I N U A : H O W T O I N T E R P R E T A T A B L E A unique feature of this book is the use of a system of five continua that correspond to different ideas about how morphological structure is best analysed and presented. The simpler option of proposing binary [+/−] oppositions is tractable only in determining extremes, and indeed every extreme position turns out to be populated by two or more of the theoretical frameworks under consideration. The damage that insisting on binarity would do in the comparison of multiple theories – which may differ in subtle ways across potentially independent dimensions – is misrepresentative ‘lumping’ of close, but distinct, positions, on the one hand, and over-dramatised ‘splitting’, on the other, resulting in the implication of greater- than-actual incompatibility between non-identical frameworks. Broadly, then, in the interest of facilitating understanding, to start with, and collaboration, in due course, among morphologists, a subtler tool than the ‘us versus them’ dynamic of binary choices is preferable. Of course, once one ventures into the denser thickets of what languages can do with morphological structures, a continuum divided into five points of distinction as is employed here may seem insufficient. This format, however, is designed to give the reader relatively quick access to an overall orientation to each theory, informed by what the respective practitioners have claimed – and more importantly, what their analyses have shown – their enterprise to be about. In this way, pair-wise or group-wise comparisons of continuum tables can indicate:
Opening the discussion 5
1. points of agreement and/or conflict that may be observed or reasonably predicted, 2. types of morphological phenomena to which one or another theory may be most readily applied, and 3. frameworks that best match (or challenge, as the case may be) the reader’s own predispositions. Table 2.27, at the close of Chapter 2, brings together information from all of the theory-specific continua, in order to bring out areas of affinity among frameworks, with the larger-scale goal of identifying clustering and complementarity for further theory-building. 1.3.1 Morpheme-based versus word/lexeme-based The first continuum, morpheme-based versus word/lexeme-based, concerns the basic units around which morphological activity is assumed to be organised. Although the full articulation of the morpheme concept remains subject to shades of interpretation, the standard interpretation of a morpheme is the atomic meaningful unit of form. Thus, in a strongly morpheme-based theory, the ‘business’ of morphology consists in describing how morphologically complex expressions come to have the meanings and attributes they do, thanks to these morphemic units and the operations that manipulate them. Morphological analysis, therefore, is analysis down to, and up from, the level of constituent morphemes. Still within morpheme-based theories, however, there is at stake the status of other-than-prototypical morphemic behaviours: elements of form with no discernible meaning (empty morphs), contrastive meanings with no observable form (phonetically null, or so- called zero-morphs), units of form that correspond simultaneously to more than one meaning (fusional, or portmanteau morphs), among others (cf. Hockett 1947: 333).2 In a strongly word/lexeme-based theory, the word (subject to definition) is the organising principle of morphological structure. Analysis below the word level, especially that which takes derived bases back to source roots, is not ignored, but neither is it of primary concern to such a theory. Derived lexical items may owe some part of their lexical character (their semantics, grammatical category, or phonology) to their source roots, but in word/lexeme-based theories, the exhaustive analysis into parts is often (but not always) seen as an excess, a hyper-segmentation which goes beyond the requirements of syntax at least, since rules of (phrasal) syntax are generally presumed not to care about the internal constituency of the words they manipulate (the Lexicalist Hypothesis (Chomsky 1970)). 1.3.2 Formalist versus functionalist The second continuum, formalist versus functionalist, has to do with a broader perspective on what linguistic theory and analysis are supposed to accomplish. This
6 Contemporary Morphological Theories
distinction bears, therefore, on the types of phenomena one considers ‘linguistic’, what sorts of data constitute true counter-examples to theoretical claims, and what role, if any, extra-linguistic factors may play in the support of, or as evidence against, an analysis. Formalist approaches focus primarily on rules, constraints, and units which are particular to language structure, usually with the goal of capturing ‘all and only’ those generalisations relevant to the characterisation of linguistic competence. Functionalist approaches are interested more in contextualising language as cognitively and socially grounded behaviour. Functionalist analyses tend to be more tolerant of gradient behaviours, appealing to mechanisms such as constraint satisfaction and trade-offs, and to considerations of relative salience of entities and patterns, as this salience may be evaluated in terms of frequency, productivity, regularity, and so on. For these reasons in particular, functionalist discussion may draw formalist fire for being fuzzy, vague, and indeterminate. Formalist approaches receive criticism in turn for being artificially ‘neat’ in the data they consider, abstracting over variation, and backgrounding the language user as part and parcel of the language use equation. 1.3.3 In-grammar versus in-lexicon The third continuum, in-grammar versus in-lexicon, refers to the ‘location’ of morphology in the architecture of a grammar. Theories which place morphology in the grammar may do so as its own autonomous component or sometimes as distributed among independently motivated components, typically syntax and/or phonology (e.g. Ackema and Neeleman 2004). Much work in generative morphology has taken an ‘in-grammar’ approach to morphology, according little role to the lexicon, other than as a repository for idiosyncrasy (e.g. Di Sciullo and Williams 1987: 3). Ensconcing this view in their models of the grammar are Pinker (1999) and Clahsen (1999; cf. Wunderlich and Fabri 1995 and §2.8, below), who relate the difference between regular and irregular morphological patterns to a dual-route system for word-formation and lexical access. On these assumptions, whereas productive, regular morphological patterns are managed via grammatical rules, unproductive, irregular morphological patterns (up to and including suppletion) are represented as stipulations in the lexical entries of individual lexical items, i.e. they are memorised. An approach which puts morphology in the lexicon, on the other hand, has a very different perspective on just what the grammar does. The lexicon from this perspective is a repository for most if not all lexical knowledge, predictable or not, and the complex lexical entries interact with grammatical structures in as many distinct ways as grammatical structure requires (Jackendoff 1975; Bybee 1985; cf. Jensen and Stong-Jensen 1984). What the former approach gains in reducing redundant lexical listing, it loses in its failure to characterise inflectional paradigms, for example, in a natural way. The latter approach, on the other hand, presents the opposite problem, a rich and rather redundant lexicon, but an accordingly streamlined grammar. Issues of mental storage versus computation are relevant at this level, with computation being the focus of ‘in-grammar’, and storage the emphasis of ‘in-lexicon’. This is not
Opening the discussion 7
necessarily an either/or proposition, however, since it is possible, according to the Split Morphology Hypothesis, for example, to handle derivational morphology in the lexicon and inflection post-syntactically (Anderson 1982). Theories assuming some version of the Split Morphology Hypothesis are thus marked in the centre column of this continuum. 1.3.4 Phonological versus syntactic formalism The fourth continuum, Phonological formalism versus Syntactic formalism, is not entirely independent of the third continuum, but neither is it fully predictable from it. Approaches which place morphology in the grammar do, for consistency’s sake, tend to formalise morphological rules to be as similar as possible to the rules assumed for an adjacent component of grammar. Word Syntax (Lieber 1992; Di Sciullo and Williams 1987; §2.15), for example, makes frequent use of hierarchical tree structure and percolation of information among nodes, whereas Lexical Morphology and Phonology (Kiparsky 1982a; §2.7) formalises lexical and post- lexical phonological rules in similar ways, distinguishing them by domain of application, rather than by making a formal distinction in rule construction. Approaches which place morphology in the lexicon, yet none the less use a formalism akin to one customarily used with some other component of the grammar, may do so to facilitate exposition, but there exists less motivation on strictly theoretical grounds for such a practice. A mark in the centre column for this continuum indicates a rule format for morphology that is qualitatively distinct from those customarily found in phonology or syntax. 1.3.5 Incremental versus realisational Borrowing a terminological contrast from Stump (2001: 2–9), the fifth continuum, incremental versus realisational, focuses on the input/output conditions of the morphological component. A choice along this dimension will entail a very different picture of just what morphology ‘does’. In an incremental approach, the meaning and other attributes of morphologically complex expressions are built up gradually as a more or less additive process (thus ‘incremental’). From this perspective, every attribute or element of meaning not present in a lexical root must be added to that root in the morphology. This addition can happen, metaphorically speaking, either through the concatenation of morphemes or through the application of morphological rules. In a realisational approach, by contrast, the input to morphology is more abstract. A lexical base (whether root, lexeme, or lexical stem) and some set of morphosyntactic properties (appropriate both to that base and to the context in which the complex expression finds itself) jointly determine the morphophonological ‘spell-out’ of the fully inflected word in that context. Incremental methods are better motivated the closer the relation between form and meaning is to one-to-one. Where the overt morphology does not match so directly, uniquely, and reliably with the meanings and grammatical functions
8 Contemporary Morphological Theories
of the word as a whole, i.e. where the overt morphological marking either over- determines or under-determines the whole, an incremental approach will be forced into abstract, ad hoc elements in the analysis such as phonetically null (zero-) morph(eme)s, or rules which apply to introduce content but effect no discernible phonological change. There are many undesirable consequences of countenancing null elements in an analysis, even if their ‘distribution’ is in some way constrained, not the least of which is learnability – how does a person know a zero when s/he ‘hears’ one, and how does a person recognise which zero s/he ‘heard’? Realisational frameworks can, in principle, avoid the zero-morph trap because the association between a word and its morphosyntactic features effectively serves as ‘input’ to the morphology. In other words, the meaning ‘licenses’ the presence of particular exponents; the meaning does not ‘depend’ on the introduction of meaningful pieces. What an incremental approach gains in concreteness of representation, it loses when faced with un-morpheme-like behaviour. On the other hand, what the realisational perspective gains in formal versatility – namely, empirical coverage – it loses in its apparently unnecessarily complicated treatment of the most transparent morpheme- like instances of word-formation via concatenation. This descriptive power would not seem as problematic if edge affixation were no more common than other types of morphological marking. As it is, however, the disproportionate amount of concatenative morphology found cross-linguistically looks rather like an accident on the realisational approach. The system of continua adopted in this book certainly does not exhaust possible alternative theoretical distinctions that one could use in the classification of morphological theories. The chosen set of five, however, allows for some interesting similarities and differences to come out, and the dimensions are substantial enough that any given linguist can quickly identify the theory or theories which best match his or her own predispositions, as can be appreciated in Chapter 2. 1.4 T H E R O A D A H E A D Chapter 2 surveys fifteen current theories that address morphology: A-Morphous Morphology, Autolexical Syntax/Automodular Grammar, Categorial Morphology, Construction Morphology, Distributed Morphology, Lexeme–Morpheme Base Morphology, Lexical Morphology and Phonology/Stratal Optimality Theory (OT), Minimalist Morphology, Natural Morphology, the Network Model, Network Morphology, Paradigm Function Morphology, Prosodic Morphology, Word-based Morphology, and Word Syntax. Each of these approaches is presented and discussed with reference to the dimensions described in §1.3 above. In Chapter 3, each of the theoretical frameworks is brought to bear on data from three morphologically interesting languages: inflection of nouns in Scottish Gaelic (regular and productive non-concatenative morphology, §3.1), verb agreement in Georgian (disjunctive distribution of semantically compatible agreement markers, §3.2), and gerund formation in Sanskrit (inflectional suffix choice sensitive to
Opening the discussion 9
the presence of derivational prefixation, §3.3). The analytic methods and relative performance of each approach may suggest to you whether or not a given framework is well suited and satisfying for your morphological endeavours. Chapter 4, by way of conclusion, synthesises the presentation and findings, and points outward and forward to some areas of linguistic study where a thoughtful and coherent analysis of morphology can potentially shed light on apparent conundrums, and thereby improve research design and clarify results. QUESTIONS FOR CONSIDERATION 1. In many cases, investigators come to morphology out of a different linguistic subdiscipline. To what degree do one’s theoretical interests and predispositions predictably influence one’s approach to morphology? 2. By gravitating toward one end or the other of the proposed continua, a framework’s view of what constitutes important issues in word structure will shift. What phenomena are left in the background with each such shift of focus? 3. Imagine creating a theory that would settle into the middle column of every continuum: in other words, a ‘just right’ or ‘middle path’ theory of morphology. What would that theory look like, and what would such an approach leave out of consideration? NOTES 1. The forms of a lexeme need not in fact all be derived from a unitary root. For example, the paradigm of go has the formally plausible go, goes, going, and gone, but it is saddled with the apparent interloper went as a past form (a suppletive form, based on an etymologically distinct root). For maximum value as related linguistic signs, however, the form–meaning connection among the members of a given lexeme’s paradigm is best kept intact. 2. These points logically arise before the distribution of morpheme alternants (allomorphs) is addressed in any analysis. For this reason, Distributed Morphology (DM, §2.5) recasts the notion of morpheme as abstract – but nevertheless atomic – units in syntactic representations, manipulable as such by transformational rules before ever being assigned their respective phonological shapes. This forestalls a number of these issues of deviation from the one-to-one correspondence of form and meaning defining the classical morpheme.
Chapter 2
Theory profiles
2.0 A B R I E F P R O L O G U E After attending a public physics lecture on ‘observable quantities’ in Berlin in 1926, Albert Einstein called the speaker, Werner Heisenberg, aside and proceeded to offer him the following reality check: ‘whether you can observe a thing or not depends on which theory you use. It is the theory which decides what can be observed’ (Heisenberg 1968, reprinted in Salam 1990: 99). An analogous caution applies to our present labours of linguistic description. 2.1 A -M O R P H O U S M O R P H O L O G Y Table 2.1 Continuum table for A-Morphous Morphology. Morpheme-based Formalist In-grammar Phonological formalism Incremental
Ꮻ Ꮻ
Ꮻ Ꮻ Ꮻ
Word/Lexeme-based Functionalist In-lexicon Syntactic formalism Realisational
Many of the assumptions which coalesced in the form of Stephen R. Anderson’s (1992) book are laid out in an extended series of articles stretching back over at least fifteen years. After pointing out a resurgence of interest in morphology as a field of inquiry in linguistics after an extended drought, Anderson (1977) goes on to assert that ‘the notion of a separable morphological “component” is probably untenable’ (17). The name A-Morphous Morphology is intended to directly challenge the traditional role of the morpheme as a primitive in word structure, focusing instead on lexical roots or stems, and operations applied thereto.1 Taking seriously a fundamental asymmetry between lexical roots and morphological operations constitutes a disruptive shift from the perspective of morpheme-based morphology (cf. Aronoff 1976, 1994; Anderson 1988b: 162–4; 1992: 48–72; Zwicky 1992: 338). In A-Morphous Morphology (and its immediate predecessor, the Extended Word and Paradigm (EWP) framework, from which there is little discernible break),
Theory profiles 11
primary attention is given to inflection (Anderson 1977, 1988a, 1992: chs 4–6).2 In Anderson (1992), for example, derivation, cliticisation, and compounding each are allotted one chapter to inflection’s three chapters. The lexicon in A-Morphous Morphology is not the minimal ‘idiosyncraticon’ containing only the truly unpredictable facts about words (Zwicky 1992: 338) – a view inherited from Bloomfield (1933: 269), but rather it is here an un-list-like collection of linguistic knowledge that a speaker may have, governed by rules of varying generality (Anderson 1992: 183). Anderson takes the relevant word-like unit to be the stem (using the word lexeme rarely, if at all), derivation to be a lexicon-internal phenomenon (cf. LMBM §2.6 and PFM §2.12), and inflection to ‘fall “outside the lexicon” in the sense that [inflectional rules] represent knowledge not of particular words, but rather of the form taken by words as a consequence of the syntactic structure in which they appear’ (Anderson 1992: 183–4). The model of the grammar (see Anderson 1982: 594), then, entails the Split Morphology Hypothesis (Perlmutter 1988; Booij 1993; cf. Beard 1995), placing inflection effectively ‘in’ the syntax and derivation in the lexicon. An interesting perspective on the purported split may be gained through the comparison of the attributes of inflectional and derivational word-formation rules (WFRs) in A-Morphous Morphology (Anderson 1992: 123, 185): Inflectional WFRs are characterised by: a. A formal Structural Description, specifying conditions on the lexical stem in the input and conditions on the aspect(s) of the morphosyntactic context realised by the particular WFR; and b. A formal Structural Change, which may involve ‘not only affixation but also other phonological changes such as metathesis, substitution, deletion, etc.’ (123); whereas Derivational WFRs are characterised by: a. A formal Structural Description, specifying the class of input stems the rule can apply to and any additional conditions; b. A formal Structural Change, specifying the alteration the rule performs in creating the phonological form of the derived stem from the form of the input stem; c. A Syntactic Structural Description and Change; and d. A Semantic Structural Description and Change. Inflectional WFRs are not additive, or invasive, beyond the level of the phonological form. Because of the inflectional feature content of a particular syntactic terminal node, its associated morphosyntactic representation (MSR) licenses the introduction of inflectional exponents via inflectional WFRs, i.e. inflection presupposes the morphosyntactic representation, rather than creating it, as the metaphor goes in morpheme-based frameworks. Derivational WFRs can potentially effect a broader range of changes in the input, but this is done without reference to particular (morpho-)syntactic contexts.
12 Contemporary Morphological Theories
In contrast with the positions of DM (§2.5), LMBM (§2.6), and Word Syntax (§2.15), A- Morphous Morphology accords little significance to word- internal derivational hierarchical structure, since syntax appears not to need, or even have, access to that sort of information (the Lexicalist Hypothesis (Chomsky 1970)). The intricate relationship of inflection and syntax, however, leads Anderson (1992: 84) to conclude that the Lexicalist Hypothesis must be relaxed in inflection, although it may safely be assumed to hold in derivation. Compounding is a hybrid case in A-Morphous Morphology, because the ‘formation of compounds seems to involve a genuinely syntactic combination of lexical elements below the level of the word’ (Anderson 1992: 292), and thus there is motivation for a syntactically accessible hierarchical structure in headed compounds, in contrast with ordinary derivation. Probably the most noteworthy – and controversial – aspects of A-Morphous Morphology have to do with the implementation of inflectional WFRs to ensure that the ‘inflectional formatives of a word [are placed] in their correct relation to one another’ (Anderson 1992: 123). The null hypothesis is that no special ordering mechanisms will be required, and an unordered list of morphosyntactic features will be sufficient to direct the phonological realisation automatically. This cannot be the case, however, for two reasons: 1. one and the same inflected word may bear two or more distinct values for the same morphosyntactic feature (e.g. agreement in Person and/or Number for multiple arguments of a verb) (Anderson 1977: 23), and 2. of two or more contextually motivated inflectional rules, there are numerous cases cross- linguistically where only a subset of these rules actually apply, implying a disjunctive relation between particular rules (Anderson 1986: 7–8). Rather than a full conjunctive deployment of all applicable rules, or the more limited (but still reasonable) expectation that every motivated feature be realised at least once in the inflected form, the actual details of realisation require that some provision be made in an adequate grammar for rule ordering. In response to the first issue, Anderson (1977) proposes that words in syntactic contexts have MSRs (mentioned above), i.e. inflectional feature matrices whose contents are internally unordered by default, but which gain layers just in case ‘a rule of the grammar assigns features to an element, and that element already carries specifications for those features’ (21; see also Anderson 1992: 94). For example, an MSR with complex [−F +G], if further assigned the value [+F], will not unify to *[+F −F +G], but rather to the layered structure [+F [−F +G]], with any and all duplicate features (whether they bear contrastive values or not) appearing in an outer layer with respect to earlier-assigned and unduplicated feature-value pairs. Layering is in principle unlimited, but there seems to be no practical need for more than three layers in any one MSR. Similarly, there is no overt constraint that layering is limited to agreement (or so-called phi-/Φ-)features and so, to the extent that layering is not invoked except in cases of repeated or conflicting person, number, gender, case, or animacy specifications, this generalisation is missed. If layering is
Theory profiles 13
triggered during the sequential creation of an MSR (it must be sequential in order to determine, in cases of duplicate features, which instance is inner, and which outer), the inherent features of a possessed noun should be inner with respect to those of a possessor and, correspondingly, the internal arguments of a verb should be inner with respect to external arguments. In Anderson (1977: 21), a sketch of an alternative formalisation is made, provisionally positing more specific features such as ‘[±1st person possessor], [±plural possessor], etc., but this would be of little interest.’ This is true, certainly, and a fairly ad hoc response to the situation, but it is limited by the binary nature of features in A-Morphous Morphology (cf. n-ary features in, for example, PFM, §2.12 below). In the case of Georgian ‘inversion’ (e.g. Harris 1981, 1984; Anderson 1984), Anderson (1992: 141–56) proposes a ‘purely morphological transformation’ whereby an inner layer of the MSR is moved to an outer position. Thus, inflectional WFRs which happen to be keyed to particular layers (i.e. have particular layers specified as part of their Structural Description) will be effectively ‘tricked’ into applying to a different layer, producing the observed agreement marking mismatches (see §3.2 for some discussion). In order to ‘force’ features into particular layers, however, Anderson (1992: 147) invokes a purely formal dummy placeholder, apparently the only instance of a zero in A-Morphous Morphology. Such uses of the MSR device allow A-Morphous Morphology to engage in a measure of ‘virtual Relational Grammar’ (see Blake 1990), while technically avoiding a backward reach into syntax proper. On the issue of disjunctive rule ordering, A-Morphous Morphology relies on a version of the Elsewhere Condition (EC; Anderson (1969), Kiparsky (1973), not to mention Pān·ini). The Pān·inian Principle, often mistaken for the full EC, is a precedence principle, whereby the most narrowly defined of a set of competing rules precedes the other competitors in application, and thus rules may apply conjunctively or disjunctively and still respect the Pān·inian Principle (cf. PFM, §2.12). Anderson’s (1992) EC formulation includes a (weak) disjunctivity rider: ‘whenever one rule is more specific than another in the sense that the forms subject to the first constitute a proper subset of those subject to the second, the application of the more specific rule precludes the later application of the more general, less specific one’ (132; see also Anderson 1986: 4). Anderson (1992: 132, fn. 30) notes that this formulation entails disjunctive application only if the more specific rule applies, and furthermore applies first. Subtly, therefore, this EC allows most of the five logically possible outcomes of trying to apply two rules, a specific one S and a general one G (cf. Janda and Joseph 1992b: 255): −S, −G neither applicable −S, +G only G applies; S is not applicable (as opposed to blocked by EC) +S, −G only S applies; G, which must be applicable given the ‘proper subset’ definition above, is blocked by the EC +G, +S G applies first; S is not blocked by the EC, and therefore applies as well. This condition is claimed to account for disjunction between rules, between a stem and a rule, and between stems as well (Anderson 1986: 4, 1992: 133–4). This
14 Contemporary Morphological Theories
rincipled disjunctivity is not empirically justified (Janda and Sandoval 1984), p however, and so Anderson (1992: 129) adds not only the device of stipulated rule blocks (the rules within which blocks may, but need not, realise the same or similar inflectional properties), but also the option of extrinsic (ad hoc) rule ordering within these blocks, substantially weakening the predictive power of the account overall. The result is an observationally adequate description, but there is little insight into why the observed order obtains, rather than any number of readily describable alternative patterns. Also unaccounted for is the tendency for disjunctively related exponents to have similar if not identical linear placement restrictions with respect to the stem (cf. PFM, §2.12). Although it makes rather less use of the ‘word’ and the ‘paradigm’ than one might expect from a ‘word and paradigm’ (WP) type of theory, A-Morphous Morphology makes a number of important and provocative contributions in its denial of the relevance of the morpheme as a basic unit of language. A-Morphous Morphology borrows some trouble by adopting the Government and Binding (GB) approach to syntax, which is not particularly amenable to the A-Morphous innovations. Unlike DM (§2.5) and LMBM (§2.6), however, A-Morphous Morphology does not focus on the formal interface as much as the logical necessities such an interface would entail, and is therefore somewhat vague. Anderson borrows Chomsky’s (1981: 92) phrase, ‘merely a matter of execution’, on two occasions: (1) instead of taking a stand on whether lexical insertion should happen at D-Structure (DS) or S-Structure (SS) (Anderson 1992: 91, fn. 16), and (2) when tentatively considering whether Generalised Phrase Structure Grammar (GPSG; Gazdar et al. 1985) might not have a better account of agreement than GB (Anderson 1992: 109). Of course, this is a theory of morphology, and so some flexibility with respect to syntactic- theoretic detail is no great flaw. The architecture of grammar given in Anderson (1982: 594) is much more specific, and already in that article it is suggested that SS is the locus of ‘lexical insertion’ and that derivation was ‘in the lexicon’, but the diagram clearly includes ‘inflection’ in a component marked ‘Phonology’. This is misleading, however, because MSRs are present at SS, so ‘inflection’ here must refer to feature realisation, the application of inflectional WFRs. As an analytic tool, the formalism of A-Morphous Morphology is generally transparent, and accommodations are made for both affixal and more processual operations. Trouble spots are generally restricted to truly controversial areas (e.g. the EC and language-specific ordering). That compounds, clitics, and morphophonology are treated as well in Anderson (1992) is especially helpful, although it remains clear that Anderson’s answer to his own (1982) question, ‘Where’s morphology?’, is an ambivalent ‘everywhere, yet nowhere’ – that is, in no single place.
Theory profiles 15
Recommended reading for A-Morphous Morphology: Anderson (1977) Anderson (1982) Anderson (1984) Anderson (1986) Anderson (1988a) Anderson (1988b) Anderson (1992). 2.2 A U T O L E X I C A L S Y N T A X / A U T O M O D U L A R G R A M M A R Table 2.2 Continuum table for Autolexical Syntax/Automodular Grammar. Morpheme-based Formalist In-grammar Phonological formalism Incremental
Ꮻ
Ꮻ Ꮻ
Ꮻ Ꮻ
Word/Lexeme-based Functionalist In-lexicon Syntactic formalism Realisational
Autolexical Syntax directly addresses the interface between syntax and morphology. Sadock (e.g. 1988) has put forward a modular, non-serial, theory of grammar, in which semantic, syntactic, and morphological modules operate separately, yet simultaneously constrain the class of well-formed expressions in a language. In this way, a potential expression may be semantically interpretable, but not syntactically parsable, or vice versa, and in either case the expression would be ruled out. The same is true with respect to morphological structure. The suggestion is, then, that one can ‘troubleshoot’ any unacceptable expression and in doing so trace the source of the problem to one or more of the components. Sadock (1991) proposes a subsystem (not a module in the sense of Fodor (1983)), which he calls the Interface. This subsystem has ‘direct access to all varieties of grammatical information’ and uses this information to coordinate ‘the several representations produced by the autonomous modules’ (36). The lexicon is held to be a part of the Interface subsystem, and it, likewise, does not constitute a module in its own right: ‘it forms the axis around which the several autonomous modules pivot’ (29). A well-formed expression of any size in a language corresponds to a triple of acceptable output representations {rsyn, rsem, rmor} from the three components posited in this framework (20; cf. the triples in Categorial Morphology, §2.3 below). A lexical entry in Autolexical Syntax consists of a set of representations, one for each component, and these three representations define the grammatical use of the lexeme (30, 222 fn. 4). Under the rubric of lexemes, Sadock introduces entries for affixes, clitics, bound roots, and stems, including morphologically complex stems. The fact that entries are rendered in a comparable format, however, requires the reader/user to look carefully at each line in the entry to determine the o ntological status of the unit described.3
16 Contemporary Morphological Theories
Sadock (1988: 281) proposes a classification scheme for lexemes: Table 2.3 Lexeme classification scheme for Autolexical Syntax. (Source: Sadock 1988: 281)
Syntax Semantics Morphology
I
II
III
IV
V
VI
VII
VIII
+ + +
+ + −
+ − +
+ − −
− + +
− + −
− − +
− − −
where ‘+’ means ‘has a representation in that module’ and ‘−’ means ‘has no representation in that module’. Of these eight classes, class VIII is ruled out in principle as being empty in every regard, a victim of ‘intermodular suicide’ (281). Sadock identifies instances of classes I, III, V, and VII, i.e. those classes which have at least a morphological representation. Class I is exemplified by the ordinary lexical stem, such as dog, with a representation at all three levels: dog syntax = N[0] semantics = F−1 morphology = N[−0] Thus dog is syntactically a noun (bar-level 0), semantically a function of one variable, and morphologically a noun stem. Somewhat counterintuitively, the minus (−) on a morphological bar-level representation indicates the morphological domain of analysis. In this scheme, a greater negative integer does not mean a smaller morphological unit, but rather a larger one, such that [−0] is a stem, [−1] is a(n inflected) word, and [−2] is a ‘super-word’, i.e. a word plus an attached clitic element. Lexical stems and larger expressions – those ‘placed’ by the syntax – have a specific representation in the syntactic module. Class III is exemplified by pleonastic and purely functional elements like dummy it, infinitive to, and complementiser that (Sadock 1988: 280). Class V is a derivational affix, which is semantically a property expression (Prop), a function on the meaning of the stem, e.g. the German diminutive -chen, as in Bäumchen ‘little tree, sapling’, but which has no independent representation in the syntax (281–2): -chen syntax = nil semantics = Prop morphology = N[M1, ntr.]/N ‘M1’ refers to a particular morphological rule in Sadock (1988: 274) – namely, category-changing derivation on a stem – and the slash formalism is parallel to that used in Categorial Grammar (see §2.3 below). The ‘ntr.’ condition on the affix is a condition imposed on the output of affixing -chen to a stem: the result will be of
Theory profiles 17
neuter gender (compare Baum ‘tree’ (m.)). In this sense, Sadock claims, the affix is the head of the construction (akin to the ideas of Williams (1981), but with a more substantial, rather than positional, definition of ‘head’). Class VII is a stem- forming element like the -s in the non-head of certain German compounds, e.g. Freiheitskämpfer ‘freedom fighter’, where the -s is not simply the genitive marker (gen. Freiheit): -s syntax = nil semantics = nil morphology = N[M2, CF]/N[F] Such elements have no syntactic or semantic representation; they are present in the morphology only, for the creation/marking of a combining form (CF), a stem (formed by N lexemes of arbitrary subclass N[F]) for use in compounding only, by means of morphological rule 2 (M2; Sadock 1988: 247, 282). A more complete classification takes the five attested classes of lexemes and contrasts them in terms of the general formal content of their representations in the three modules (289): Table 2.4 Representations for lexeme classes. (Source: Sadock 1988: 289) Stem
Inflectional
Derivational
Incorporating Clitic
Syntax
X[0]
———
———
[Y[1]X[2]__]
[ZnX[2]__]
Semantics
Property or Relation
———
F(X[0])
F(X[2])
F(X[2])
[X[−1]X[−0]__]
[Y[−0]X[−0]__]
[Y[−0]X[−0]__]
[W[−2]Y[−1]__]
Morphology X[−0]
These formulations are deliberately abstract, the variables allowing for a range of instantiations for each type. Sadock (2012) represents a more detailed account of his model, with greater elaboration of the autonomous module architecture. The constituent elements are consistent with the earlier theory, but with more space given to logically and tactically distinguishable ‘types of information’ that a natural language expression may carry. Sadock presents the model as it has evolved as shown in Figure 2.1. Each of the proposed modules operates independently, with module-specific features and combinatoric rules, capable of rendering (if at all possible) a structural representation for a given natural language expression. The organisation of information as determined by one module’s grammar ‘need not correspond to its organisation in another, though there will be definite limits on the degree to which the various autonomous descriptions of a single natural-language expression can diverge from one another’ (1). This revised scenario splits the semantic module into function-argument and role structure modules, the syntactic module now handles constituency independent of a distinct module of linear order, and the
18 Contemporary Morphological Theories
function-argument structure role structure
morphological structure
Interface
linear order
syntactic structure morphophonology
Figure 2.1 Architecture of Automodular Grammar. (Source: Sadock 2012: 6, © Cambridge University Press)
orphological module now handles combination while the morphophonological m module governs the pronounceable dimension of expressions.4 The synchronising function of the ‘all-important’ Interface component5 ensures ‘a certain degree of compatibility . . . with respect to any pair of autonomous representations’ (Sadock 2012: 12, 24), and does so by means of the enforcement of three general (intermodular) principles of correspondence: 1. Lexical correspondence (inviolable): ‘If the value of a lexical item occurs n times in dimension Dj, then the dimensionally appropriate values of that lexical item must be present n times in every other dimension of analysis’ (25); 2. Category correspondence: by default, ‘certain categories in one module [. . .] match certain categories in another module’ (24), with deviations producing mismatches (for example, Latin deponent verbs such as arbitror, arbitrārī, arbitrātus sum ‘think, judge, observe’, which have active semantics
Theory profiles 19
but in most forms, including all finite forms, systematically bear exponents more typical of passive voice (cf. Baerman 2007)); and 3. Geometric correspondence: by default, ‘geometric relationships [e.g. dominance, c-command] between elements of an expression in one dimension . . . correspond to the same relationships between the corresponding values of those elements in any orthogonal dimension’ (Sadock 2012: 24). With these three principles holding sway over sets of representations, the degree to which the modules will bring forth truly incoherent sets of representations for the same concrete natural language expression is greatly limited. By not nesting levels of representation hierarchically, each within that ‘above’ it, as has customarily been assumed in generative models, Automodular Grammar (like Autolexical Syntax before it) does not entail that, for example, syntactic structure imposes hard constituent limits on morphological structure (Sadock 1991: 5–10, 51). Indeed, the sorts of phenomena that have motivated the Autolexical/ Automodular enterprise at all points in its development are those that give rise to questions of representational mismatch (e.g. clitics, noun incorporation, and other discrepancies between linear order and semantic scope). There is likewise an independence between morphological structure and morphophonological constituency (e.g. juncture phenomena, external sandhi, and syllabification of some clitics). Bearing specifically upon the Automodular version of the morphological module in its capacity as ‘the grammar of word structure’, the module ‘deals only with productive morphological facts’ (Sadock 2012: 148). Assuming two basic units – namely, the morphological word and a subword category stem, the basic morphological operations are functions from one unit-type to another (Sadock 2012: 148): Table 2.5 The four fundamental operations of the Morphological module. (Based on: Sadock 2012: 148) To From
Stem Word
Stem
Word
derivation derivational cliticisation
inflection cliticisation
The stem-based operations in this model comprise the traditional domains of morphology, and Sadock sets the word-based operations outside the discussion of the morphological model proper. The productive morphological processes that belong to the modules, rather than the lexical entries, are presented in the format of lexical rules that declare structural alterations in the corresponding specifications of lexical entries to which the rule applies; for example, the contrast between Latin agricola ‘farmer’ and servus ‘servant’ is handled by including in the lexical entries for the respective noun stems the information that is true for all their inflected forms (Sadock 2012: 151):
20 Contemporary Morphological Theories
agricol- syntax: N[masc.] F/A: Fa morph: Stem [N, DEC-1] mphon: / agrikol /
serv- syntax: N[masc.] F/A: Fa morph: Stem [N, DEC-2] mphon: / serv /
The paradigmatic relations that may be said to hold among the inflected forms of individual lexemes are here parcelled out among lexical rules that supply the information that a stem would need to be used as a full morphological word: nominative singular syntax: [nom., sg.] morph: Stem [N] mphon: 1. / . . . / 2. / . . . /
→ → →
Word [N, nom., sg.] / . . .a / in [[DEC-1], . . .] / . . .ʊs / in [[DEC-2], . . .]
The result of the Interface’s reconciling the lexical entries with the lexical rule produces fully specified and appropriately marked nominative singular words. This establishes the spirit of the approach, as it takes lexemes (and their stems) as basic, and accounts for morphological form and content separately at the level of modules, but linked via the intermodular correspondence principles. It would seem to cast the framework as more in line with Item and Process (IP) than with Item and Arrangement (IA) descriptive practice, and indeed Sadock sees ‘no particular difficulties’ in describing non-concatenative operations in the model, although the sole worked example is an edge- oriented subtractive pattern in Alabama verb stems, cleanly trimming stem-final segments, rather than altering or reordering processes that would require more morphophonological machinery (Sadock 2012: 153–4). Non- productive morphological patterns, by contrast, are consigned to the lexicon, where they stand as ‘generalizations over distinct lexical items’ (cf. MM §2.8 and the Network Model §2.10).6 The relationship between the structure of lexical entries and the six modules assumed in Automodular Grammar implies that full lexical entries are twice the size of those discussed in Autolexical Syntax. In practice, however, there is rarely the need to spell entries out completely, because the generalisations under discussion tend to touch multiple modules, but not all of them. In the case of English have (of possession), this generally transitive verb (syntax a) may in certain dialects also have the syntax of an auxiliary (i.e. invertable, negatable; syntax b). Furthermore, its somewhat idiosyncratic inflectional paradigm rates separate stipulated morphophonological subentries. The fairly comprehensive lexical entry proposed for this verb is as follows (Sadock 2012: 170; notes added): have (of possession) F/A: Faa RS: ‘have’, ANC, PAT
(combines with two arguments) (non-agent and patient macro-roles)
Theory profiles 21
syntax: a. V in [__, NP] b. %V[TNS] __ in [VP[V[TNS], __], NP] morphology: Stem[V] i. Stem in [__, PAST] mphon: / hæd / ii. Stem in [__, PRES, 3s] mphon: / hæz / iii. Stem in [__, PRES] mphon: / hæv /
(transitive syntax) (may pattern with auxiliaries) (projects a verb paradigm) (irregular past shape) (irregular 3sg. present shape) (present shape)
The level of detail provided permits one to know what structures may include instances of have, and corresponds to the information types that the Interface module would have to have access to in order to evaluate the sets of representations that the various modules would independently submit, so to speak, for an expression containing have. The present shape perhaps does not require stipulation as such, since in itself it is not irregular and is predictable from its root; it might be better cast as a default stem for a more generalisable description of verbs. Hypothetically, if this were a regular verb (where ‘regular’ corresponds to ‘follows the productive pattern with no need for stipulation’), there would be only / hæv / listed under mphon, with no further specification as to use and no alternate stems listed. A non-finite form such as having is unexceptional given a stem / hæv / – whether specified as [PRES] or not – as is the -ing form for all other (non-modal) verbs in English, and by virtue of this predictability, its omission in the entry is appropriate, to be handled by the corresponding productive rule in the morphology module. Owing to the correspondence principles above, many implicational relations between the features of different modules enable convenient expository under- specification in the ordinary case. Where there are no violations of the correspondence principles, there are correspondingly no representational mismatches, and in the end there is nothing for the framework to ‘do’. The management of structures is distributed to the corresponding dedicated and autonomous modules, none of which takes the output of another module as its input. It is both a strength of the approach and a perhaps misleading consequence of its elaboration that the preponderance of the Autolexical/Automodular literature is dedicated to reconciliation of mismatches. These ‘problems’ correspond to the figure that stands before the ground of smooth, unexceptional operation of the Interface module as it evaluates and coordinates compatible representations. Although the most salient cases for analysis in this modular theoretical framework are those that involve representations in conflict – in fact, a given natural-language expression may correspond to multiply discrepant structures without n ecessarily reaching ungrammaticality – the unmarked case of a morphological structure actually being consonant with its respective co-structures is not beyond comment. Given the multiplied opportunities for mismatches to arise in the Autolexical, let alone the Automodular, architecture, the degree and frequency of conformity across
22 Contemporary Morphological Theories
sets of independently calculated structures from the modules may stand more in need of explanation than discrepancy does (cf. Spencer 1993). Recommended reading for Autolexical Syntax Sadock (1985) Sadock (1988) Sadock (1991) Sadock (2012). 2.3 C A T E G O R I A L M O R P H O L O G Y Table 2.6 Continuum table for Categorial Morphology. Morpheme-based Formalist In-grammar Phonological formalism Incremental
Ꮻ
Ꮻ
Ꮻ Ꮻ
Ꮻ
Word/Lexeme-based Functionalist In-lexicon Syntactic formalism Realisational
The theoretical background for Categorial Morphology is to be found in the broader scope of the Categorial Grammar (CG) framework (see, for example, Karttunen 1986; Wasow 1993: 195–6; Steedman and Baldridge 2011), referring in turn to the category theory of Montague Grammar (Montague 1970, 1973; Partee 2001). A leading assumption of CG is that, in lieu of phrase structure rules, the ‘lexical entries of words encode virtually all the information’ relevant to the combination of said words into phrases (Karttunen 1986: 44). The theory posits syntactic categories that are either basic or derived, the former bearing atomic labels (e.g. NP, S; not to be confused with X0 syntactic heads) and the latter complex labels that express (1) the category of the node that may dominate them, (2) the category of the node of a required sister, and (3) the direction in which the sister may appear with respect to the labelled node in question. Thus the complex label S\NP corresponds to (a VP headed by) an intransitive verb, because with an NP sister to its left (\), the category forms an S. A determiner, on the other hand, would bear the complex category NP/ CN, indicating that in combination with a following common noun, a noun phrase is formed. In Categorial Morphology, this basic formalism is extended for word internal applications. Schmerling (1983) notes, in trying to gain ground for a specifically morphological treatment grounded in category-based structures and manipulations, that the prototypical CG perspective represented a framework assuming ‘distinct phonological, syntactic, and semantic systems, while invoking neither “morpheme” nor “levels”’ (222; cf. Montague 1970). Schmerling proposes that the core of a language consists of a set of expressions (A) and an indexed set of operations. The set A contains not only the basic expressions (i.e. morphological simplexes), but also
Theory profiles 23
‘all the expressions derived from these by repeated application of the operations; it contains nothing else’ (Schmerling 1983: 223). This modified version of Montague Grammar is characterised as a formalisation of the Item and Process (IP) approach to word formation (although the resulting picture of the lexicon is more populous than a morpheme-based theory typically requires). In the IP spirit, the work of morphology in this context focuses not on the position of discrete meaningful pieces but rather recursive category-assignment operations, separate from the words they participate in defining. A category-assigning rule is assumed to be tripartite, containing (1) the index of the operation employed in the rule, (2) the index of the input categories, and (3) the index of the output category of the rule (Schmerling 1983: 223–4). The idea of operations applying at the edges of expressions, of functors (the derived categories) taking arguments and together forming a larger expression of a (usually) distinct category, is the modus operandi of CG theory. Segment-level alternations, as word-internal operations, are assumed to be triggered by edge concatenation (226–7). This assumption is in trouble on empirical grounds for mutations and ablaut that are not triggered by elements outside the word in question, but rather are patterns that can have morphological significance in their own right, e.g. English man~men. Cases like these involve affixation only under remarkably abstract assumptions (see §3.1 below), and actually support Schmerling’s early argument that morpheme-as-thing (IA-type) analyses are unnecessarily limited. Her approach to portmanteau forms such as French du and au involves a substitution operation of the ‘amalgamated’ form for the sequence, de + le and à + le, respectively (228–30). Cliticisation is proposed to be handled similarly, through a substitution of the clitic group for the host (226). At a more concrete level, if any given operation is taken to be the equivalent of any other, then non-concatenative morphology is no different from concatenative morphology. If, on the other hand, we consider the relative power and latitude of a substitution operation as opposed to an operation which takes an expression as an argument and does something specifically to that expression, it seems that non- concatenative operations do not achieve equality in Schmerling’s model. The next major step in Categorial Morphology is Hoeksema’s (1985) dissertation. Written independently of the conclusions in Schmerling (1983), Hoeksema’s work acknowledges that there is more to morphology than edge affixation, but decides to forgo those complications until after a solid theory of Categorial concatenation is in place. Hoeksema takes the more conservative approach to Montague metatheory, whereby expressions are represented as triples: a phonological projection, a categorial projection, and a semantic projection. Formally (12): For every lexical entry L: L = Hoeksema (1985) does not focus on phonological details, so the phonological projection, where mentioned at all, is typically just the standard orthographic form of the expression in question. Similarly, the details of the semantic projection are left fairly under-specified – where necessary, the semantic projection takes the form
24 Contemporary Morphological Theories
of expressions of intensional logic (13). The categorial projection, Hoeksema’s (1985) true interest, is given significantly more detailed discussion. Basing the ‘word syntax’, as he puts it, on the general framework of Categorial Grammar, ‘the categorial representations will be members of the set defined by the recursive statement’ (13): X is a category if: (i) X is a member of the set of primitive categories PC (i.e. N, NP, and S); or (ii) X is of the form V/W, where V and W are categories; or (iii) X is of the form V\W, where V and W are categories. Now the primitive category set is truly minimal, and it entails some rather complex derived categories at times, e.g. (NP\S)/NP = transitive verb, i.e. an expression such that, if it finds an NP to its right, will form an expression NP\S, which in turn, if it finds an NP to its left, will form an S (17). The information is ‘in there’, but it takes some patient unpacking. Hoeksema (1985: 17–22) has a clear morphemic bias, since he defines one-place versus two-place operations, based on whether concatenation is involved (two- place) or not (one-place). Again, as with Schmerling, this makes concatenative and non-concatenative morphology qualitatively different. One-place operations include substitutions and zero-conversion (alias transpositions), whereas two-place operations include affixation and compounding (17–18). One-place operations are set aside almost entirely for the remainder of the book (subsequent chapters focus on compositionality and different types of compounding). It will be useful at this point to summarise the approach to affixation. The two-place operations employed in the Categorial Morphology of Hoeksema (1985: 19) are right- cancellation (RC) and left-cancellation (LC), common in Categorial Grammar: RC (A/B, B) = A
LC (A, A\B) = B
These operations, incorporated into lexical rule schemas of prefixation and suffixation, are as follows (19): Pref (v, w) = Suff (v, w) = Using these schemas, phonological projections are simply concatenated, categories are cancelled and resolved into new, derived categories, and semantic functors take scope over their arguments. This remains restricted to affixation, however. In Hoeksema and Janda (1988), now in light of both Schmerling (1983) and Hoeksema (1985), the basic Categorial Morphology formalism is presupposed. From the very first expository section, ‘Addition’, context sensitivity beyond the purely categorial is assumed. Prefixation and suffixation, jointly referred to as extrafixation, are the only even potentially context-free operations (Hoeksema and Janda 1988: 204). Addition operations which are context-sensitive may be sensitive to phonological properties of their arguments (e.g. phonological constraints on the English suffix -en in soften, tighten) or to prosodic constituents of varying sizes and qualities
Theory profiles 25
(e.g. consonants, vowels, clusters, syllables, stressed vowel/syllable, etc.). In the same way, infixes are regularly placed with reference to one of these categories, rather than with reference to a morpheme boundary per se. Infixes and certain clitics are generally placed just within the edges of expressions, and a mechanism proposed by Bach (1984), called ‘wrapping’, is invoked to handle these cases. The first step is to distinguish the first and last elements in a string from the non-first and the non-last, respectively: Let x be the string x1 . . . xn 1. FIRST (x) = x1 2. RREST (x) = x2 . . . xn 3. LAST (x) = xn 4. LREST (x) = x1 . . . xn-1 Once these basic operations are defined, the operations R[ight]WRAP and L[eft] WRAP can be defined in terms of them: RWRAP (x, y) = FIRST (x) y RREST (x) LWRAP (x, y) = LREST (x) y LAST (x) The disposition of y with respect to the discontinuous parts of x needs to be determined, especially in the case of clitics, but also prosodically in general for issues of syllabification or metrical foot assignment, for example, and so the further complex operations are defined (Hoeksema and Janda 1988: 209): i. ii. iii. iv.
LWRAP-pref (x, y) = (LREST (x) (y LAST (x))) LWRAP-suff (x, y) = ((LREST (x) y) LAST (x)) RWRAP-pref (x, y) = (FIRST (x) (y RREST (x))) RWRAP-suff (x, y) = ((FIRST (x) y) RREST (X))
This allows the placement of a morpheme in second position (iii and iv) or in penultimate position (i and ii), with prosodic or other dependency to the left (ii and iv) or to the right (i and iii). As may be seen from the above, Hoeksema and Janda (1988) are very much about responding to empirical challenges with independently motivated formal mechanisms in an enriched version of Categorial Grammar and (especially prosodic) phonology. As suggested by Schmerling (1983: 223), the operations in Hoeksema and Janda (1988: 212ff.) are indexed with respect to the level of analysis at which they apply (e.g. segments, syllables, words, phrases). The potential power of this indexation may be worrisome to some, but at least the levels mentioned are independently available in any general theory of grammar. A distinction between operations and the morphological rules which employ them is useful (cf. Zwicky 1987a), especially for cases where the same or very similar operations figure in multiple rules (German umlaut, Gaelic initial lenition; see Janda and Joseph (1986)). In this way also, a single rule may perform multiple operations, so as not to unnecessarily fragment operations which pattern together (cf. PFM, §2.12 below). Hoeksema and Janda (1988) raise a pair of related predictions which follow automatically from the formal nature of Categorial Morphology: ‘Rules that combine
26 Contemporary Morphological Theories
RWRAP and suffixation and rules that combine LWRAP and prefixation do not occur’ (213), and ‘Prefixation (suffixation) on level x is sensitive only to the properties of the leftmost (rightmost) constituent on that level’ (218). Fula consonant mutation would seem to challenge the latter prediction (Lieber 1992: 166): waa ‘monkey’ waa-ndu baa-ɗ i mbaa-kon
Class 11 Class 25 Class 6
Although these are otherwise apparently well-founded generalisations, it should be noted that they are both phrased with respect to extrafixation, despite the article’s explicit focus on process morphology. Categorial Morphology has a long and respected ancestry, although it has not particularly caught on outside of the company of practising Categorial grammarians. Since it is a challenge to motivate this approach without first motivating a Montague view of linguistic metatheory, there are some inevitable obstacles to the accessibility of an analysis cast in this framework. As Hoeksema and Janda (1988) show, however, there is room under the umbrella for more than concatenation (compare Word Syntax, §2.15 below), and this is clearly a(n unanticipated) bonus in empirical coverage. Recommended reading for Categorial Morphology Hoeksema (1985) Hoeksema and Janda (1988) Schmerling (1983). 2.4 C O N S T R U C T I O N M O R P H O L O G Y Table 2.7 Continuum table for Construction Morphology. Morpheme-based Formalist In-grammar Phonological formalism Incremental
Ꮻ
Ꮻ Ꮻ Ꮻ Ꮻ
Word/Lexeme-based Functionalist In-lexicon Syntactic formalism Realisational
The framework of Construction Morphology, as it has been developed by Booij (2010), takes on board a number of the fundamental assumptions found in theoretical approaches that have emerged under the rubric of ‘Construction Grammar’ (e.g. Fillmore et al. 1988; Goldberg 2006; Croft 2002), which in turn grow out from the context of Cognitive Linguistics (e.g. Croft and Cruse 2004; Evans and Green 2006). Construction Grammar in general holds a central position for its namesake, the construction, which links properties across phonological, syntactic, and semantic domains into symbolic units. Such units vary with respect to specificity, from fully
Theory profiles 27
specified, or substantive, in terms of all aspects of all constituent elements, to fully schematic, in which only the broadest abstract outline of the composition of the construction may be stated. In these terms, a conventional and somewhat opaque idiom like paint the town red ‘have an extravagant evening out’ is largely substantive, in that the verb may appear in any contextually appropriate finite or non-finite form (for example, we painted the town red, they could have been painting the town red), but otherwise the free substitution of any one of the individual words is severely limited, resulting in loss of the idiom’s specialised meaning (?colourv/?a/?suburb/?blue). At the other end of the continuum, a general construction such as subject-predicate is so under-specified that it (or something analogous to it) underlies a great proportion of clauses in a given language. Aside from the general functional requirements of each constituent element, however, an actual instantiation of this construction (a construct) can vary across the range of creative and more or less complex ways that clauses may be built with respect to syntactic category, word choice, internal constituency, and so on. Each step toward the substantive, then, can be seen as defining a subschema, which inherits all the attributes of the parent, and introduces one or more particular attributes that the specification in question conventionally brings with it. For example, the choice of an intransitive verb imposes certain limits upon the range of further constituents for the predicate. The concept of a construction as a form-meaning pairing has been used in Construction Grammar to emphasise the degree to which grammatical constructions and words may be placed on a continuum. On analogy with the term lexicon, constructicon has been offered as a more comprehensive set of constructions, to be distinguished ‘only with regard to their internal complexity’ (Michaelis and Lambrecht 1996: 216), with individual words constituting ‘the limiting case of a construction’ (Fillmore et al. 1988: 501). The scope of constructions at this lower limit is the whole word, with subword patterns being abstractable from them, rather than discrete and autonomous building blocks up from which words are constructed (Blevins 2006; see Word-based Morphology §2.14). Indeed, morphological constructions in this light can, like phrases, be fully specified (dog-s), partially schematic (noun-s), or fully schematic (noun-plural), or seen more hierarchically, dog-s instantiates and inherits from noun-s, which in turn instantiates and inherits from noun-plural. Construction Morphology is firmly word-based, locating bound subword elements as ‘part[s] of morphological schemas, and their meaning contribution is only accessible through the meaning of the morphological construction of which they form a part’ (Booij 2010: 15). The very idea of degrees of schematicity and the establishment of subtypes of (morphological) constructions suggests the adoption of a hierarchical lexicon (Flickinger 1987), and Construction Morphology assumes just such a mechanism. Theories of inheritance within hierarchies are of course relevant here, and in particular the choice of inheritance dynamic – monotonic (unchangeable once declared) versus default (potentially overridable in subtypes) – comes into play. Riehemann (2001: 274) prefers the former, at least provisionally, on the grounds that monotonicity in its stability aids learning of the hierarchy, despite in some cases requiring a greater number of subtypes
28 Contemporary Morphological Theories
overall to accommodate subpatterns. Booij (2010: 27–9), on the other hand, opts for default inheritance from mother to daughter to better reflect the fact that exceptionality of a subtype in one regard typically does not entail other exceptional behaviours. Booij (2010: 27) characterises the default model as ‘[a] crucial notion’ in that, when a child-type must simultaneously inherit from distinct mother-types and therefore may need to unify the feature sets, an incompatibility of feature specifications may be ignored or otherwise resolved, rather than failing to unify, as monotonicity would seem to imply. The discerning of a principled resolution hierarchy for conflicting features is ongoing. Gurevich (2006: 31) is agnostic between the two modes, but opts for default inheritance in her discussion of Georgian inflection, at least partly because it ‘is intuitively better for contrasting regular and irregular patterns’ (63). To illustrate inheritance, both Riehemann and Booij address the example of derived deverbal adjectives, German adjectives in -bar and Dutch adjectives in -baar, respectively, corresponding to English V-able adjectives such as abstractable. Riehemann (2001) adapts formal elements from Head-driven Phrase Structure Grammar (HPSG; Pollard and Sag 1994) including Attribute- Value Matrices (AVMs) for the representation of constraints on types. In separating out the various subtypes of German deverbal adjectives in -bar, Riehemann distinguishes a prototypical, fully productive, ‘regular’ type that is built upon a transitive verb (one which governs a direct object in the accusative case) as its base. reg-bar-adj PHONOLOGY
1
bar
trans-verb PHON MORPH-B
1 CAT | VAL | COMPS
SYNSEM | LOC
CONT | LZT
CATEGORY
VALENCE
ACT
4
HEAD
NP[acc] 2 : 5
UND 2
3
adj SUBJ COMPS
NP: 5 3
SYNSEM | LOCAL
CONTENT | LZT
possibility_rel ARG1
2
ARG2
4
Figure 2.2 AVM for the productive type of German deverbal adjectives in -bar. (Source: Riehemann 2001: 268)
Theory profiles 29
Other semi-productive subtypes are separated according to their subcategorisation frames, with certain miscellaneous outliers set aside for semantic or morphological reasons. bar-adj
poss-bar-adj trans-bar-adj reg-bar-adj
essbar ‘edible’
fruchtbar ‘fruitful’
dative-bar-adj
prep-bar-adj
intr-bar-adj
unentrinnbar ‘inescapable’
verfügbar ‘available’
brennbar ‘inflammable’
Figure 2.3 Hierarchy of German deverbal adjectives in –bar. (Source: Gurevich 2006: 31; cf. Riehemann 2001: 264)
Each adjective in the hierarchy carries as part of its phonology attribute the value bar concatenated at the end of the verb’s stem, formally: [phon 1 ⊕bar], drawing from a phon value inside a morphological base attribute [morph-b ]. In other words, the suffix is not seen here as a sign, as coming into these adjective constructs from a position of possessing its own lexical entry, but rather as being a constraint ‘liv[ing] only in the phonology’, and inherited as such by the types in the hierarchy (Riehemann 2001: 268–9). It follows from this that morphological exponence other than affixation is similarly to be handled in the phon attribute of the complex word. So-called zero-derivation amounts therefore to an alteration in other constraints with respect to a morphological base, but not in the value for phon. As an example of non- concatenative morphological marking, Riehemann (2001: 275) represents Hebrew derivation of instrument nouns (e.g. masreq ‘comb’) from verb roots (s-r-q ‘comb’) by individually indexing the consonants as a list value for a root attribute < 1 , 2 , 3 > and placing them with respect to a language(-family)-specific skeleton attribute which stipulates the schematic pattern . Together, root and skeleton provide the evaluation < 4 > of the attribute morph, which is structure-shared with the phon attribute of the whole word. In this way, the complex meaning of the whole pertains to the pattern as a whole, and the application to the context of combing to the unification of the root (sign) and the pattern (non-sign).7 In Booij’s presentation on Dutch derivational morphology, the schemas are represented in minimal sufficient detail (in comparison to some of the fuller AVMs in Riehemann’s approach). To represent the derivation of an adjective blus-baar ‘extinguishable’ from a verb blus ‘extinguish’, the word formation schema [V-baar]A is invoked (Booij 2010: 42–3). It is significant that the category
30 Contemporary Morphological Theories
A subscript pertains to the whole construction, rather than to the suffix, which itself is held to bear no syntactic category (cf. Williams 1981). Together with any formal (typically segmentable) correlates of a word formation schema, Booij’s presentation regularly takes care to establish a semantic representation of what the construction can be said to result in at a holistic level, expressing this informally and in terms of constituent meanings where relevant. For example, the construction schema responsible for Dutch deverbal adjectives in -baar carries through the action specified by the verb base, giving the fuller, but still basic, schema as follows: [ Vi - baar ]A ↔
[able to undergo semi]j
[ Ni - ise ]Vj
[action transforming someone or something into semi]j
Correspondingly, English denominal verbs in -ise would prompt the inference of the following schema: ↔
While there exist many exceptions to the above semantic characterisation (burglarise, winterise), and likewise to the N-as-base condition (equalise, brutalise), it remains a reliable core use of the pattern (fetishise, magnetise, diphthongise). The variation in attributes found in this word-family from a morpheme-based perspective would suggest that there are several homophonous suffixes - ise in English, and these may or may not be related in a systematic way. In a Construction Morphology account, however, the assumption that constructions are linked in inheritance hierarchies within which aspects of form and/or meaning are shared by default allows – and in fact obliges – one to collect relevant data first with an eye toward the most general, under-specified schema, perhaps simply [x-ise]V, as criterion. From the most comprehensive pool of examples thus developed, any major types and sometimes subtle subtypes may be sorted along categorial, semantic, and perhaps other morphological or phonological criteria. It is specifically the theory’s expectation of hierarchical relations among the subschemas of a construction that lifts this process above ad hoc taxonomy. The primary function of constructions in a morphological context is to express generalisations that structure the lexicon of existing words, and then, secondarily, to provide source patterns for analogical extension to new words (Riehemann 2001: 262; cf. Bybee 1995, §2.10). In relation to this, two further interconnections of morphological schemas may be articulated. The first involves the stating of relations between morphological schemas of equal complexity that have the same base for derivation (Dutch werk-er/werk-ster ‘worker (m./f.)’, padvind-er/ padvind-ster ‘boy/girl scout’), whether or not that base corresponds directly to an appropriate sense for inheritance purposes (e.g social-ism/social-ist), or that imply the existence of a base that does not in fact correspond to an existing word (English aggress-ion/aggress-ive, but *aggress is lacking; compare possess, etc.). Although more usually used for relations among the inflected forms of a common lexeme, Booij (2007: 36–7) terms such co-derivative relations paradigmatic and marks them in the following manner:
Theory profiles 31
< [Vi] ↔ actioni > ≈ < [Vi-er]Nk ↔ [subj of actioni]k > ≈ < [Vi-ster]Nj ↔ [female subj of actioni]j > < [X -ism]Ni ↔ semi > ≈ < [X -ist]Nj ↔ [person with property Y related to semi]j > < [Vi] ↔ actioni > ≈ < [Vi-ion] ↔ [concept or result of actioni]k > ≈ < [Vi-ive]Aj ↔ [showing characteristic of subj of actioni]j >
These paradigmatic relationships allow for the regular connection between -ism/-ist pairs to be registered with a schematic correspondence between morphological schemas, rather than deriving one complex from the other through an intermediate truncation rule (Booij 2013: 264). The status of bases formed by truncation, but not attested as words in their own right within the language under discussion, is questionable, as is making an arbitrary decision as to directionality of derivation, since the paradigmatic relationship could permit the coining of a novel instantiation to start at either end of the correspondence, or at opposite ends for different bases and/or in different speech communities. The second type of connection between distinct morphological schemas consists in what Booij (2010: 41–50) characterises as schema unification, a complex pattern that can give rise to new complex words via a ‘short-cut’, on the basis of a critical type-frequency of exemplars. For example, the English schema [un-X]X ‘not X’ is productive with verbs, adjectives, and adverbs, r esulting in a high frequency of appearance, while the schema [XV-able]A has its own broad distribution, as we have seen. These affixes furthermore freely co-occur in morphologically complex instantiations such as unbelievable and unforgettable. The suggestion is that, on the basis of frequent co-occurrence in concrete instantiations, a complex schema unifying the two independent schemas can be abstracted, in the above case: [un-A]A + [V-able]A = [un-[V-able]A]A When such a unified schema has been abstracted, one ‘may derive multiply complex adjectives of this type in one step from a verbal base’ without forcing the invocation of an intermediate adjective in -able that may not be attested widely, or perhaps not at all (Booij 2007: 38). This does not sidestep the issue of category-based hierarchy that morpheme-based frameworks tend to use in modelling word structure, but it permits an account of particular pockets of joint productivity that does not depend on the establishment of a frequent intermediate form in -able (or in un-) before the corresponding un-X-able can be coined. Toward this end, Gurevich (2006: 48) specifically makes ‘empirical evidence of productivity’ a necessary condition for positing any given schematic construction. Constructions may only be ‘inferred on the basis of multiple stored examples’ and, as such, they ‘provide a convenient locus for analogical extensions’. One would not expect, for this reason, a unification of the unproductive English schema [X-dom]N with any prefix whatsoever, regardless of that prefix’s own independent type-frequency.
32 Contemporary Morphological Theories
Recommended reading for Construction Morphology Booij (2007) Booij (2010) Gurevich (2006) Riehemann (1998) Riehemann (2001). 2.5 D I S T R I B U T E D M O R P H O L O G Y Table 2.8 Continuum table for Distributed Morphology. Morpheme-based Formalist In-grammar Phonological formalism Incremental
Ꮻ Ꮻ Ꮻ
Ꮻ Ꮻ
Word/Lexeme-based Functionalist In-lexicon Syntactic formalism Realisational
Proponents of Distributed Morphology (DM) have made it a point to offer basic introductions to their framework in the literature (Halle and Marantz 1993; Harley and Noyer [1999] 2003; Embick and Noyer 2007; Bobaljik 2012). This practice is in principle quite helpful for orienting researchers who are accustomed to assuming the existence of ‘words’ and a ‘lexicon’ thereof to a distinct ‘approach to morphology’ (Halle 1990; cf. Pullum and Zwicky 1991) that accords to neither of these concepts status as ‘privileged objects’ (Embick and Noyer 2007: 292). The overview given by Harley and Noyer ([1999] 2003: 464–5) foregrounds three distinguishing core properties for DM: 1. Late insertion, that is, syntactic terminal nodes are not associated with phonological content; only in the mapping from the output of syntax to the level of Phonological Form (PF) are phonological expressions (which DM terms Vocabulary items) inserted into the purely abstract syntactic structures. 2. Under-specification of said phonological expressions means that the vocabulary items themselves are not indexed for their possible insertion sites – forms are selected with reference to the relative complexity of their respective insertion contexts, mediated in general by principles of subset-based precedence (Halle and Marantz 1993: 123; cf. A-Morphous Morphology §2.1, LM&P §2.7, and PFM §2.12). 3. ‘Syntactic Hierarchical Structure All the Way Down’ refers to DM’s opting for ‘a single generative system [that] is responsible both for word structure and phrase structure’ (Embick and Noyer 2007: 290; cf. Word Syntax §2.15). In light of these assumptions, the concrete words that a given language actually uses are the realisations of hierarchical arrangements of nodes, determined in the
Theory profiles 33
first instance by the operation of narrow syntax. Once the work of constructing/ deriving the syntactic output, or S-structure, is complete, the relevant morphosyntactic heads in the tree structure authorise the presence and distribution of atomic abstract (form-free) morphemes in syntactic terminal nodes.8 To the degree that the resulting configurations do not correspond to the eventual, empirically verifiable distribution of inflectional exponence, however, manipulations of morphosyntactic heads are assumed to bridge any gaps or infelicities. The namesake claim of DM is that ‘the machinery of what traditionally has been called morphology is not concentrated in a single component of the grammar, but rather is distributed among several different components’ (Halle and Marantz 1993: 111–12; cf. Anderson 1992: 2). This would seem to leave ‘morphology’ fragmented and nearly epiphenomenal – a collection of effects attributable to the joint, and presumably minimally coordinated, functioning of universal and language-specific conditions on the independently motivated components of syntax and phonology. Correspondingly, at least rhetorically, under none of the proposed architectures in DM is there an autonomous module of morphology. There is nevertheless, in each succeeding model, a distinct region labelled as housing the morphological operations. In Halle (1990), a procedural architecture places a morphological component post-PF, but feeding into a phonological component. The words and morphemes of the Vocabulary are fed into the model before a trifurcation into syntactic representations. Vocabulary
Morphemes
Morphology
DS Readjustment
Words
SS
PF Spell-out
LF Phonology
Figure 2.4 Architecture of the grammar. (Source: Halle 1990: 157)
In the formal launch of DM, Halle and Marantz (1993: 114–15) introduced a representational level of Morphological Structure (MS),9 which is positioned between the output of syntax and a structural ‘lock-down’ enforced at the level of PF, in keeping with Minimalist principles of economy of movement. This model does not include explicit indications of information flow, and so appears less serial than the article’s accompanying discussion implies.
34 Contemporary Morphological Theories
DS (D-structure)
SS (S-structure)
LF (Logical Form)
MS (Morphological Structure)
PF (Phonological Form) Figure 2.5 Architecture of the grammar. (Source: Halle and Marantz 1993: 114, © Massachusetts Institute of Technology, by permission of the MIT Press)
Harley and Noyer ([1999] 2003: 465) elaborate on Halle and Marantz’s model, supplying access to three ‘lists’ of form and/or meaning elements that inform the derivational model – namely, the sets of morphosyntactic features that feed the operation of syntax, the Vocabulary items that are inserted into the PF representation, and the non-linguistic ‘world’ knowledge contained in the Encyclopedia, which allows for the reconciliation of PF and Logical Form (LF) at a ‘Conceptual Interface’, tentatively equated with ‘Meaning’. None of these lists, however, is intended to coincide with the concept of a generative lexicon (Harley and Noyer [1999] 2003: 466), but rather to replace it in distributed fashion, standing outside the model proper, feeding the right sort of information into the grammar at the appropriate moment (note the indications of information flow). The lists serve as a distributed alternative to what is seen as a generative or ‘computational’ lexicon (Marantz 1997: 203). For DM, the trade-off for removing the computational lexicon in this way is the rise of more specialised, albeit more numerous, apparata. Embick and Noyer (2007: 292) present an architectural diagram that more explicitly enacts a shift away from relying on an extension of syntax in advance of a PF deadline, and toward housing morphological operations, including Vocabulary insertion itself, within what may be termed a ‘long PF’ that terminates in the input to phonological implementation (Embick and Noyer 2007: 293, fn. 5). In line with the ‘Strong Minimalist Thesis’ (see Lasnik and Uriagereka 2005), the traditional levels and/or components of D- structure and S- structure are done away with here, and so the move of the morphological operations into an already-phonological but not-yet-articulated space is effectively obligatory (no S-structure, thus no MS). This change expands – and apparently integrates into
Theory profiles 35
List A: Morphosyntactic features
Syntactic operations (Merge, Move, Copy)
Morphological operations
Logical Form
Phonological Form (Vocabulary Insertion, Readjustment)
List B: Vocabulary items
Conceptual Interface (‘Meaning’)
List C: Encyclopedia Figure 2.6 Architecture of the grammar. (Source: Harley and Noyer 2003: 465 (simplified))
the reimagined PF – the sandbox within which discrepancies between what the syntactic derivation hands down and what is actually pronounced may be ironed out. The means for implementing any necessary amendments is common across all DM models – namely, a battery of manipulations including, but not limited to, the following: 1. morphological merger ‘joins terminal nodes under a category node of a head’ and thereby ‘forms a new word from heads of independent phrases’, e.g. the merger of tense specification (under a Tns head) with the main verb in English (Halle and Marantz 1993: 116; cf. Marantz 1988); 2. fusion unites two distinct functional heads standing in a sister relationship into a single terminal node bearing all the morphosyntactic features from the heads so fused, such that only a single Vocabulary item may be
36 Contemporary Morphological Theories
LISTS ACCESSED Syntactic terminals [Morphosyntactic features]
STAGES OF THE DERIVATION Syntactic derivation
Vocabulary
(Spell-out) Morphology PF
Encyclopedia
LF
(Interpretation)
Figure 2.7 Architecture of the grammar. (Adapted from: Embick and Noyer 2007: 292, 301)
inserted – namely, the most fully specified Vocabulary item that is not distinct from that feature set (Halle and Marantz 1993: 116; cf. Bobaljik 2012: 147–8); 3. fission separates off the realisation of compatible but as- yet- unrealised features from an inserted Vocabulary item and, after a subsidiary terminal node is generated in the structure, a less marked Vocabulary item is inserted in that node (Halle 1997: 432; compare the notion of extended (redundant) exponence of a single feature, ruled out as such in DM); 4. impoverishment rules delete semantically motivated morphosyntactic feature specifications of (abstract) morphemes ‘in the context of other morphemes’, just in case a particular feature specification lacks an overt exponent under statable co-occurrence conditions, simulating disjunctive or blocking- like effects at the allomorph level (Halle and Marantz 1993: 154; Embick and Noyer 2007: 310–14; see Halle 1997; see Bonet 1991; cf. Aronoff 1976: 43–5); and 5. readjustment rules are so-called morphologically-conditioned phonological rules that serve to ‘change the phonological form of morphemes after Vocabulary insertion’ (Halle and Marantz 1993: 158; see Pyatt 1997). The net effect of such operations can at times be dramatic indeed.10 The operations may function to unite what is underlyingly separate (morphological merger and fusion), split what is underlyingly unitary (fission), delete was is underlyingly present (impoverishment), and in the case of readjustment, alter, insert, or simply replace that which remains ‘out of order’ before moving to the stage of pronounceable output.
Theory profiles 37
In DM, therefore, it is important to arrive at the right number and positioning of abstract terminal nodes for the correct insertion of inflectional exponents. A suitable metaphor could be one of building structures to suit prospective residents (the morphosyntactic properties), then remodelling the units to permit cohabitations, separations, and so on. All the while, one knows ‘what to do’ because one knows ‘what is about to happen’. That is, the stimulus – and the check – on morphological operations posited in a DM derivation reside in the ultimate distribution of the phonological expressions realising the morphemes: one creates the map that reconstructs the pathway back to the assumed atomic nodes of syntactic-structural origin. Because in DM morphosyntactic features are attributes of terminal nodes only, stem selection is sensitive to the addition of particular (potentially phonetically null) affixes. The selection of a past stem rang, for example, is determined by the presence of a terminal node bearing the feature [+past] in which no overt morpheme is to be inserted. If [+past] were a feature of the head V, the appropriate stem could be selected without this appeal to inter-morpheme dependency, and the zero- morpheme could be dispensed with altogether. The assumptions of DM, however, entail that [+past] have its own node, and that this is separate from the stem node. Localising morphosyntactic features in syntactic terminal nodes as (or in conjunction with) functional heads allows DM to avoid the sort of ‘spell-out’ rules divorced from ‘independently motivated syntactic phenomena’ that are found in rule-based realisational theories (Halle 1990: 155; inferential, in Stump’s typology (2001: 1)). In exchange, however, the morpheme-based realisation in DM requires an intra- word hierarchical constituent structure that is not part of inferential alternatives (e.g. A-Morphous Morphology §2.1, PFM §2.12, or Word-based Morphology §2.14). Vocabulary insertion in DM is context-sensitive, only possible after the atomic terminal nodes have been resolved via any applicable morphological operations into the required content and position. Only at this point, therefore, can the context be identified with certainty and the correct morphemes even begin to compete for insertion. Competition among rules introducing phonological expression is truly relevant only among those Vocabulary items compatible with the insertion context, or more precisely, non-distinct from the features present in the terminal node. In DM, the criterion for precedence is appearance ‘in the most complex, most highly specified context’ (Halle and Marantz 1993: 123). A sample rule block is given here to demonstrate the contrast between conditioned allomorphy (the first three rules below) and context-free insertion (the remainder; Halle and Marantz 1993: 126): [+participle, +past] [+past] [+past] [+past] [+participle] [3sg.]
↔ /-n/ / X + ___ where X = hew, go, beat, . . . ↔ Ø / Y + ___ where Y = beat, drive, bind, sing, . . . ↔ /-t/ / Z + ___ where Z = dwell, buy, send, . . . ↔ /-d/ ↔ /-ing/ ↔ /-z/ ↔ Ø
38 Contemporary Morphological Theories
It may be noted that in this system, a phonetically null (zero) affix is insertable both by stipulation (to realise [+past] for an enumerated list of verbs)11 and context-free as an absolute default.12 DM rejects the technical notion of the disjunctive rule block as posited, for example, in A-Morphous Morphology (§2.1; Halle and Marantz 1993: 113). DM’s sets of rules are ordered first by the complexity of the feature(s) to be realised and then, in case of ‘ties’ such as the realisations of [+past] above, by the context of insertion (the verb-list membership). Should this intrinsic sequencing fail, rule application is further backed up by the option to impose extrinsic (i.e. not independently motivated) rule ordering, as needed, in order that ‘the correct output can be obtained’ (Halle and Marantz 1993: 120). In the case of English verb-inflection, the rules above are indeed ordered intrinsically, although the /-ing/ and /-z/ affixes are not in serious competition with the others. The condition on insertion that a morpheme not be ‘featurally distinct’ from the node into which it is to be inserted would technically allow the /-ing/ into the competition, but as a spurious competitor only. For DM, this seems an advantage, since the structure of the grammar needs no further special mechanism to block the insertion of inapplicable Vocabulary items. Recommended reading for Distributed Morphology Embick and Noyer (2007) Halle (1997) Halle and Marantz (1993) Harley and Noyer ([1999] 2003) www.ling.upenn.edu/~rnoyer/dm/ 2.6 L E X E M E – M O R P H E M E B A S E M O R P H O L O G Y Table 2.9 Continuum table for Lexeme–Morpheme Base Morphology. Morpheme-based Formalist In-grammar Phonological formalism Incremental
Ꮻ Ꮻ Ꮻ
Ꮻ
Ꮻ
Word/Lexeme-based Functionalist In-lexicon Syntactic formalism Realisational
Lexeme–Morpheme Base Morphology13 (LMBM), despite the limited scope suggested in its name, may be called a theory of language because of the role its originator sees for a morphological component: ‘All the borders between all linguistic modules [are] defined as morphological interfaces comprising algorithms which convert the representations of one module to those of the other’ (Beard 1995: 389). From this description, it might seem that LMBM would be a morphologist’s paradise, since it makes the grammar apparently morpho-centric. The theory is
Theory profiles 39
ambitious in that its implementation requires a revision of almost every traditional component of grammar, and just within morphology the model assumes the Separation Hypothesis (between content and form of elements) and the Split Morphology Hypothesis (derivation and inflection are distributed in different modules; cf. §2.1), but moving in the opposite direction, LMBM assumes a unified Morphological Spelling (MS) component that operates between the syntactic and phonological rule systems to realise all morphological marking, derivational and inflectional. The array of assumptions driving an LMBM analysis can represent a formidable learning curve, and so adopting the framework for a particular analysis is likely to run into resistance from those encountering the framework for the first time. In LMBM, a base component creates hierarchical structures which stand as general potential inputs to both the lexicon and the syntax. The content of such structures is some number of basic (underived) lexemes (defined as the major categories N, V, and Adj only). A subset of a putative universal set of forty-four basic grammatical functions are assigned to nodes in the structure (Beard 1995: 391–5; Table 2.10). Derived lexemes are created in the lexicon from the base-generated structures through an amalgamation metaphor, whether through head- to- head raising or through bracket erasure. If derivation is not selected, then every node in the base-generated structure must be accounted for (somehow filled, with a lexical or a functional syntactic head) according to the general rules of GB syntax. It is crucially important to note that the output of the syntax and the lexicon is quite abstract, and the only phonological content is the underlying phonological representations of the basic lexemes in the structure. Morphological information, by which is meant anything that is realised by bound morphology or closed-class free morphemes (including adpositions, pronominals, auxiliaries), is spelled out in the MS component. Ordering of affixes is determined based on the assumption that grammatical features in representations are ordered. Inherent features of the lexeme are spelled out first, then those of any derivational functions picked up in the lexicon, and finally any inflectional features which were acquired through the operation of syntactic rules. The MS component need not ‘see’ the layering of features: it is simply that the ordering is determined in up to three distinct stages, and the Affix Ordering Generalisation (inflection ‘outside’ of derivation) is consistent with this layering (whether or not this generalisation is empirically sustainable). Since all that the MS component gets as phonological input is the stem of the lexemes, it follows that morphological realisation proceeds from the ‘inside out’.14 There are several ways in which LMBM tries to ‘have it both ways’, theoretically speaking. In order to account for those aspects of structure which are shared between derivational morphology and syntax components, Beard strengthens the notion of the base component, which serves as the common input to the other two. In order to keep the effects which motivate the Split Morphology Hypothesis without losing the generalisation that many of the same sorts of marking processes are used in both inflection and derivation, LMBM posits the late-applying MS component, which
40 Contemporary Morphological Theories
formally implements all of the grammatical functions and features distributed in the lexicon and the syntax (the Integrated Spelling Hypothesis (Beard 1995: 101)). In this way, derivation and inflection are functionally distinct, but formally united. In LMBM, the notion of Case is redefined as a purely morphological notion (avoiding the overlap it has had in other frameworks with thematic roles and/ or hierarchical structure). Given the universal set of grammatical functions, these functions are expressed by various syntactic constructions and morphological markings. Because the relation between grammatical function and morphological Case is typically not one-to-one, Case is seen as a morphological means of spelling out, in part or in whole, grammatical functions (Beard 1995: 254). These grammatical (as opposed to semantic) functions serve a crucial role in LMBM, and so it is important for a practitioner of LMBM to accept the validity of the grammatical functions as a closed and universal set. Table 2.10 Primary and secondary grammatical functions. (Adapted from: Beard 1995: 391–5) Agent Patient Subject Object Possessivity Possession Measure Material Partitivity Distinction Absolute
Means Route Manner Ession Duration Iteration Accordance Purpose Exchange Cause Sociation
Location Goal Origin Inession Adession Anteriority Posteriority Superession Subession Transession Intermediacy
Prolation Proximity Opposition Perlation Circumession Termination Concession Distribution Exception Privation Thematicity
(White = primary functions; light grey = primary spatial functions; darker grey = secondary functions)
LMBM assumes that any nominal entity in a sentence bears one (or two) of the above functions. A nominal may bear two functions if one is primary (spatial) and the other secondary, e.g. [Goal[Posterior]] He went behind the camera (249). An important innovation in LMBM is the disposal of several syntactic categories. This change is entailed when grammatical morphemes, both free and bound, become part of morphological spelling and consequently do not require a structural position in syntactic trees. It has long been noticed that there are functional parallels between adpositions and Case marking. Beard takes this as an indication that the functional parallels motivate a formally unified treatment. The tradition of classifying adpositions as [−N, −V] lexical items, despite their closed-class status, has been misguided, according to Beard, who suggests that adpositions and Case marking co-operatively serve to identify grammatical functions of NPs, and thus that there are no syntactic PPs at all. This is a strong claim, with an at least prima facie counter-example in the Celtic so-called ‘inflected prepositions’ (cf. Stewart and Joseph 2009), but it does follow from the cross-linguistic distribution of adpositions with respect to Case marking (extended argumentation in Beard 1995: 229–77).
Theory profiles 41
LMBM maintains a strict distinction between abstract grammatical functions and the formal pieces involved in the realisation of those functions, i.e. the Separation Hypothesis. The separation in LMBM is more than just a logical conceit – the architecture of the grammar directly reflects this separation, since the grammatical functions are available even in the base component, but no phonological representations other than the stems of lexemes are available before reaching the (post-syntax, post-lexicon) MS component.15 It must be acknowledged that LMBM takes the spirit of the GB post-syntactic level of Phonological Form (PF) very seriously. LMBM finds itself caught between two goals: 1. to serve as a replacement to Word Syntax (§2.15) as a morphological interface with GB syntax, and 2. to remain true to the several ways in which LMBM architecture diverges from the GB architecture. The latter goal includes large and sweeping revisions – for example, the fact that the base component would replace D-structure; the grammatical functions would more than replace GB’s Case and Theta theories; and the reassignment of all function words (not merely affixes) to spell-out by the MS component would fundamentally change tree-structure representations. By translating grammatical functions into an abstract set, LMBM hopes to achieve cross-linguistic applicability in a way that theories which have a richer array of lexical categories and structural positions often do not. If the grammatical functions do indeed prove a viable approach, the focus of work in syntax and morphology would likely, almost necessarily, change extensively. There are some apparent logical problems of sequencing, such as having both a generative lexicon and a generative syntax, and the switching back and forth from one component to the other that sentence-building in LMBM would seem to require. There is also the apparent counter-modular need for the base to have access to the stock of lexemes in advance of submitting the base-generated output to the lexicon. Even though, according to Beard (1995: 361), the proposed meshing of LMBM with GB theory would require only ‘a modest adjustment’ in GB, LMBM has a distinct agenda as far as linguistic theory and investigation go. LMBM leaves an autonomous syntax with considerably less to work with than GB is accustomed to. Although Beard (1995) does occasionally allude to the speakers of languages in the theoretical exposition, the metaphors selected are more generally grounded in terms of the automatic implementation of systems of deductive algorithms, the mapping function between components that morphology serves. Rhetoric can get a little mystical sometimes: ‘The lexicon has two options . . . If the lexicon chooses the former tack . . .’ (339–40). Considerable thought has gone into both macro- and micro-level issues in developing LMBM, but the revolution in orientation that LMBM’s acceptance would require stands as a not insignificant obstacle to push through.
42 Contemporary Morphological Theories
Recommended reading for Lexeme–Morpheme Base Morphology Beard (1986) Beard (1987) Beard (1988) Beard (1995) Beard and Volpe (2005) www.facstaff.bucknell.edu/rbeard 2.7 L E X I C A L M O R P H O L O G Y A N D P H O N O L O G Y / STRATAL OPTIMALITY THEORY (OT) Table 2.11 Continuum table for Lexical Morphology and Phonology/ Stratal Optimality Theory (OT). Morpheme-based Formalist In-grammar Phonological formalism Incremental
Ꮻ Ꮻ Ꮻ Ꮻ
Ꮻ
Word/Lexeme-based Functionalist In-lexicon Syntactic formalism Realisational
The literature on Lexical Morphology and Phonology (LM&P) is at the same time rich and somewhat convoluted. It represents a convergence between a morphological approach (level ordering) and a phonological approach (rule strata) with similar but not always identical theoretical assumptions about causes and effects in morphophonology. The empirical motivation for this approach is the varying distribution of sound- structural patterns, from idiosyncratic alternations affecting one or a small group of derived lexical items to so-called ‘across the board’ generalisations. Whereas the former types of alternation are exceptional, and so can only be stated with considerable conditioning beyond the purely phonetic, the latter, more general sound patterns are closer to uncontroversially phonological rules. In recognition of the continuum of distribution, but in the hopes of reducing the machinery to something more binary, LM&P theorising offers a particular configuration of modules in the grammar, coupled with distinct modes of application within, and interaction between, said modules. This modelling focus follows from the concern for getting the surface facts right by means of positing as general a rule set as possible, or as unified an underlying representation in lexical entries as possible, or ideally, both of these at once. LM&P assumes that all but the most recalcitrant alternations are effected by a rule of some sort. The recurrent question when looking at a morphologically complex expression is ‘which came first?’. The linear order of affixes with respect to the root is taken to reflect the sequence of rule application. On the
Theory profiles 43
underived lexical items
level 1 morphology
level 1 phonology
level 2 morphology
level 2 phonology
level n morphology
level n phonology
syntax
post-lexical phonology
lexicon
Figure 2.8 Modular interaction in LM&P. (Source: Kiparsky 1982a: 4)
(controversial) assumption that all sound- structural rules are primarily (or entirely) phonological, the issue of modularity in grammar arises again and again. Morphological processes add material, and then phonological processes ‘iron out’ the discrepancies between what biuniqueness would predict and what actually occurs. LM&P assigns particular affixes to strata in the lexicon, and links rules to these strata on the basis of the sound-structural effects of their application and the linear location of these effects. Rules associated with a lexical stratum may apply multiple times – that is, cyclically – within that stratum, but it is not a necessary condition for lexical status that they do in fact reapply (Archangeli 1987: 6). Lexical rules apply in the lexicon as words are built, apply only in environments that include a morphological boundary in the structural description (so-called derived environments), and are subject to lexical exceptions, whereas post-lexical rules are exceptionless, general rules that are insensitive to morphological structure. Kiparsky prefers the more restrictive tactic of positing a minimum of strata in the absence of independent motivation. The alternative LM&P proposal of Halle and Mohanan
44 Contemporary Morphological Theories
(1985), however, imposes no a priori limit on the number of strata in an analysis, permitting descriptions with a single lexical stratum, but at the same time invoking no fewer than three lexical strata to describe Malayalam. An acknowledged forerunner of LM&P, Chomsky and Halle (1968) appealed to different sorts of boundary markers in the phonological representations (‘+’ for morpheme, ‘#’ for word), at least typographically on a par with phonemic segments, which phonological rules could refer to at no cost. The approaches of Siegel (1979) and Allen (1978) kept the boundary markers, but made them something that classes of morphemes were sensitive to, determining licit attachment sites, and thus creating level-ordering. LM&P replaces distinct boundary markers with distinct types of rule application (lexical and post-lexical), and posits distinct components in the grammar (which is held to include the lexicon) to oversee the proper application of the rules. In this way, rules are limited to strata within components, and the insertion of particular morphemes serves to trigger the application of certain lexical rules. In general, it can still be said that -ity in English is a ‘stratum 1 affix’, but this is only determinable indirectly, on the basis of the stratum 1 phonological rules which its insertion triggers, e.g. Trisyllabic Shortening and Obstruent Voicing: brief → brevity. Level- ordering foregrounds the assignment of affixes to strata, casting any attendant sound-structural changes as concomitants of morphological rules.16 The phonological concerns of most LM&P practitioners, however, not only shaped the framework’s descriptive formalism, but also led to a de-emphasis on those aspects of morphology that do not correlate with phonological alternations. Developing a synchronic derivation of song from sing via a level 1 lexical phonological rule of ablaut, for example, is a clear importation from diachronic developments. This rule is very limited indeed in its distribution in the lexicon of English, and it is not phonetically motivated; LM&P requires the synchronic grammar to be responsible for ‘creating’, or at least deriving, this correspondence nevertheless (Kiparsky 1982a: 12). In the case of Trisyllabic Shortening, more specifically a rule that ‘shortens a vowel if followed by at least two more vowels, of which the first is unstressed’ (Kiparsky 1982a: 35), this level 1 rule may be stated in a way that does not directly refer to morphological structure: V → [–long] / __C0ViC0Vj where Vi is not metrically strong.
This permits the rule to have its impact on morphologically complex words, in fact disproportionately so, because with affixation come more vowels, e.g. sane → s[æ]nity, declare → d[ɛ]claration. The rule as written is apparently general, referring only to its phonological aspect, but special measures must then be taken to prevent it from applying incorrectly to the many words which, all else being equal, should be subject to the rule as rendered. It is not the case that borrowed words and names should be immune from linguistic analysis, but the failure of English Trisyllabic Shortening or predominant stress patterns to apply to Amazon, Oberon, Menomini, Ticonderoga, or Goolagong is neither surprising nor conclusive. Kiparsky (1982a) warns with gravity17 against solving such problems of limited rule applicability by assuming that a given
Theory profiles 45
morphophonological process is an integral part of building morphological structure. He feels that such an appeal to the morphologisation of a putative lexical phonological rule is ‘the most unfortunate treatment of all’, in that it constitutes a claim ‘that there are as many “Trisyllabic Shortening” rules as there are suffixes that can trigger the process’ (39). This last statement clearly establishes LM&P as morpheme-based and incremental: it is at least an IP theory of morphology, if not IA&P. Zero-morphemes are tolerated in LM&P, creating derived environments that legitimise the application of lexical phonological rules that would otherwise have to be seen as applying to an edge, rather than spanning a morphological boundary. The use of such zeroes permits not only the formation of guideV → guideN (agent), but by means of Elsewhere Case preemption of a general rule by a specific rule (Pān·ini; Anderson 1969; Kiparsky 1973), the stipulation of guideV as one of the legitimate inputs for the rule suffixing -Ø also serves to block the creation of the more freely productive *guider (Kiparsky 1982a: 7).18 As in all such cases, there are learnability concerns with phonetically null elements, made the more acute in a blocking capacity. The theory as it is designed needs such elements to sustain lexical rules as not morphological in themselves. Booij and Rubach (1985) suggest that there is a further lexical component, a post-cyclic lexical rule block that, as the name implies, follows the application of all cyclic lexical rules, yet still participates in determining the shape of particular words, and therefore is distinct from the post-lexical rule block as well. This move places these authors in a position where they must posit functionally parallel rules in different components, a problem which they dismiss on the grounds that the amount repeated between such rules is not 100% (15–18). Making redundant formulation an all-or-nothing issue, however, is an innovation with Booij and Rubach, since generalisations can be lost in sometimes very subtle ways (cf. the meta-templates and meta-redundancy rules of Janda and Joseph 1992a). Booij and Rubach’s investigation of clitics in Dutch and Polish leads them to the claim that not only are clitics in the lexicon, but also they are affixed to bases in the lexicon, because they correlate with sometimes quirky alternations in the shape of the host (35ff.). The examples chosen to demonstrate this are from Dutch: namely, reduced phonetic forms of verb=clitic combinations, where the verbs in question are limited to high-frequency basic auxiliary-like vocabulary (‘have’, ‘can’, ‘be’, ‘must’, ‘will’, and the like), and the clitics in question are weak subject pronominals, such that, for example, underlying /heb-ik/ ‘have I’ reduces to [hɛk] in ‘casual and substandard varieties’ (35). Full lexical verbs (‘fall’, ‘run’) do not permit such reduction, and so whatever rule is posited in order to account for the reductions must be limited to this particular lexical domain and social register. Given what is known about the promiscuity of clitics vis-à-vis the distribution of affixes (note that Booij and Rubach (1985: 36) also include the reduction of /dat-ik/ ‘that I’ to [dɑk]), this means that there is an entry in the lexicon for the combination of every clitic and every potential host element in the language, a considerable expansion in some aspect of the lexicon. This claim is maintained despite their positing a separate operation of cliticisation in the syntax, for all and only those clitics that do not
46 Contemporary Morphological Theories
c orrelate with alternations in their hosts (i.e. those that are phonologically uninteresting) (50). Rather than unify cliticisation in one location, Booij and Rubach claim that there are lexicalised and non-lexicalised clitic–host combinations, correlating conveniently with alternating and transparent surface forms, respectively. Although the use of LM&P as such has dropped in recent years, it has seen a partial makeover of sorts in the context of the rise of Optimality Theory (Prince and Smolensky 2004), specifically in the form of what has be named Stratal OT (Bermúdez-Otero 1999; Bermúdez-Otero 2010/2014; Kiparsky 2000). The theory as developed retains the importance of cyclic application and stratal assignment, but since OT is a constraint-based programme, it is the constraints and their relative hierarchies that comprise the crucial units of analysis. In lieu of arbitrary numbers for rule strata, Stratal OT posits cycles of phrase-or postlexical-level, word-level, and stem-level phonology. Within each of these level-phonologies, constraints are independently ranked, and the three levels are evaluated in parallel. Bermúdez-Otero (2010/2014) clarifies that the distinction between LM&P and Stratal OT is more than one of translating formalisms, and that certain principles assumed as fundamental in LM&P, e.g. Strict Cyclicity, do not play a role in Stratal OT. The morphosyntactic structure in which a string to be analysed appears is likely to contain a number of nested domains, corresponding to the levels mentioned above: that is, phrase-level domains are triggered by phrase nodes, word-level domains by complete words, and stem-level domains are triggered by the stems used for derivation. Each of the domains is evaluated according to its level’s ranked constraints, and the benefit in phonological terms is that the issue of opaque surface forms is attributed to the interaction of the different level-phonologies, such that within any single level, no opacity may arise. Traditional, fully parallel OT models that posit unitary constraint rankings for particular languages are notable for encountering opacity hurdles that can only be solved with extraordinary measures that weaken the theory’s predictiveness by reducing the limits on its descriptive power (e.g. Output-Output (O/O) Faithfulness constraints, Sympathy theory; see Kiparsky 2000: 352–5). To the degree that Stratal OT permits the function-based segregation of constraint rankings (Bermúdez-Otero 1999: 72), the framework gains an advantage over standard OT analyses that do not have access to information concerning levels, and as a consequence must treat all the ranked constraints consistently throughout a particular analysis. With domains of different levels in the Stratal OT model, and with principled differences among these domains as to how they are identified, demarcated, and evaluated phonologically, some of the results left unexplained in LM&P – including any independent grammatical correlates of the levels assumed for the affix strata and corresponding lexical rule levels – may receive a more substantial motivation in the new regime. Recommended reading for Lexical Morphology and Phonology and for Stratal OT (items marked † pertain to Stratal OT only) Archangeli (1987) Bermúdez-Otero (1999)†
Theory profiles 47
Bermúdez-Otero (2010/2014)† Kiparsky (1982a) Kiparsky (1982b) Kiparsky (1985) Kiparsky (2000)†. 2.8 M I N I M A L I S T M O R P H O L O G Y Table 2.12 Continuum table for Minimalist Morphology. Morpheme-based Formalist In-grammar Phonological formalism Incremental
Ꮻ
Word/Lexeme-based Functionalist In-lexicon Syntactic formalism Realisational
Ꮻ Ꮻ Ꮻ Ꮻ
Claiming the spirit of Chomsky’s (1993) Minimalist Program as an inspiration, Wunderlich (1996) introduced an approach to morphology (primarily inflection) named Minimalist Morphology (MM). The two guiding general principles in the design of this framework are economy of entities, modules, and processes, and learnability, which discourages reference to abstract elements that cannot be directly observed (e.g. phonetically null, or zero-affixes) or facts that can only be determined by psycholinguistically unwieldy, large-scale comparison and contrast mechanisms (e.g. arbitrary inflectional classes). In MM, morphology is assumed to be a module distinct from syntax, and the job of morphology in this framework is to account for all and only the actually occurring word-forms of a given language. Toward this goal, MM proposes the partial architecture in Figure 2.9 (Wunderlich and Fabri 1995: 239). In this model, word-sized units pertain to structured lexical entries, such that base forms and any phonologically distinct stems related to said bases (up to and
structured lexical entries
phonological rules
wfc1
affixation
paradigms
wfc2
wfc3
Figure 2.9 The inflectional module of the grammar. wfc = well-formedness constraint. (Source: Wunderlich and Fabri 1995: 239, © De Gruyter)
48 Contemporary Morphological Theories
including suppletion, e.g. fully unpredictable inflectional pairs like go/went) are linked within lexeme- specific, non- monotonic inheritance hierarchies, and not derived by rule. Recurrent subpatterns appearing in multiple lexemes may be acknowledged within schematised ‘generalised lexical entry templates’ (Wunderlich 1996: 96), but these are precluded from morpheme status on a par with affixes by definition in MM, even in cases where a segmental or suprasegmental alternation could stand as a reliable predictor (exponent) of an inflectional feature value. In this regard, there is built into this theory a convenient fiction – namely, that the concepts of regularity, productivity, and concatenation are all of a piece. Furthermore, as a corollary to this, it is assumed that non-concatenative patterns always and only constitute unproductive subpatterns. For the German and English examples in the MM literature at least (Clahsen 1999; see also in this connection Pinker and Prince 1994), this merger of independent notions is quite safe, even obvious (e.g. as seen in English non- concatenative, unproductive, irregular verb inflection pairs such as ride/rode, fight/ fought), but there are in fact clear empirical challenges to linking these terms (see §3.1). Nevertheless, the segregation of the different types of morphological marking is institutionalised in MM via the qualitative difference between the structured lexicon and the subcomponent of affixation. Affixes are held to have their own lexical entries, marked with an ad hoc feature [+min]. The rules adding these affixes to lexical stems, together with the semantic/ grammatical content of these affixes, e.g. the German 2nd-person agreement marker -st for verbs, may be represented in the following declarative format (Wunderlich and Fabri 1995: 262): /st/; [+min]; +2 / +V Exponent [PHON] Output [CAT] Input [SUBCAT] Affixation in MM is a strictly information-adding proposal, such that an affix may add only ‘+’ valued feature specifications. Any ‘−’ values in a representation of any complexity are the result of default insertion in the absence of a ‘+’ value, or in other words, an affix may not be explicitly specified as producing a [−2] output, for example, even if the affix in question correlates with all and only 1st-and 3rd-person arguments (262). The structured lexicon, affixation, and paradigm subcomponents are each subject to well-formedness constraints (wfc1, 2, and 3, respectively, in Figure 2.9), which ensure that the otherwise free operation of concatenation and the concomitant morphological feature co-occurrences that this engenders do not violate any universal, logical/semantic, or language-specific conditions. As can be seen, all paths in the inflectional component lead to paradigms, which are identified in MM as precisely the interface between morphology and syntax. In this way, MM is explicitly an ‘early insertion’ theory, with fully inflected forms feeding syntax, rather than morphology receiving assembly instructions from the syntax (Wunderlich 1996: 102; cf. DM’s ‘late insertion’, §2.5). Some basic entries from structured lexicons are illustrated in the Figure 2.10 (96). These verb stem sets show form changes through substitution and/or s tem-forming
Theory profiles 49
German strong verb with several formally and functionally distinct stems: {werf, wirf, warf, würfe, worfn} ‘throw’ [ v r f ]+V
[ . I . . ]–1
[ . a . . ]+pret
[ . . . . ]imp
[ . y . . X ]+conj
[ . o . . n ]+part
English irregular verbs: drive (vowel change and suffixation) [ d r aj v ]+V
[ . . ow . ]+pret
[ . . I . n ]+part
drink (vowel change only) [ d r I ŋ k ]+V
[ . . . . ]+part V
[ . . æ . . ]+pret
put (no formal change)
[p
t ]+V
[ . . . ]+pret
+part
Figure 2.10 Example hierarchical lexical entries. (Source: Wunderlich 1996: 96)
50 Contemporary Morphological Theories
augmentation. Each subordinate stem here inherits the category [+V] from the base form (the root node of the respective tree), and is further specialised to fulfil one or more particular inflectional functions. Note that in the case of put, the subordinate stem inherits both its full phonological shape and the [+V] category from the base form, and it is specified to fulfil the disjunction of [+pret +part]. Thus all required inflectional stems are furnished without recourse to zero-affixation (respecting learnability). Although the upper boundary of functions that may be chained in a single disjunction is not stipulated, it need not be, owing to the default mechanism that falls out naturally from inheritance trees. This assumption of unstated feature values is embraced in MM in connection with a clear preference for under-specification whenever possible. Minimalist Morphology aims to keep the mechanics of morphological formation streamlined, and to balance out the desired freedom of application for affixation, which would otherwise overgenerate (Wunderlich 1996: 94, 96), there are a number of constraints that ensure desired outcomes (i.e. actually attested forms in paradigms) without recourse to stipulating ‘winning’ candidates. Among the principles adduced in MM are Principles of Affixation (97): Monotonicity: The output of affixation must be more informative than the input [thus neither semantically/functionally empty nor impoverishing (feature-deleting) affixation]. Adjacency: The input requirement of affixes must be met locally. Affix Order: The order of affixes must conform to the hierarchy of functional categories, i.e. affixes that express lower-ranked categories must be attached first, relatively closer to the verb stem. In this last principle resides the functionalist underpinning of this framework: the hierarchy of functional categories to which MM subscribes is both semantically based and hierarchically ranked, each category possessed of marked [+] and unmarked [−] values. The hierarchy is hypothesised to be universal, and to hold sway in both inflectional morphology and syntax (Wunderlich and Fabri 1995: 247; Wunderlich 1996: 98; cf. the Mirror Principle of Baker (1985), a version of which is generally assumed in DM §2.5 and Word Syntax §2.15). The sequence proposed for verb-forms in particular is the following: Outer Inner C[omplementiser] > person > number > gender > mood > tense > aspect > voice (> verb) (Wunderlich and Fabri 1995: 246–51; cf. Bybee’s (1985) Relevance Hierarchy, see §2.10 below).19 Deviations from the sequence above may emerge in the course of diachronic change, but the alteration is predicted to be temporary and under semantic and systemic pressure to return to the universal order (Wunderlich and Fabri 1995: 249; Wunderlich 1996: 97–98). What is more, only adjacent categories in this hierarchy are subject to fusion such that they are, or diachronically come to be, expressed by one and the same exponent (98). MM places the notion of inflectional paradigm in a central, organising position,
Theory profiles 51
mediating the operation of inflectional morphology and the projection of inflected morphological objects into the syntax (102–4). Whereas frameworks that assume late (post-syntactic or otherwise syntactically driven) insertion of lexical items typically have no need for paradigms as such in the grammar, and shunt such structures aside as descriptive or pedagogical epiphenomena, MM’s paradigms are templates that serve to allow, from among the multitude of potential morphological objects that may be constructed from the structured lexical entries and the affixes of a given language, just those inflected forms that actually enter into well-formed syntactic constructions. As mentioned previously, the free operation of affixation would overgenerate word-forms beyond necessity, and the solution to this excess production adopted by MM is the mechanism of paradigm construction. That is to say, MM does not simply assume that paradigms are received, unquestionable lists of forms, collections of ‘winning’ candidates that could be selected by stipulation just as easily as by principled competition. Rather, MM posits an ‘affix-driven’ construction model that not only determines the feature combinations that inform the number and content of cells in paradigms, but also takes into consideration the distribution of particular affixes and stems. The precise range of outputs created (via positive ‘+’ specification) or allowed (via under-specification) by each lexical item (stem or affix) implicated in a given word-form candidate decides the cell or range of cells that the word-form may, all else being equal, compete to occupy.20 There are two sorts of principles in MM that direct this virtual traffic: namely (Wunderlich 1996: 99), Paradigm Principles: Completeness: ‘Every cell of a paradigm must be occupied,’ thus defective paradigms, those lacking one or more logically possible forms that other lexemes of the same category possess, are not predicted and require explanation; and Uniqueness: ‘Every cell of a paradigm is uniquely occupied,’ enforcing anti- synonymy, i.e. no cell permits two or more options in free variation. and Selection Principles: Output Specificity: ‘Word-forms with more feature specifications take precedence over those with fewer feature specifications,’ recalling that in MM only [+] values constitute specification; Input Specificity: ‘Word-forms with underlying (lexically specified) feature values take precedence over those with derived values,’ predicting that a specified stem precludes the redundant introduction of the same value via affixation, also known as blocking; Simplicity: ‘Strings made of fewer affixes take precedence over those made up of more affixes’; and Reduction: ‘Disjunctive information [ a ˅ b ] is reduced to [ b ] by the presence of forms expressing [ a ],’ that is, only contrasting feature specifications are considered in comparing specificity. Each Selection Principle has a functional motivation, yet the simultaneous evaluation of a set of candidate word-forms in light of these principles does not necessarily – or
52 Contemporary Morphological Theories
even typically – yield a single consensus ‘victor’. Wunderlich (1996: 99) notes that different languages may show evidence of distinct relative rankings with respect to the Selection Principles, and proposes that a constraint satisfaction model on the lines of Optimality Theory (OT, Prince and Smolensky 2004; see §2.7 and §2.13) would provide an adequate resolution in cases of competition. Recommended reading for Minimalist Morphology Fabri et al. (1996) Wunderlich (1996) Wunderlich and Fabri (1995). 2.9 N A T U R A L M O R P H O L O G Y Table 2.13 Continuum table for Natural Morphology. Morpheme-based Formalist In-grammar Phonological formalism Incremental
Ꮻ Ꮻ Ꮻ Ꮻ
Ꮻ
Word/Lexeme-based Functionalist In-lexicon Syntactic formalism Realisational
Natural Morphology (NM) calls for maximising the precision of distinctions among rule types and components of grammar. This sounds like a formalist objective, but at its root NM is motivated by goals consistent with those of the functionalist school of Natural Phonology (Donegan and Stampe 1977, 1979). Just as Natural Phonology had distinguished so-called natural (fundamental, universal, automatic) processes from language-specific acquired rules in phonology (Stampe 1979), so too does NM seek to distinguish rules of morpho(pho)nology both from automatic phonology and from morphology proper.21 Dressler (1985b: 3–4) holds the view that there is an interface between morphology and phonology – namely, morphonology, which is not in itself a component with permanent defining attributes of its own, but rather is defined in an ongoing dynamic with its neighbours to varying degrees in different circumstances. He distinguishes the following ‘upward’ progression of rule types: 1. phonological rules (PRs), which include ‘automatic, exceptionless phonemic alternations’, even in derived environments such as the (neutralising) voicing assimilation that affects the shape of the English plural marker -s (3, 12); 2. morphonological rules (MPRs), which are ‘redundant’ (268) ‘co-signs of morphological rules/categories’ (117), and as such may resemble PRs, but are not automatic and typically involve opacity, e.g. an MPR that derives [waɪvz], given /waɪf+z/ (12);
Theory profiles 53
3. allomorphic morphological rules (AMRs), which are also typically opaque and which furthermore govern morphologically and/or lexically conditioned allomorphy, e.g. adjusting /ɔks+z/ to ox-en, adjusting /ʧaɪld+z/ to childr-en via a stem-umlaut AMR and an application of the same plural suffix AMR as in oxen, and explicitly applying only a stem-umlaut AMR in a case like foot ~ feet, invoking neither the /+z/ suffix as above nor a ‘zero’-suffix, the latter rejected as ‘needless’ and ‘undesirable’ in principle (13); and 4. morphological (spell-out) rules (MRs), which ‘mediate between morphological meaning and form’, and are therefore distinguishable from the other three types, all of which ‘rewrite morphological forms’ through their application (12). The rule that attaches the /+z/ plural marker in the corresponding cases above, logically (or derivationally) prior to the voicing PR, the stem-voicing MPR, and/or the AMR adjusting the allomorph to -en, is at the level of a spell-out MR (11–13). Dressler (1985b) clearly distinguishes the segments involved in morphonological alternations from their domain of application, and he seeks to separate rule types according to phonological, morphological, lexical, stylistic, and other conditions on application (145–50). A sound-structural rule in this framework can have many sorts of conditions beyond the phonological, and this poses no formal or logical problem, because a wide range of linguistic and extra-linguistic constraints are ‘in the model’. There is an attempt to counter this expressive power, however, in the expounding of a potentially arcane system of ‘demerits’ (by virtue of its detail and multi-dimensionality), scores of 1–5 that may be assigned to a given rule as a mark of its relative ‘naturalness’ – in this case, its proximity to automatic phonology – according to generality of application, phonetic distance between alternants, effect on pronounceability and perceptibility, productivity (in several senses), and so on (57ff.). NM considers not only the description of language data, but also the purpose of each element in the context of its use. Dressler judiciously puts forward some of the logical pitfalls of functionalist argumentation: 1. circularity (can independent criteria for markedness and/or naturalness be identified), 2. ad hoc devices (unboundedly many goals to be served simultaneously by language use in a finite context), and 3. teleology in variation and change (quasi-mystical ‘group-minds’ or ‘community grammars’). Dressler counters with the issues of systemic coherence in the NM model, the impossibility of any complete explanation, and the – perhaps under-appreciated – extra-linguistic (functional) factors that control variation and change (Dressler 1985b: 26–67).22 In this connection, two driving assumptions shape the entire NM system of morphological description, prediction, and explanation:
54 Contemporary Morphological Theories
1. The goal of language use is effective communication, and 2. Language is a semiotic system in the sense of Peirce (1965). Words are primary signs in NM, morphemes are secondary signs (‘signs on signs’), and phonemes are tertiary signs. With signs as an organising principle in the service of communication, the clear demarcation of the constituent signs in a string best facilitates the interpretation of the signs and the recovery of the primary meaning. It is predicted, therefore, that the more sign-like a morpheme is, the more efficient it is as a means of communication. Segmentability being systematically favoured, any process or rule which serves to obscure morpheme boundaries (e.g. much of morphonology) is predicted to be contrary to the goal of clear and efficient communication, and thus there will be pressure from within the system to ‘iron out’ the alternation and thus to converge on a constant form–meaning correspondence, i.e. as a good sign should (300–6). Where such convergence does not obtain, the explanation is presumably to be found in a conflicting function which inhibits the (re)unification. Mayerthaler ([1981] 1988) leans strongly toward universal functions which all languages must address, e.g. the symbolising or encoding of semantic concepts. In order to accomplish a meaningful characterisation of universal naturalness, he draws most of his supporting data from language change and language typology. Mayerthaler draws a number of broad distinctions concerning the relative markedness of related pairs of semantic concepts regularly expressed in language (e.g. definiteness/indefiniteness, animate/inanimate, present/preterite, etc.) and determines that ‘prototypical speaker attributes’ (including the ‘here and now,’ 1st-person pronominals, etc.) are universally less marked, and thus ‘the more important and constructive [their] role is for the organisation of natural languages’ (15). For this to have any empirical content, there must be some translation, some correspondence of semantic naturalness (unmarkedness) in the form of language. This relationship between meaning and form is what Mayerthaler calls optimal symbolising. If morphology is sign-based, the more semantically marked a feature to be symbolised is, the more ‘featured’ (essentially, longer) the symbolisation will be: ‘What is semantically “more”, should also be constructionally “more”’ (19). The following is the proposed scheme of symbolising types (18): A. Featureless (no overt marking) B. Featured (some overt marking) 1. Additive Featured (increased content) a. Particle Additive (affixed) b. Modulator Additive (segmental lengthening) 2. Modulator Featured (segmental substitution). As signs go, then, the optimal symbolisation is B.1.a., or affixation, and the longer the better for the symbolisation of a semantically marked element (cf. Latin positive longus – comparative longior – superlative longissimus). Zero-conversions and non- concatenative morphology are, of course, predicted to be inefficient symbols, so their use, especially in a systematic way, is a puzzle for NM (subtractive morphology
Theory profiles 55
is not addressed). Mayerthaler is quick to point out that the predictions of NM are always relative tendencies, rather than categorical statements. This correspondence between markedness and symbolisation is a type of (weak) homomorphism, and Mayerthaler terms it constructional iconism (17–25). Homomorphism of this sort is a desirable condition from the perspective of NM, but it is admittedly an idealisation which must often be disrupted in the service of competing linguistic (and extra-linguistic) functions. Dressler (1985b: 301) modifies the simple ‘bigger is better’ sign evaluation metric of Mayerthaler with an appeal to the practicalities of perception and production – an efficient sign ‘must be neither too big nor too small’. Along with iconicity in the form of a sign, Dressler also stresses the value of a biunique relationship between the signifier and the signified, so that either is readily accessed from the other (301). The combined pressure of iconicity and biuniqueness motivates the prediction that operations which fuse or delete morphemes, whether in part or in total, are diachronically unstable and synchronically rare (306). Ambiguity in the input or output of any rule is a strike against it as a natural rule, and thus homophonous morphemes are to be disfavoured (313). Dressler lays out the following scale, ranked in descending order of morphotactic transparency (316–17): 1. 2. 3. 4. 5. 6. 7.
Intrinsic allophonic PRs intervene, Extrinsic allophonic PRs and/or resyllabification, Neutralising PRs intervene, MPRs, MPRs with fusion, AMRs, and Suppletion.
Dressler notes that rules often change type over time, and contrary to all expectations of naturalness, the change tends to be in the direction of decreasing transparency. A proposal in this regard is that the transitions between PRs and MPRs, as well as between MPRs and AMRs, are ‘discrete and gradual’ in nature, rather than abrupt, and furthermore that there is a ‘typical and unidirectional . . . evolution’ among the rule types in exactly that sequence (PR > MPR > AMR) over time (Dressler 1985b: 22; cf. Janda 1986: 600–11). NM posits that, all else being equal, biuniqueness and productivity go hand in hand, with the former implying the latter (Dressler 1985b: 329). In this way, the addition of new words to the lexical stock, which should employ the most productive means available, ought to involve the application of the clearest (i.e. unique, and perhaps transparent, too) signifier for the signified in question. This reasoning verges on the circular, but through an appeal to the diachronic loss of transparency in rules, NM can allow for, if not actually account for, the development of polysemous morphemes and the rise of new productive morphemes displacing the old. Almost coextensive with, but inversely related to, the scale of morphotactic transparency (excluding suppletion) is the scale of indexicality: as a rule becomes more context-sensitive, the presence of the output ‘points’ more clearly to the presence of
56 Contemporary Morphological Theories
its conditioning environment, and the greater the phonetic distance between input and output, i.e. the greater the change the rule effects, the more indexical the rule is (thus intrinsic allophonic rules have almost no indexical value). Although NM assumes fuzzy transitions from one rule type to another, it is nevertheless a modular theory, such that the application of PRs presupposes the application of MRs. A subtle consequence of this modularity is the quantum leap in the indexicality of a rule once it becomes morphologised. MRs ‘precede’ PRs, and thus they have a certain priority over PRs. MRs furthermore have semanticity, which phonemes and allophones (in themselves) do not. For these reasons, it becomes much less troubling that morphotactic opacity increases over time, since indexicality and semanticity increase correspondingly (Dressler 1985b: 309–11, 333–4). Wurzel (1989) turns the focus of NM specifically on inflectional morphology, from the perspective of systems as coherent and consistent wholes. Not to dismiss the role of language typology, but rather to take individual languages as extensions of types, Wurzel refers to System-Defining Structural Properties (SDSPs), which organise and lend stability to inflectional systems. Since inflectional classes are based on paradigms, and paradigms are based on inflectional markers, and markers in a given language ‘are not part of any universal inventory of markers’ (63), introducing inflectional classes into a discussion of morphological naturalness is inevitably challenging (cf. MM, §2.8). Wurzel speaks of morphological norms at the language- specific level, rather than in terms of naturalness in general. For example in Modern German, since weak verb formation is both increasingly common and the only productive rule for new verbs in the language, this weak pattern is (currently) the norm for German (64–5). He couches the range of SDSPs as parameters (75): 1. an inventory of categorial complexes and categories assigned to them, 2. the occurrence of basic-form inflection or stem inflection, 3. the separate as opposed to combined symbolisation of categories of different categorial complexes, 4. the number and distribution of formally distinct inflectional forms in a paradigm, 5. the types of markers occurring, and their relations to the categorial complexes concerned, and 6. the existence or non-existence of inflectional classes. Given these SDSPs, one may construct a ‘typological characterisation and classification of inflectional systems’ (75). In very congruent systems, the SDSPs act almost as laws, while in more mixed systems, the SDSPs stand more as defaults, i.e. as what happens when no extraordinary circumstances come into play (82; cf. Zwicky 1986). System congruence (congruity in Wurzel) is not something that can necessarily be assessed through cursory inspection. Rather, it involves extensive and exhaustive comparisons, e.g. (Wurzel 1989: 83): 1. of (abstract) marker types (e.g. suffixes), 2. of particular markers,
Theory profiles 57
3. of the number of distinct inflectional forms in different paradigms, 4. of co-occurrence of various markers, and 5. of all the different markers realising particular inflectional categories. Given the SDSPs and the mass of empirical data that one would collect in discerning a particular language’s set of parameter settings, it is not hard to imagine that a system- internal pressure toward increased congruence is proposed as a motivator of morphological change. It is also not surprising, given the focus in NM on conflicting functional motivations, that these SDSPs come into conflict with system-independent considerations, i.e. the more universally oriented issues identified by Mayerthaler ([1981] 1988). In this connection, Wheeler (1993) urges caution with respect to some of NM’s proposals of hard-and-fast hierarchical arrangements among naturalness principles, demanding empirical verification of each putative pair-wise precedence, rather than theory-internal or purely logical presupposition (100–1). As point of departure, Wurzel (1987: 92–3) had proposed the following hierarchy of naturalness principles, arranged in order of greatest to least in influence vis-à-vis the other principles: System-dependent principles are as follows: I. The Principle of System Congruity, favouring uniform inflectional systems that accord with a language’s morphological type; II. The Principle of Class Stability, favouring independent motivation and maximally general implication patterns in inflectional classes; System-independent principles are as follows: III. The Principle of Uniformity and Transparency, favouring biuniqueness (one-to-one matching) between morphological form and content; IV. The Principle of Constructional Iconicity, favouring a correlation between degree of markedness at a semantic level and formal marking; and V. The Principle of Phonetic Iconicity, favouring correspondences between semantic content and sounds used to indicate the presence of said content (à la onomatopoeia and sound symbolism, and so not specifically morphological). In light of examples from Romance, Germanic, Slavonic, and Turkic, Wheeler contests the definitions and hierarchical arrangement, most specifically those involved in Wurzel’s principles III and IV. Toward this end, Wheeler adds two principles, and furthermore revises Wurzel’s principle IV so as to relativise its originally stated preference for a more absolute correspondence between semantic markedness and formal signification thereof. One added principle – namely, the Avoidance of Counter-iconicity; that is, subtractive marking – is held to dominate the other system-independent principles. Also added is a principle of Markedness in Syncretism, which tolerates identity of forms for morphosyntactically more marked properties, but strongly disfavouring syncretism with a form signifying an
58 Contemporary Morphological Theories
unmarked property. In Wheeler’s estimation, a better-supported hierarchy would be as follows: I. System Congruity, II. Class Stability, III. Avoidance of Counter-iconicity, IV. Uniformity and Transparency, Markedness in Syncretism, Constructional Iconicity (relativised), and V. Phonetic Iconicity. This arrangement claims that although some relations of dominance among naturalness principles may be demonstrated in cases where joint satisfaction is impossible, other principles may not be definitively so ordered, and may in fact be subject to language-specific patterns of conflict resolution (109; renumbered to reflect unranked relations). In summary, NM predicts that the most efficient morphological system will exhibit iconicity and biuniqueness to the highest degree possible, avoiding syncretism in general, and avoiding zero-marking on all but the most basic (semantically least marked) expressions. To the degree that languages do countenance syncretism and zero-derivation, this is claimed to be the result of conflict with other systemic as well as extralinguistic pressures, and further, that such language states are rare, unstable, and subject to change at the earliest opportunity. Although the testing ground for these intuitively plausible hypotheses is (necessarily) based on the description of synchronic morphological systems, the methodological focus is frequently on comparison with some other language state, to evaluate the relative naturalness of the states. Indeed, Mayerthaler states in this regard, ‘[W]e do not believe in the possibility of a synchronic linguistics in the sense that it would be possible to write an adequate grammar excluding the dimension of time’ ([1981] 1988: 4). It is perhaps not an accident, therefore, that a theory that holds multiple gradient scales as central organising principles – scales of naturalness that hold scope over universal, typological, or language-specific, system-particular domains – focuses on language comparison and change. As a tool of the typologist, historical linguist, or dialectologist, among others, NM could surely have an appeal in its functionalist and system-wide (‘macro’) orientation. Recommended reading for Natural Morphology Dressler (1985a) Dressler (1985b) Dressler (1985c) Dressler (2005) Dressler et al. (1987) Mayerthaler ([1981] 1988) Wurzel (1989).
Theory profiles 59
2.10 T H E N E T W O R K M O D E L Table 2.14 Continuum table for the Network Model. Morpheme-based Formalist In-grammar Phonological formalism Incremental
Ꮻ Ꮻ Ꮻ
Ꮻ Ꮻ
Word/Lexeme-based Functionalist In-lexicon Syntactic formalism Realisational
Joan Bybee, a leading force in the Natural Generative Phonology movement (Hooper 1976, 1979), has proposed the Network Model, a functionally oriented view of morphology, seen first and foremost as an organised system. In the preface to her (1985) book, Bybee observes that it might appear strange to some that her attention had moved from a very concrete approach to phonology to settle on ‘a different set of issues’ (v). On the contrary, an approach to morphology follows naturally from the careful division of morphophonemics from articulatory-and perceptual- based phonology (‘phonology proper’). By emphasising that morphophonology is morphologically conditioned, and therefore part of the domain of morphology, Bybee likens the arbitrary nature of much of morphophonology to the arbitrariness found throughout morphemics, l’arbitraire du signe. As a functionalist theory, the concern is not with descriptive segmentation of morphemes, because there are simply too many deviations from a one-to-one form– meaning correspondence. Bybee’s goal is ‘to propose certain principles in a theory of morphology whose goal is to explain the recurrent properties of morphological systems, including fusion and allomorphy, which are traditionally viewed as problems [many-to-one and one-to-many meaning–form connections, respectively], in terms of the general cognitive and psychological characteristics of human language users’ (3). With cognition as a concern, psycholinguistic experimentation is an important source of evidence for the claims of the Network Model. Similarly, because the goal is to explain recurrent patterns, cross-linguistic data from linguistic typology is also of importance. Many formalist theories, by contrast, tend to de- emphasise evidence of these sorts, because they introduce gradient patterning, rather than neat categorical behaviour. In contrast with the Word Syntax approach (§2.15 below), the Network Model suggests that the lexicon is not merely structured; it is richly structured, with connections at many levels. Phonological connections, syntactic (categorial and subcategorial) connections, and semantic connections link words and parts of words simultaneously. Multiple links constitute lexical associations of differing strength and character, and generalisations about lexical subclasses can refer to constellations of links, including links from different grammatical domains (cf. Jackendoff 1975). For example, much of the exposition of the network model in Bybee and Moder (1983) is done in terms of the ablauting strong verbs in English, such as sing/sung and string/strung. Bybee and Moder show that the oldest members of this class were
60 Contemporary Morphological Theories
monosyllables ending in a velar nasal. Later additions to the class have diversified this condition, allowing for a final velar and/or nasal, as in dig/dug and spin/spun, respectively. Rather than deriving morphologically complex words by rules per se, Bybee’s model appeals to patterns among the various links in the lexicon to identify morphological patterns. Thus, morphological analysis is radically not about Items and Arrangements or Items and Processes, nor is it about Words and Paradigms. Morphological analyses are implicit in the lexical connections that individual speakers make in their own lexicons. Patterns defined by links can be referred to as schemas, either source-oriented or product-oriented, as conditions guiding the coining and interpretation of novel forms (Bybee 1985: 129; Bybee and Moder 1983: 255). The individualisation of morphological analysis is not a surrender to chaos and unpredictability, however, since the empirical experience of speakers acquiring and processing their language, especially within the same community, is very likely to be comparable. With comparable experience, the reasoning goes, will come motivation for largely coinciding lexical structure. In this way, quite contrary to the ‘ideal speaker–hearer’ approach often appealed to in (Chomskyan) linguistic theorising, the Network Model is based in the experience and general cognitive processes of natural language users. Bybee (1985) claims that derivational and inflectional morphology are not qualitatively distinct phenomena, but rather ‘a gradual . . . distinction, the basis of which is relevance’ (5). Not only, then, is morphology restricted to the lexicon, but also form and function are distinguished quite clearly, although in practical terms, each dimension on its own is gradient. ‘The semantic relevance of an affix to a stem is the extent to which the meaning of the affix directly affects the meaning of the stem’ (4). This is potentially a vague and variable gradient, but Bybee purports to avoid ‘ethnocentrism’ by drawing claims about relative relevance from a cross-linguistic comparison of fifty languages in widely different language families and geographical regions, thereby escaping (to the greatest degree possible) genetic or contact confounds (8). From this typological evidence, Bybee claims support not only for the categories which she posits along the continuum, but also for the relative ranking of each grammatical meaning. Specifically, with reference to verb morphology, ‘the categories of valence, voice, aspect, tense, mood and agreement are ranked for relevance to verbs in that order’ (4–5). Part II of Bybee (1985: 137–205) discusses in depth what is understood by ‘aspect’, ‘tense’, and ‘mood’ in her model, in order to clarify the categories for further testing and to preempt spurious counter-examples which might follow from differing definitions of what constitutes a tense, for example. Bybee’s assumption of a cline of relevance allows her to make predictions about exponent form, on the one hand, and sequencing on the other. Bybee (1985: 4–5) claims that exponents of more relevant grammatical meanings will be found closer to the verb stem than will those of less relevant meanings, and more relevant exponents are more likely to involve morphophonological alternations in the affix, the stem, or both (Bybee 1985: 33–43). The Network Model has a variety of independently proposed solutions to problematic issues in deviations from one-to-one form–meaning matching (Carstairs
Theory profiles 61
1987). Morphophonology is considered a historical relic of earlier phonetically motivated alternation now housed in the morphology, rather than something to be processually recapitulated in putatively synchronic phonological rules. Fusional morphology is similar, and is claimed to follow from frequent co-occurrence of morpheme pairs. Affix genesis is rooted in semantic bleaching and phonetic erosion (without calling this by the name ‘grammatic(al)isation’). These explanations, in their broadest senses at least, are generally agreed upon in historical circles. Perhaps surprisingly, Bybee (1985: 50–8) finds a place for the basic-derived distinction in the lexicon, the natural occasion to appeal to a scale of relative (un)markedness. Unmarked (i.e. zero-marked) word-forms are predicted to be semantically unmarked, or at least no more semantically marked than the most unmarked word-form in the paradigm. Markedness is of course a concept which is frequently criticised for cross-investigator inconsistencies, but the typological and psycholinguistic bases for Bybee’s analysis, including, for example, the sequence of acquisition of forms, add weight to her argumentation. Proposals in this area are not without difficulty, however. Bybee (1985) implies that considerable support exists for the claim that semantically unmarked forms are ‘morphophonemically simpler’ than more marked forms (6). Certain stem allomorphy facts from Sanskrit would seem to deviate from this prediction, in that for those paradigms with weak and strong stems, and especially for those with three grades of stem, the weak(est) stems (in white) are found in oblique cases to the exclusion or near-exclusion of the direct cases, where the strong stem (in grey) generally predominates, e.g. in the masculine forms of the possessive adjective bhagavant ‘fortunate’ (Stump 2001: 170): Table 2.15 Masculine forms of the Sanskrit possessive adjective bhagavant ‘fortunate’. (Source: Stump 2001: 170, © Cambridge University Press)
Nominative Accusative Instrumental Dative Ablative Genitive Locative
Singular
Dual
Plural
bhagava¯n bhagavant-am bhagavat-a¯ bhagavat-a¯ bhagavat-e bhagavat-as bhagavat-i
bhagavant-a¯u bhagavant-a¯u bhagavad-bhya¯m bhagavad-bhya¯m bhagavad-bhya¯m bhagavat-os bhagavat-os
bhagavant-as bhagavat-as bhagavad-bhis bhagavad-bhyas bhagavad-bhyas bhagavat-a¯m bhagavat-su
Even though it would appear that the nominative singular is the only ‘zero-marked’ form in the paradigm, the weak stem, which is always ‘morphophonemically simpler’, is used in the oblique forms, and never in the nominative singular. This is primarily a suggestion for a redefinition of basic versus derived, however, since on grounds of predictability, the weak stem is usually predictable from the strong stem, but the reverse is less reliable, meaning that ‘morphophonemically simpler’ can be a misleading diagnostic for the directionality of derivation. It would seem, therefore,
62 Contemporary Morphological Theories
that in the Network Model, the interlexical connections might more reliably point to a basic form than a guideline framed in terms of the relative number of phonemes. A complex perspective has emerged from Bybee’s particular programme. This involves the difference between regular and irregular, productive and non-productive, and type versus token frequencies of words. In Bybee and Moder (1983: 251), irregular inflectional forms, particularly those involving morphophonological alternations, are claimed to be ‘scheduled for levelling, since they disrupt the one-to-one correspondence between sound and meaning’. In Bybee (1995), however, an explanation for the endurance of certain disruptive alternations is explained with reference to token frequency (lexical strength, in Bybee’s terminology): ‘irregulars will tend to regularise unless they are sufficiently available in the input to create a strong lexical representation. Thus if the irregular past has low token frequency and is thus more difficult to access, a regular form might be created’ (428). The more frequently a verb is used, the more able it is to sustain irregularity in its paradigm, should any such irregularity exist. The verb to be is cross-linguistically very likely to show some irregularity in its paradigm, and Bybee’s claim is that the reason is the frequency with which forms of the verb to be are used in everyday speech. The pressure of conventional usage ‘outweighs’ the pressure of regularity. The sometimes elusive notion of productivity (see §4.2) is also a function of frequency in the Network Model, but in this case it is type frequency, the proportion of the vocabulary in the relevant grammatical category which participates in a particular pattern. The higher the type frequency, the more likely the class is to act as a default, and consequently the more likely it is to be employed for analogy, as in cases of doubt, neologism, and language acquisition. The chances for the expansion of the pattern’s input set increase as a result, and this means an increase in the pattern’s productivity. Formalist theories, on the other hand, tend not to worry as much about pattern frequencies overall, with the exception of theories such as Network Morphology (§2.11) and PFM (§2.12), in which default patterns play a central role in rule application. There are many parallels between the Network Model and the concerns of Natural Morphology (§2.9 above), just as there are parallels in the perspectives of their respective antecedents – namely, Natural Generative Phonology and Natural Phonology – with regard to the types of evidence that each allows, the larger systemic questions which each seeks to address, and the focus on the isomorphic sign as a driving influence in change. For these reasons, it seems a little strange that two schools cite each other’s work almost not at all (with the exception of Dressler 1985b, where three references to Bybee are given). Bybee offers Network Model as a tentative theory-name, but not until Bybee (1995: 428). Since similarities with the Network Morphology programme (§2.11 below; named just prior, in 1992–3) are very limited, perhaps the Network Model may ultimately need another title. As for the appeal of the Network Model as it stands, however, its attention to cognition and typology make it a likely stimulus for new research programmes in psycholinguistics, and its clear and falsifiable predictions make it a standing challenge to those engaged in the description of synchronic systems and diachronic changes.
Theory profiles 63
Recommended readings for the Network Model Bybee (1985) Bybee (1988) Bybee (1995) Bybee (2007) Bybee and Moder (1983). 2.11 N E T W O R K M O R P H O L O G Y Table 2.16 Continuum table for Network Morphology. Morpheme-based Formalist In-grammar Phonological formalism Incremental
Ꮻ
Ꮻ Ꮻ
Ꮻ Ꮻ
Word/Lexeme-based Functionalist In-lexicon Syntactic formalism Realisational
Network Morphology has been developed by the Surrey Morphology Group. An integral part of the theory is the computer language DATR (Evans and Gazdar 1996), which was designed with lexicon modelling in mind. The lexical knowledge modelled in Network Morphology is based on the common computational principles of hierarchy and inheritance. Network Morphology lexica are strongly hierarchical, and individual lexical entries are typed feature matrices, analogous to representations in Head-driven Phrase Structure Grammar (HPSG; Pollard and Sag 1994) and Generalised Phrase Structure Grammar (GPSG; Gazdar et al. 1985), adjusted for direct computational implementation as lines of programming code. The authors of articles in Network Morphology move frequently between feature notation and tree diagrams, which is helpful to a reader who may not always be able to picture the dependency relations in the compressed featural format. Based on the concept of feature inheritance, the Network Morphology lexicon begins at the very top with the type word, which branches into subtypes according to syntactic categories. New subtypes are motivated each time there is a subset of lexemes which differs from the default feature set in some systematic way. A subtype must have some specific feature value which differs from the larger class; this feature value overrides the feature value the subtype would inherit by default from the supertype. In this way, dependent types largely cohere with their parent types, and sister types cohere in the defaults they jointly inherit from a common parent node. Lexical classes and subclasses are thus defined, and this allows generalisations to refer to individual nodes or hierarchically related nodes. Simultaneously, this suggests that generalisations will not hold over disparate classes, i.e. those not so related in the hierarchy (this expectation is not exactly air-tight, but there is a systematic method proposed to handle referrals across branches, discussed below).
64 Contemporary Morphological Theories
Corbett and Fraser (1993: 126, 136) provide a more concrete example. The declension classes in Russian are generally claimed to number three or four. An example paradigm for four typical nouns will show the reason for the ambivalence (115). Table 2.17 Russian noun declension classes. (Source: Corbett and Fraser 1993: 115, © Cambridge University Press)
Nom. sg. Acc. sg. Gen. sg. Dat. sg. Inst. sg. Loc. sg. Nom. pl. Acc. pl. Gen. pl. Dat. pl. Inst. pl. Loc. pl.
I zakon ‘law’
II komnata ‘room’
III kost’ ‘bone’
IV v’ino ‘wine’
zakon zakon zakona zakonu zakonom zakone zakoni zakoni zakonov zakonam zakonam’i zakonax
komnata komnatu komnati komnate komnatoj komnate komnati komnati komnat komnatam komnatam’i komnatax
kost’ kost’ kost’i kost’i kost’ju kost’i kost’i kost’i kost’ej kost’am kost’am’i kost’ax
v’ino v’ino v’ina v’inu v’inom v’ine v’ina v’ina v’in v’inam v’inam’i v’inax
It has been noted that declensions I and IV are formally quite similar, contrasting clearly with both II and III (Corbett 1982). Network Morphology allows for the capturing of gradient similarities with a hierarchical approach to the lexicon. The following hierarchy (cf. Corbett and Fraser 1993: 126) shows the formal affiliation of Russian declension classes: 1. Nominal A. Adjective B. Noun 1. N_O (traditional o-stems) a. N_I, e.g. zakon b. N_IV, e.g. v’ino 2. N_II, e.g. komnata 3. N_III, e.g. kost’ This hierarchy captures ‘the fact that there are four main declension classes [in Russian], but that the differences between N_I and N_IV are not as great as those between either of them and the other declensional classes’ (127). N_O is a ‘super-node’ from which N_I and N_IV inherit their shared properties (127). Since Network Morphology revolves around type hierarchies, it is important to note that each node in the network corresponds to a class of lexemes, characterised by common attribute-value pairs, called facts. Facts are inheritable downward in the network, unless overridden by specific facts listed at an intervening node in
Theory profiles 65
the path, down to and including the node in question. For this reason, facts about inflectional classes are composites of inherited facts and stipulated subclass-specific facts. In order for a declension class to be ‘well typed’, the composite of facts must constitute a complete set of rules of inference (i.e. facts) for a full inflectional paradigm appropriate to the lexeme-class.23 To turn this hierarchy briefly into a DATR representation24 (adapted from Corbett and Fraser 1993: 135–6): NOMINAL: == “” == hard == “” == “” == “” == “” == (“” “” _m) == (“” “” _m’i) == (“” “” _x). NOUN: < > == NOMINAL == (“” _e) == (“” _i) == “< “” mor gen pl>” == (“” _ej) == _a == n. N_O: < > == NOUN % traditional o- stems == (“” _a) == (“” _u) == (“” _om). N_I: < > == N_O == masc == “” == (“” _ov).
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
This fragment (of a fragment) of a grammar is designed to show both default inheritance (< > == X) and the node-specific facts which introduce new information ( == masc).25 Using fact-indices (the line numbers on the right, above), and given the following lexical entry for the noun zakon ‘law’: Zakon
< > == N_I == law == zakon == inanimate
25
the rules of inference used in the inflected forms in a paradigm of class N_I are as follows:
66 Contemporary Morphological Theories Table 2.18 Rules of inference for Russian declension class N_I.
Nom. sg. Acc. sg. Gen. sg. Dat. sg. Inst. sg. Loc. sg. Nom. pl. Acc. pl. Gen. pl. Dat. pl. Inst. pl. Loc. pl.
I – zakon ‘law’
Rules used
zakon zakon zakona zakonu zakonom zakone zakoni zakoni zakonov zakonam zakonam’i zakonax
1, 23, 25 1, 4, 23, 25 1, 18, 25 1, 19, 25 1, 20, 25 1, 11, 25 1, 12, 25 1, 4, 12, 25 1, 2, 3, 13, 24, 25 1, 7, 15, 25 1, 8, 15, 25 1, 8, 15, 25
Network Morphology also permits rules of referral (Zwicky 1985, 1992; Stump 1993a), whereby systematic formal parallelisms not handled by defaults are f ormalised as a stipulated referral to another form in an analogous paradigm, e.g. for N_III, the value for is referred to the corresponding value under N_I, whereas the value for N_III’s is referred to that of N_II. These referrals are ways of expressing parallelisms not predicted by hierarchical inheritance patterns. Network Morphology offers a rich formal system for the representation of lexical patterns. It was designed with computational implementation in mind, and so there is a practical advantage for such uses in choosing this framework. It is clear from the above examples and from the hierarchical lexicon approach in general that all morphology is handled in the lexicon – derivation mapping from one lexeme to another, and inflectional patterns handled through defaults and overrides as one moves down the path from the most general lexical class to specific lexical entries. This clearly implies that Network Morphology is realisational in its approach, since the formal markings are values for abstract attributes in the feature representation. Once a large enough grammatical fragment is built and particular lexical entries are introduced into the model, the programme’s output is the full inflectional paradigm of each lexeme, marked with and values (see Corbett and Fraser 1993: 139–41). The formalism and level of detail needed for computational implementation might be off-putting or even irrelevant for some potential consumers of morphological theory. Some might also question the licence to split subtypes of subtypes with no defined limit. The inheritance metaphor, however, makes clear predictions, and the possibility of computational implementation of grammars compiled using this model make for a very appealing (virtually) empirical check on the correctness of predictions. Since correct output does not necessarily guarantee the optimal description, Network Morphology’s reliance on default inheritance supplies the impulse to minimise redundancy in the lexical representations. As a descriptive tool and as a computational input, Network Morphology is designed with the future of linguistic research in mind.
Theory profiles 67
Recommended reading for Network Morphology Brown (1998) Brown and Hippisley (2012) Brown et al. (1996) Corbett and Fraser (1993) Fraser and Corbett (1995) www.surrey.ac.uk/englishandlanguages/research/smg/index.htm 2.12 P A R A D I G M F U N C T I O N M O R P H O L O G Y Table 2.19 Continuum table for Paradigm Function Morphology. Morpheme-based Formalist In-grammar Phonological formalism Incremental
Ꮻ
Ꮻ Ꮻ
Ꮻ Ꮻ
Word/Lexeme-based Functionalist In-lexicon Syntactic formalism Realisational
Paradigm Function Morphology (PFM) is a lexeme-based realisational theory of inflectional morphology. Stump’s work owes much to the theory and metatheory of Arnold Zwicky, and PFM spells out in detail some of the leading ideas of Zwicky’s (e.g. 1987b) Interface Program. Although PFM’s introduction in the literature is generally taken to be Stump’s (1991) article, some important precursors may be gleaned in his less formally oriented (1990) piece: The proposed framework embodies a conception of the boundary between inflection and derivation that is wholly at odds with the split morphology hypothesis. In particular, this framework does not treat inflection as an extralexical phenomenon but instead presupposes that all morphological processes operate in the lexicon. It does not presume that all rules of derivation inherently precede all rules of inflection but instead allows some intermixture of inflection with derivation . . . [I]nflectional and derivational processes are distinguished according to the kinds of expressions that they produce . . . . (116–17) With the exception of Matthews (esp. 1972) and Carstairs (e.g. 1983, 1987), PFM gives unusual prominence to the paradigm as an organising principle in morphology. Many theories have a nodding acquaintance with the paradigm, but treat it as an epiphenomenon, something with pedagogical or perhaps only curiosity value (e.g. Anderson 1992: 79–80). This does not mean that PFM treats the paradigm as a primitive, however. The paradigm in PFM is a set of cells defined by the universal and language- specific co-occurrence restrictions on morphosyntactic features and their permissible values. Every cell in the paradigm, therefore, corresponds to a complete well-formed set s of morphosyntactic properties (i.e. feature-value pairs). For the paradigm of a
68 Contemporary Morphological Theories
lexeme L, the form Y which occupies a given cell is the realisation of the corresponding set s on the root X of L. The eponymous paradigm function (PF) is a mapping from a root-pairing to a (word-)pairing : that is, to an inflected word with the property set appropriate to the cell it appears in. A paradigm function is in turn ‘defined in terms of more specific realisation rules’ – the individual rules of morphology realising the language’s morphosyntactic properties (Stump 2001: 33). The formalism of PFM is both rigorous and interpretable within the traditions of formal linguistics. That said, however, there are a few potential barriers to clear interpretation. Each realisation rule, for example, bears a triple subscript: the rule block n that the rule belongs to, the proper subset t of s that the rule participates in realising, and the lexeme class C whose paradigm function the rule participates in defining. Rule format: RRn, {t}, [C] () =def All of this appears before the root-pair is encountered, and well before the effect of the rule on the root X, i.e. Y’, is encountered. If strict attention is not paid to a sometimes quite long string of subscripted abbreviations, it can be difficult to keep track of the point being made in each rule. For example, from Old Norse (Stewart 2000): RR2,{MOOD:indic,VOI:act,TNS:past,PER:3,NUM:pl.},[V]() =def Stump (2001) integrates and builds upon ideas in the articles which preceded it, making the whole more accessible. In PFM, rules of all sorts, and consequently PFs as well, are seen as static well- formedness conditions holding between lexical roots and stems, between stems and inflected words. This is in keeping with other non-derivational approaches to linguistics, e.g. HPSG, but the different perspective can be misleading if one takes the descriptive model to represent a derivation in the traditional sense of the word. The step-by-step demonstration of rule evaluation is therefore more on the lines of a logical proof, but the fact that a proof generally looks not unlike an incremental building-up of complex morphological structure (at least in the horizontal dimension) certainly renders it an ‘apparent derivation’. A key concept in PFM is that of the rule block, mentioned in passing above. Stump (2001: 33) likens the block to Anderson’s (1992: 129) use of the same term. An important difference exists, however, between the two conceptualisations. A-Morphous-type blocks were motivated as a response to cases of disjunctive rule application; there is no independent motivation or principle which allowed the rule block to cohere. PFM blocks, by contrast, correspond to the traditional notion of a position class, whereby ‘rules belonging to the same block compete for the same position in the sequence of rules determining a word’s inflectional exponence’ (Stump 2001: 33). ‘Same position’ here is more literally construed than the disjunctions in Anderson (1986, 1992), such that a PFM block of realisation rules corresponds to a ‘slot’ in a word’s sequence of inflectional affixes. PFM rule blocks, therefore, are organised according to the distributional facts of exponence, and not of the more
Theory profiles 69
abstract notion of disjunctive rule application. PFM gets disjunctive application for free, as it were – since no more than one exponent can appear in a given slot, no more than one rule from the same block may apply in the definition of a given PF. Reference to slots while at the same time eschewing morphemes as objects opens PFM for some criticism, because (as happened to the MSRs in A-Morphous Morphology, §2.1 above) zeroes can take up residence in vacant positions. To counter this possibility, and in keeping with the mathematical ‘function’ mentality, absolutely no structural zeroes are allowed for in PFM. Where no rule in a block is applicable to , a universal realisation rule applies, the Identity Function Default (IFD), mapping the input on to itself (Stump 2001: 53, 143): Identity Function Default (IFD): RRn, { }, U () = def Here, n ranges over all rule blocks, { } is the empty set of morphosyntactic properties, and U is a variable over all lexeme-classes.26 The IFD, therefore, is effectively the last rule in every rule block, guaranteeing that a proof never fails because some slot in the PF was undefined for lack of an applicable rule. There is no question of ‘adding Ø’ – the IFD evaluation of the block is ‘no change’. On the issue of rule ordering, PFM denies the need for extrinsic rule ordering. By Pān·ini’s Principle (no disjunctivity rider required, cf. §2.1): given any complete set s of morphosyntactic properties appropriate to a particular lexeme class and any lexeme X in that class, ‘the value of the . . . PF for the root-pairing is always the result of applying the narrowest applicable rule’ in each of the blocks mentioned in the PF schema (Stump 2001: 52; emphasis in original). A PF schema identifies which rule blocks are involved in the definition of the form realising the set s on the root X of lexeme L, e.g.: PF() = def Nar3(Nar2(Nar1(Nar0()))) Narrowness, then, is evaluated between realisation rules in terms of the relative specificity of the set of morphosyntactic properties realised by each rule. This is the method for enforcing the Pān·inian Principle, i.e. proper subset precedence. The Identity Function Default is, for PFM, the ‘default default’, meaning that where no special case is called for, the IFD takes over. The default-override relation is crucial in PFM, as it is in Network Morphology (§2.11) and Construction Morphology (§2.4). Defaults are what lexemes in a particular class ‘inherit’ by virtue of class membership, provided that they are not simultaneously members of a more select class (a proper subset of the larger class) which is subject to a special override rule. The Narrowness relation is simply a principled (rather than extrinsic or arbitrary) and formal way of deciding, between two realisation rules that are applicable in a given case, which would override the other (subject to potential further override by some third rule, narrower still than either of the two). Defaults are therefore layered, and the prediction is that the Pān·inian Principle will always be adequate for the unique determination of the narrowest applicable rule, given the joint assumptions that blocks are position classes and that no block is ever undefined by virtue of the IFD (Stump 2001: 21–5).
70 Contemporary Morphological Theories
Lexemes each possess a stem set consistent, in number of stems at least, with the other members of the same lexeme class. In the general (non-suppletive) case, stems will be related to the root or another stem by rules of stem formation, purely formal operations (Stump 2001: 183–6). If stems occupying corresponding positions in the stem sets of comparable lexemes in distinct inflectional classes are not characterised by parallels in general phonological shape, purely morphological (morphomic, in the sense of Aronoff (1994: 22–9)) rules of index assignment come into play, marking stems as ‘strong’ versus ‘weak’, for example, or assigning arbitrary numerical indices (Stump 2001: 190–4).27 Rule block 0 in any given language is a block of stem selection rules that identify the morphosyntactic properties each stem may (partially or wholly) realise (Stump 2001: 175–9). In this way, regular (and/ or productive) stem-internal non-concatenative marking may be handled by stem formation rules, and the Separation Hypothesis is still respected, since rules of selection and formation are in principle independent. The evaluation of particular realisation rules is stated as a default phonological entity, which implies that the default shape of the exponent may be overridden under specific circumstances. An unordered set of morphophonological rules constrains the evaluation of each realisation rule in any instance of its application. For any given application of a randomly chosen rule, any number of morphophonological rules (including none) may affect the phonological shape of the rule’s evaluation. Where whole blocks of realisation rules or an identifiable subset of rules in a block are subject to one or more particular morphophonological rules, a morphological metageneralisation may be stated concerning those rules to account for this subregularity.28 Spencer (2013) brings together a proposed extension of PFM – namely, Generalised Paradigm Function Morphology (GPFM) – with the goal of moving beyond the inflectional emphasis of PFM and toward accounting for lexical relatedness more generally. Incorporating the inflectional machinery of ‘classical’ PFM within a larger, formally and theoretically compatible framework, GPFM posits a more detailed theory of the lexical entry than PFM had required for its operation. The GPFM lexical entry includes four attributes: namely, FORM, SYN, SEM, and LI (lexical index) (Spencer 2013: 178–81).29 For example, the lexical entry for writeV contains the following attributes: FORM(write) = write SYN(write) = Verb SEM(write) = [WRITE(x, y)] LI(write) = write Spencer alludes to an intervening round of function application at each of these attribute levels, taking the pairing of the lexeme in question and no further specification (here, ), and yielding the values above, taking advantage of the Pān·inian Principle to allow the specificity of the lexeme as argument to apply, and to disjoin all other potential values (Spencer 2013: 189–90; cf. Kiparsky 1982a for the recasting of lexical entries as identity rules). With this step assumed, it is subsequently elided for expository purposes.
Theory profiles 71
Whereas the PF of ‘classical’ PFM mapped pairs on to the cells of inflectional paradigms, GPFM uses the corresponding GPF construct not only to achieve the realisation of inflection but also to map lexeme-derivational relation pairs to derived lexemes, complete with the corresponding alterations to the lexical attributes; for example, in deriving writerN from writeV, the GPF for realising subject nominalisation (SN) applies to the lexical entry of write as follows: GPF() ≡ fform() fsyn() fsem() fli()
= = = =
write er Noun [PERSON(x), [WRITE(x, y)]] SN(write)
It is of course not coincidental that the output of the GPF application corresponds exactly to the content of the lexical entry of writerN (Spencer 2013: 199–200). With regard to the typology of lexical relations that the four-part lexical entry predicts, Spencer cites two main considerations: (1) the persistent problem that between clear cases of inflection and clear cases of derivation exist many less easily distinguished types of relatedness, and (2) the insights of Canonical Typology (Corbett 2009, 2010; see §4.1 below) in which the space for variable instantiations of feature combination is laid out on logical grounds before the first example is categorised. Table 2.20 shows the logical combinations of rule-induced changes that the GPFM architecture of lexical entries permits. Table 2.20 Typology of lexical relations. (Source: Spencer: 2013: 7, © Oxford University Press) FORM
SYN
SEM
LI
identity logical impossibility
− −
− −
− −
− +
inflection
+ − + − + − +
Same lexeme: − + + − − + +
− − − + + + +
− − − − − − −
derivation
+ − + − + − +
New lexeme: − + + − − + +
− − − + + + +
+ + + + + + +
72 Contemporary Morphological Theories
Of the sixteen matrices that may be formulated in this scheme, one is simply the identity relation (four ‘−’, or ‘no change’ marks); another is a change (marked ‘+’) only in the LI attribute, which Spencer characterises as ‘logically impossible’, but this is not so. It is merely pernicious, because the derived result would be a lexeme fully identical to its ‘base’ in form, syntax, and semantics, but with a covertly d ifferent identity in the lexicon.30 For the rest, the range is from a change in form only (clear inflection) to a change in all four attributes (clear derivation), with the crucial attribute change that divides the two being that of LI. Representing ‘change/no-change’ as a binary abstracts over many types and degrees of change, but the system as designed does allow for a systematic categorisation of lexical relations. Spencer (2013) provides examples of each of the ‘possible’ relations in turn in order to demonstrate the plausibility and utility of the GPFM model. Although PFM has been compared to A-Morphous Morphology (§2.1) as reflecting a similar theoretical perspective, a much closer affiliation is to be found with Network Morphology (§2.11) in the shared reliance on features, defaults, lexical classes and subclasses, and the paradigm as an organising principle. One clear distinction, however, is PFM’s directly tying rule blocks to (linear) position classes, whereas this does not seem to be a prediction in the Network Morphology approach. An empirical examination in this area should prove a useful line of study to determine whether such an assumption is fully tenable. Recommended reading for Paradigm Function Morphology Spencer (2013) Stewart and Stump (2007) Stump (1990) Stump (1991) Stump (1992) Stump (1993a) Stump (1993b) Stump (1995) Stump (2001). 2.13 P R O S O D I C M O R P H O L O G Y Table 2.21 Continuum table for Prosodic Morphology. Morpheme-based Formalist In-grammar Phonological formalism Incremental
Ꮻ Ꮻ
Ꮻ Ꮻ Ꮻ
Word/Lexeme-based Functionalist In-lexicon Syntactic formalism Realisational
Theory profiles 73
Prosodic Morphology is an outgrowth of Autosegmental Phonology (Goldsmith 1976). Proposals made in McCarthy’s (1979) analysis of Classical Arabic, and distilled somewhat in McCarthy (1981), gave rise to an approach that escapes the limitations of the two-dimensional trees of Word Syntax (§2.15 below). In McCarthy (1981) the task is to accommodate non-concatenative morphology within the same basic scheme as concatenative morphology. In order to accomplish this, McCarthy invokes the abstract multidimensional representations, or tiers, found in Autosegmental theory. If every morpheme is represented on its own tier, root and non-root morphemes are more parallel at the formal level. The asymmetry comes in the form of a prosodic skeleton, to which the segmental and/or featural content of the morphemes is mapped on an independent basis. This allows for the retention of discrete morphemes, while allowing the parts of these morphemes to appear discontinuously in the output string, a result not readily permitted in representations of two dimensions (or fewer). Thus, for example, in the Classical Arabic form kattab ‘cause to write’ (perfective active), the morphemes are /k-t-b/ ‘write’, /-a-a-/ (reducible to /a/ under assumptions of spreading) [perf. act.], and the abstract template CVCCVC [causative]. McCarthy (1981: 385) exemplifies fifteen abstract morphological classes for the (majority) triconsonantal roots of Classical Arabic, choosing to refer to the classes by the established Hebrew term, binyan(im). Whereas Autosegmental Phonology upholds as the ‘unmarked’ association pattern ‘one-to-one, then spread from the last attached segment to fill the remainder of appropriate slots’, the analysis in McCarthy (1981) requires a number of ‘marked’ patterns of association between segments and the skeletal slots, for example, in cases where the middle of three consonants spreads, permitting kattab, mentioned above, in lieu of the unattested *katbab that unmarked association would predict. The device of pre-association is used in order to allow for certain overrides of the unmarked association patterns, whereby one might declare, for instance, ‘attach edge segments, then fill remainder by spreading as yet unattached segments’. McCarthy proposes this possible pattern, with the functional explanation that failing to ensure that at least edge elements are attached before spreading may have the consequence of obscuring the root’s identity.31 Marantz (1982) capitalises on the descriptive success of McCarthy’s framework, testing it on reduplication data from several languages. Whereas McCarthy (1981) used the skeletal tier as a sort of output template to be filled in, Marantz (1982) suggests that affixes as well as stems can be segmentally under-specified, that most reduplication processes are best analysed as the affixation of a consonant– vowel (C–V) skeleton, itself a morpheme, to a stem. The entire phonemic melody of the stem is copied over the affixed C–V skeleton and linked to C and V ‘slots’. (437) In assuming a complete copying operation from which ‘leftovers’ can be ‘stray-erased’, and within which the segments or features can be overridden by pre-attached values, Marantz (1982: 444) might seem to be opting for an excess of descriptive power.
74 Contemporary Morphological Theories
Given that there are indeed languages which use total r eduplication, however, positing a single universal operation based on the limiting case is actually conceptually simpler. The fact that other languages reduplicate no more than one or two segments in all cases undercuts the universal appeal somewhat, but there is a case to be made either way. A real advantage of Marantz’s (1982) presentation is the involvement of a richer and independently motivated prosodic hierarchy (also developed in Halle and Vergnaud (1980)) in the description of the different abstract shapes that affixes can take. The limiting case, ‘normal affixation’, is the addition to a stem of a morpheme which is fully specified, all the way to the segmental level, borrowing nothing from the content of the stem (Marantz 1982: 456). Yidiny reduplication, fully under-specified at the segment level, copies the first two syllables of the stem, regardless of their C–V composition (453). The more frequently encountered reduplication types are somewhere in the middle, then, with a specific C–V skeleton, and perhaps some limited segmental and/or featural pre-association (449). From this perspective, morphological operations and different morpheme types are formally united in a plausible way. Perhaps the start from non-concatenative processes led to this more even-handed treatment of concatenative and non- concatenative morphological patterning. The skewed relative frequency of ‘normal affixation’ versus the much less common reduplication cross- linguistically is unpredicted, however. Akinlabi (1996) gives an indication of the survival of the approach into the Optimality Theory paradigm. Akinlabi, although dealing with putative morphemes which are no larger than features or sets of features, hopes to account for these as edge-oriented affixes. In the constraint-based framework of Optimality Theory (Prince and Smolensky 2004), constraints which align prosodically defined elements such as syllables, feet, and (prosodic) words are commonly employed to describe positional affinities between one level of the prosodic hierarchy and another when, all else being equal, independent positioning might be assumed. Akinlabi seeks to adjust the terrain, positing constraints which align particular morphemes to particular prosodic constituent edges (243): Featural Alignment Align (PFeat, GCat) A prosodic feature is aligned with some grammatical category. What this does not take into consideration, and what McCarthy (1981, 1982), Marantz (1982), and Halle and Vergnaud (1980) had likewise left unrecognised, is that morpheme is not obviously a part of the prosodic hierarchy. Because the phonological material in a given word owes its existence, in the general case, to some element of meaning or grammatical function, and there is therefore some dependency between a morpheme and its spell-out (‘Pfeat is the featural spellout (or content) of the morphological category in question’ (Akinlabi 1996: 243)), Prosodic Morphology sees no obstacle to positing a hierarchy:
Theory profiles 75
root > morpheme > syllable > C–V skeleton > segment > feature The comparability of morphemes and syllables is limited, since meaning attends the one but not the other. The question of where (or whether) to place ‘foot’ in the above shows the grafting of one dimension into another. To base an analysis on correspondences between the phonological and the morphological, especially when one is proposing constraints of a universal nature (as OT analyses explicitly presume), is to open oneself up to criticism of allowing too liberal a formal representation. These levels are not always spelled out exhaustively in the examples given in Prosodic Morphology (although McCarthy 1982: 213, for example, comes close), so it is easy to set aside the telescoping that is going on. For Akinlabi, the placement of a featural affix is part of the lexical entry of that morpheme; determining whether it is a prefix (i.e. placed relative to the left edge of the stem) or a suffix (relative to the right) is based on evidence for directionality of autosegmental association. A featural suffix, for example, will tend to have its effect at the right edge of the stem, but depending on the relative strength of feature co- occurrence constraints and faithfulness constraints, the suffix may be forced further in from the edge, or else be blocked from applying. This move is fairly ingenious within the formalism of OT, despite some speculative underlying assumptions. As a brief example, Chaha labialisation is claimed to be a featural suffix, realised on the rightmost stem consonant which has the potential to be labialised (coronal consonants may not be labialised, although labial and velar consonants can). The feature links once only, to the rightmost licensing consonant, able to link to an initial consonant only if the subsequent consonants in the word are all coronals. In case of a stem with only coronal consonants, the feature does not link and thus is not phonetically realised. Given these details, a constraint hierarchy of *COR/LAB >> PARSE >> ALIGNR emerges (249). A particularly strong argument in support of this approach is the factorial typology given and exemplified in an appendix to the article (283): Table 2.22 Factorial typology of constraint rankings. (Source: Akinlabi 1996: 283, © Cambridge University Press) Co-occurrence Co-occurrence Parse Parse >>a>> Alignment
Alignment Parse Co-occurrence Alignment Co-occurrence
Alignment
Parse
Parse Alignment Alignment >>b>> Co-occurrence Parse Co-occurrence
Nuer continuancy Chaha labialisation Japanese mimetics Aka voicing Athapaskan [–continuant] Aka voicing Zoque palatalisation
Except for the ‘Japanese mimetics’, however, one would hardly know from the phenomena named that this was a morphological analysis. The categories realised by the various featural affixes are backgrounded throughout the article, in an effort, it would seem, to amplify the phonological aspect of the model. Simultaneously, therefore, Akinlabi (1996) displays the inheritance from the
76 Contemporary Morphological Theories
earlier work in Prosodic Morphology and leaves intact a more concatenative ideology for morphology. Prosodic Morphology, although undergoing some significant transformations in its transition into constraint-based (OT) analyses, is an approach that the phonologically minded may take to readily. Despite the several caveats in the above, there is clearly value in this method of representing the phonological aspect of morphology. One must remember, however, that the insights of multi-tiered representations can collapse (for better or worse) into the same plane if viewed from a different angle.32 Recommended reading for Prosodic Morphology Akinlabi (1996) McCarthy (1981) McCarthy (1982) McCarthy and Prince (1993) McCarthy and Prince ([1993] 2001) McCarthy and Prince (1994) Marantz (1982). 2.14 W O R D -B A S E D M O R P H O L O G Y Table 2.23 Continuum table for Word-based Morphology. Morpheme-based Formalist In-grammar Phonological formalism Incremental
Ꮻ Ꮻ
Ꮻ Ꮻ Ꮻ
Word/Lexeme-based Functionalist In-lexicon Syntactic formalism Realizational
There is a break in descriptive tradition that can be located with respect to the model put forward by Bloomfield (e.g. 1933), departing from what had been customary in pedagogical and reference grammars, most saliently the traditional grammars of Classical Greek and Latin. This method of presentation has been called Word and Paradigm (WP) morphology, owing to its custom of focusing on inflectional classes by means of (1) presenting exemplary paradigms of each class’s pattern of forms, and (2) within grammatical categories, abbreviating the patterning down to a small subset of so-called principal parts, which together could serve as reliable predictors of the paradigmatic distribution of forms, and therefore of inflectional class. In this way, language students (and presumably, by extension, speakers in general) would not need to memorise all of the inflected forms of all of the inflectable words. The necessary analogies for deducing needed word-forms are ‘resident in existent form inventories’ (Blevins 2005: 20). The status of WP as ‘standard and traditional’ seems to have been resemanticised as ‘musty and antiquated’ in the Structuralist and Generative environments that took
Theory profiles 77
centre-stage in the twentieth century. As descriptive grammars became more about cataloguing ‘recurrent partials’, reifying each as a sign-like morpheme with a consistent – if sometimes strategically vague – significance, and recasting analogical relations in the format of rules, the dynamic of building with atomic elements left the more static, even monolithic, perspective of consulting trusty, unchanging exemplars as a method fit for storage among the mothballs. Indeed, the WP mode of presentation spoke only occasionally, and always from the margins, as morphology first went morphemic, and subsequently slid by and by into the adjacent domains of phonology and syntax. In Hockett (1954: 210) there is mention of the time-honoured WP approach, albeit as an afterthought, in an article aimed at contrasting the ascendant Item and Arrangement (IA) and slightly less ascendant Item and Process (IP) approaches. In Robins (1959), a reply to Hockett (in part) and to trends in morphology in general, the case is made that WP, if properly ‘defended’, could be just as adequate as IA or IP to the descriptive enterprise. Matthews (1972) takes pains in the description of Latin verb morphology to demonstrate problems incurred by the insistence on morpheme- based analysis, problems that are a direct consequence of dealing in biunique, indissoluble form–meaning pairings, and which a WP approach does not invite.33 The above historical preamble has been introduced in this section in light of the relative dominance of non-WP approaches in the morphological literature more widely. In a series of articles, Blevins (e.g. 2003) has mounted an elaboration of what he refers to regularly as ‘traditional WP analysis’. This Word-based Morphology programme (Blevins 2006) is by no means a mere revival of the forgotten wisdom of ancestors, however. The arguments for a relevant and modern deployment of WP thinking as a theoretical framework for morphology depend on demonstrating not only the conceptual and empirical shortcomings of competing IA and/or IP treatments, but also the sufficiently frequent empirical need for the machinery assumed in WP analysis, but absent from IA and IP. Blevins locates the key source of descriptive problems in morphological theorising as inhering in the ‘post-Bloomfieldian’ morpheme-as-sign, on the one hand, and as deriving from a too-literal interpretation of the metaphor of grammar as the constructor of each individual word from minimal pieces, on the other. In contrast with constructive approaches to word-structure which show the influence of the dynamic of building up from a core, or root, unit incrementally, either through the attachment of affixes (IA) or through the application of (sequences of) processes (IP), Word- based Morphology takes a distinct tack in assuming an abstractive perspective in which it is the set of fully inflected words in a language that are basic in its morphology, and that subword units such as roots, stems, and exponents constitute recognisable patterns to be abstracted from the inventories of full word-forms. In this way, the business of morphological analysis becomes not segmentation and categorisation, but rather the identification of paradigmatic patterns among inflectionally related word-forms (declensions and conjugations) and, within these patterns, reliable implicational relations that permit, based on the existence of individual forms, the prediction of one or more other forms in the paradigm. This is the standard function of the principal parts of lexemes in reference and pedagogical grammars – namely
78 Contemporary Morphological Theories
to signal inflection-class membership – and from which information the remaining forms may in the general case be inferred without explicit listing. There are often language-specific or language-family-specific traditions of selecting the forms to be used as reference points: for example, Latin noun declension class membership may be identified via joint inspection of a given noun’s nominative singular and genitive singular forms – compare amicus ~ amīcī ‘friend’ (second declension) with passus ~ passūs ‘pace’ (fourth declension). Word-based Morphology recognises the spirit of the principal parts approach, but rather than adopting received selections wholesale, there is room to evaluate relative predictiveness among the members of particular classes and subclasses, such that a pairing that is very informative for one inflectional class may need more, or less, or different support from other forms in providing full and reliable prediction of all forms in the paradigm. For example, Xūs as genitive singular is unique to the fourth declension in Latin, and is therefore sufficient to provide ‘a “hook” into a deductive pattern’ (Blevins 2005: 14). On the other hand, a genitive singular Xī is potentially ambiguous between the second and fifth declensions, and thus the nominative can yet become decisive. The consideration of principal parts is informed by a central role for analogy in Word-based Morphology. The relations among paradigmatic word-forms are analogical in nature, construed in terms of (information-theoretical) predictiveness,34 rather than constructive activities of derivation. In view of the predominance of rule-centric discourse, Word-based Morphology’s choice to place generative rules in the position of epiphenomena, of convenient encapsulations of generalisations across whole words, may strike many as a wilful missing of said generalisations. In fact, the generalisations are not missed, but rather they reside in the forms that instantiate them. They are available for abstraction, for statement as implications among forms and characteristic of paradigmatic classes. The generalisations therefore are not the familiar ‘instructions for assembly’, and subword pieces are not candidates for lexical entry status in their own right. In cases of maximally morpheme-like behaviour – that is, transparency, segmentability, invariant form, and mono-functionality, or in other words, concatenation of the biunique – the cognitive impulse to factor out recurrent partials is strong indeed, and establishing a morpheme-based morphotactics is an attractive option for the purpose of description. The deviations from biuniqueness are many and widespread, and these have been touched on at a number of points throughout the present book. Not only is there a challenge to the construct of the morpheme in WP morphology, but furthermore, the practice of deriving full words in isolation from each other (see DM §2.5 and Word Syntax §2.15) is found to be inadequate in the description of languages such as Estonian (Blevins 2004, 2005), which require reference to combinations of stem and exponent within paradigmatic structure for an adequate assessment. The morpheme-based approach, which assumes that the respective properties of a word-form’s constituent morphemes will sum to the properties of the word-form, runs aground in describing Estonian, which makes nearly maximal use of a limited number of formal exponents in distinct combinations to fill out the nominal paradigm, as in the following:
Theory profiles 79 Table 2.24 Declension for Estonian nouns hekk ‘hedge’ and kool ‘school’. (Source: Blevins 2005: 2–5)
Nominative Partitive Genitive Illative Inessive Elative Allative Adessive Ablative Translative Terminative Essive Abessive Comitative
Singular
Plural
Singular
Plural
`hekk `hekki heki hekisse hekis hekist hekile hekil hekilt hekiks hekini hekina hekita hekiga
hekid `hekkisid `hekkide `hekkidesse `hekkides `hekkidest `hekkidele `hekkidel `hekkidelt `hekkideks `hekkideni `hekkidena `hekkideta `hekkidega
`kool `kooli kooli koolisse koolis koolist koolile koolil koolilt kooliks koolini koolina koolita kooliga
koolid `koolisid `koolide `koolidesse `koolides `koolidest `koolidele `koolidel `koolidelt `koolideks `koolideni `koolidena `koolideta `koolidega
A preceding grave accent (`) indicates the presence of a stem with ‘strong’ prosodic status, referred to as ‘overlong’ or ‘Q3’ (Blevins 2005: 2, 22, fn. 1).
In the Estonian system, shared form does not entail, or even imply, shared properties. The same stem is used in realising semantically distinct word-forms, and the same can be said for theme vowels and inflectional suffixes. It is the combination of these units that permits the identification of the form. Attempting to force a morpheme-based analysis on this array of forms forces incoherent, mix-and-match allomorphic statements, and at the same time, substantial morphemic homophony. One may find ‘recurrent units of form . . . at every level’ but no morphemes as such, because the ‘subword units cannot be brought into correspondence with grammatical properties’ in that they ‘bear no consistent meaning in isolation’ (Blevins 2005: 2). The redeployment of formal units for contrastive functions is called syncretism, and although this term is most frequently applied at the whole-word level, it may equally describe multiple uses of a stem or other formal unit. A framework that assumes biunique form–meaning units (morphemes) as basic confronts syncretism as problematic, and considerable effort may be required to bring syncretic relations into compliance with morpheme-based expectations. Word-based Morphology, in assuming that whole words are the basic units, and that paradigmatic relations are significant rather than incidental, does not predict that form and meaning are bound together at a fundamental level (the by now familiar Separation Hypothesis, cf. LMBM §2.6).35 In the case of Estonian, the distribution of formal elements involves considerable syncretism, but this state of affairs does not subvert grammatical description. On the contrary, the pattern becomes straightforward once the grammatical properties of individual forms do not inhere in the subword elements but rather are ascribed to the cells of the paradigm that the full word-form occupies. The predictive pattern across noun stems in Estonian is as follows:
80 Contemporary Morphological Theories Table 2.25 Predictive patterns in Estonian noun inflection. (Source: Blevins 2005: 1) Nominative sg.
Partitive sg. ‡
Genitive pl. ‡ Semantic pl.
Genitive sg. ‡ Partitive pl. ‡ Nominative pl. ‡ Semantic sg.
A description based on forms, rather than on inherently meaningful forms, can be stated without undue violence to the constituent units, i.e. explaining additions, deletions, or inversions of morpheme-meanings. In short, the word-forms of Estonian are constituted by systematic arrangements of morphomic units of the sort posited by Aronoff (1994; cf. Matthews 1972), units identified by the full word-forms that use them, rather than by what they may be said to mean in and of themselves. Word-forms in this model are built with crucial reference to the forms of other, related words, not through the invocation of other lexical entries, grammatical properties and all. In its embrace of the paradigm, Word-based Morphology is able to invoke evidence that accords with analogical relations such as token-frequency within inflectional paradigms and type-frequency for so-called derivational word-families (cf. Network Model §2.10). Blevins (2006: 535) cites psycholinguistic evidence that constructive approaches do not predict: namely, bidirectional priming effects between morphologically complex and simple related words, which imply that complex words, not just mono-morphemic units, may be stored in memory if they exceed a certain frequency threshold (Bybee 1985, 2007; cf. Clahsen 1999; Pinker and Prince 1994). The issue of psychological plausibility is not skirted or evaded, as one might expect in a framework that focuses on complexes (word-forms) and complexes of complexes (paradigms), appealing to each as somehow basic. On the contrary, it is not assumed that paradigms are memorised, but rather that they are deducible by analogical comparison to principal parts and exemplary paradigms. Anticipating the counter-argument of the source of analogical models in non- pedagogical contexts, it is not necessary for speakers to share a common exemplary paradigm for inflection classes to be effectively deduced; in fact, there is no condition that a class have, nor a speaker consult, only one model per pattern (mutatis mutandis for principal parts, as mentioned above). The preference for single exemplars is a by-product of exposition, as is appropriate to the compact aesthetic of pedagogical and reference grammar contexts. While the particular set of assumptions that comprise the Word- based Morphology model results in a less streamlined descriptive system than one with atomic morpheme units and rule-based derivations, the WP method can account for classical affixation as well as many other attested systems that require morpheme-based approaches to tack on devices such as inflection-class diacritic features, metafeatures, blocking mechanisms, and/or property-manipulating rules (cf. DM §2.5). The WP approach in this instantiation can be sensitive to intra-and cross-paradigmatic patterns that are not evident in the dissection of any single form,
Theory profiles 81
individually constructed. Blevins (2005: 7) is thus able to contrast the ‘signal’ of WP, the substance that may be perceived from that perspective, with the ‘noise’ in a post-Bloomfieldian analysis, the messy and unaccountable deviations from a one- form/one-meaning ideal. Recommended reading for Word-based Morphology Blevins (2001) Blevins (2003) Blevins (2005) Blevins (2006). 2.15 W O R D S Y N T A X Table 2.26 Continuum table for Word Syntax. Morpheme-based Formalist In-grammar Phonological formalism Incremental
Ꮻ Ꮻ Ꮻ Ꮻ
Ꮻ
Word/Lexeme-based Functionalist In-lexicon Syntactic formalism Realisational
The approach to morphology called Word Syntax has a special position in linguistic theory, especially in the area of GB-style syntax and its descendants. It owes much to the classic Item and Arrangement (IA) approach (Hockett 1954). In Word Syntax, morphemes are the essential building blocks of words. Bound morphemes differ from free morphemes solely in that the bound morphemes subcategorise for a stem of a certain category to attach to. The name Word Syntax is an obvious choice for this approach, because one need only glance at an analysis to see the overt parallels being drawn between words and phrases. Lieber’s (1981) dissertation is held up as an example of the Word Syntax movement in its crystallising phase. Morphology from this perspective is first and foremost about the concatenation of discrete meaningful units – namely, morphemes – and the binary-branching tree structures constitute an account of how a morphologically complex expression comes to have the meaning and morphosyntactic features it does. In Lieber (1981) and in Williams (1981), much attention is paid to the notion of headedness in morphologically complex words. Williams (1981: 248) proposes that the rightmost morpheme at any level of morphological concatenation is the ‘head’ of the construction (his Right-hand Head Rule, or RHR), i.e. that for any concatenation of two morphological elements, the element on the right determines the category and attributes of the resulting expression. Lieber (1981) proposed the mechanism of feature percolation as the means of transmitting attributes from a constituent morpheme upward to a larger construct. Williams’s RHR ‘works’ for much of English derivational morphology
82 Contemporary Morphological Theories
and compounding, since English endocentric compounds are almost invariably right-headed, and since nearly all category-changing affixes in English are suffixes. There nevertheless exist a number of languages with systematic left-headed compounding (Italian, Scottish Gaelic, Vietnamese) and even within English there are a few category-changing prefixes ([en-[noble]A]V). Inflectional affixes in English are invariably suffixal, but part of the definition of inflection is that it cannot change the category of the word to which it applies. The Spanish diminutive suffix -ito/-ita can attach to nouns, adjectives, and adverbs, producing in each case a semantic change only, crucially being transparent to the category of the word it attaches to, quite unlike a head is predicted to behave (Di Sciullo and Williams 1987: 26; cf. Jaeggli 1980). The RHR is clearly not adequate as a general principle of morphology, but perhaps a revision could redeem it? Lieber (1981) and Selkirk (1982) both reject the RHR, as originally defined, as simply being too strong. They both suggest alternatives to strict right-hand percolation, allowing for so-called ‘back-up percolation’ in cases where the whole expression has attributes present in some non-head morpheme but not present in the head (Selkirk 1982: 76): Percolation (revised) 1. If a head has a feature specification [aFi], a≠u[nspecified], its mother node must be specified [aFi], and vice versa. 2. If a non-head has a feature specification [bFj], and the head has the feature specification [uFj], then the mother node must have the feature [bFj]. This allows non-head features to be percolated to the construct, but only if the head has no non-null specification of its own to contribute for the feature in question. Prefixation is still a potential problem if multiple prefixes were to have conflicting specifications for the same feature. It may be that this situation never arises, especially if we assume strict binary branching, but there is nothing to rule it out in principle. In Di Sciullo and Williams (1987: 26), acknowledging some serious empirical problems for the RHR as originally defined, a relativised notion of head is put forward: ‘The headF of a word is the rightmost element of the word marked for the feature F’ (26). This permits a multiply affixed word to have several heads simultaneously, effectively allowing any morpheme to determine some categorial quality of the derivative. Prefixes cannot determine category, however, because the root is always to the right of them, and the root is always marked for at least grammatical category. Thus examples like ennoble still are unexplained, and the predictiveness of the original hypothesis is severely weakened. As for left-headed compounds, Di Sciullo and Williams claim that such constructions in Romance languages are not really compounds, but rather they are ‘phrases reanalysed as words’ (83, contra Selkirk 1982: 21). The argumentation is less than conclusive, given the semantic idiosyncrasy of some of the expressions and the failure of agreement in at least some cases. The bottom line for Di Sciullo and Williams’s proposed amendments to those of Lieber (1981), Williams (1981), and Selkirk (1982) is a weaker model overall, and a smaller – but enduring – empirical problem.
Theory profiles 83
Fabb (1988a) proposes doing almost all word formation in the syntax (at least all productive affixation; cf. MM §2.8), with separate affixal nodes in the phrase marker, and concatenation via head movement. Di Sciullo and Williams (1987: 87) disapprove of such intermingling of syntax and morphology as engendering a loss of generality in both morphological and syntactic rules. Developments in GB syntax converged with the idea of inflection in syntax, such that verb inflection (and sometime noun inflection as well) is performed (or, alternatively, ‘checked’) by the movement of lexical heads through a sequence of functional heads, each of which contains a morphosyntactic value appropriate to the clause in question, and often associated with overt inflectional morphology. Once head movement is complete, an inflected lexical head appears in S-structure as input to Phonological Form (PF). Whether the affixes are actually represented in the tree structure under the appropriate functional heads is a decision not without implications (cf. DM §2.5 above). The so-called Lexicalist Hypothesis (Chomsky 1970) makes a qualitative distinction between syntax and the internal structure of words. Despite some formal similarities including apparent hierarchical relations among at least the derivational morphemes in a morphologically complex word, syntax does not have access to, and therefore cannot make reference to, any internal structure of the words which might appear in syntactic constructions. This point is recast in Selkirk (1982): ‘The category Word lies at the interface in syntactic representation of two varieties of structure, which must be defined by two discrete sets of principles in the grammar’ (2). ‘Doing affixation in the GB syntax’, as Fabb (1988a) would have it, is clearly contrary to the Lexicalist position. In 1992, Lieber re-entered the fray with an overtly syntactic approach to word formation. Specifically in response to the lexicalised phrases – which Di Sciullo and Williams (1987) looked to as a safety-net against the falsification of the RHR – Lieber sees a need to intermingle principles of phrase-building and principles of word-building (21). Again it is claimed that all morphemes have lexical entries, and most, if not all, have syntactic categories of their own. Morphemes are thus X0 elements to be inserted in syntactic tree structures. Allowing for unlimited recursion at the X0 level, any number of morphemes may be adjoined into a complex X0 without untoward results in the X-bar syntax (37). The assimilation of morphology to syntax is fairly completed by the introduction of the notions of complement, specifier, and modifier morphemes. Alongside the existing notion of head, Lieber assumes that parallel terms mean parallel behaviour ‘above and below the word level’ (39). Some conventional parameter settings found in syntax are modified and rebranded Licensing Conditions (38): 1. Xn *. . . X(n-1, n). . ., where recursion is allowed for n=0. 2. Licensing Conditions a. Heads are initial/final with respect to complements. i. Theta-roles are assigned to left/right. ii. Case is assigned to left/right. b. Heads are initial/final with respect to specifiers. c. Heads are initial/final with respect to modifiers. 3. Pre-or post-head modifiers may be Xmax or X0.
84 Contemporary Morphological Theories
With the above as general conditions holding of morphemes as well as words in this expanded view of syntax, the onus is on Lieber to demonstrate that full parallelism obtains. The cost of maintaining this assumption, however, is a series of ad hoc replies to empirical problems: 1. English synthetic compounds are left-headed because the construction is a holdover from Old English, when the parameter-settings were different (62–3); 2. Right-headed compounds in French (the only kind that matter, according to Di Sciullo and Williams (1987: 83–6)), such as radioactivité, are dismissed as non-productive, learned, neo-classical vocabulary, with no bearing on parameter-settings (66); and 3. Variable adposition patterning in Dutch is the result of treating the parameters as defaults rather than as true parameters (70–1). The resulting correspondence between phrasal and word syntax is rough at best. The predictions which follow from Lieber’s assumptions are quite strong, if the specifier and complement morphemes she sketches can be maintained. The Word Syntax approach to morphology has the formal advantage of making morphology similar or identical to the independently motivated syntactic component. The greater the insistence on assimilation, however, the more adjustments and riders there are to be included in the statement of syntactic rules and principles. Giving each morpheme a lexical entry, but at the same time suggesting that the lexicon is no more structured than a random collection of such entries (Lieber 1992: 21)36 makes one wonder what the lexicon is really good for, other than standing as a legitimiser for the putative equivalence of all morphemes, bound or free.37 As was mentioned in the introduction to this section, the Word Syntax framework has had considerable influence on the treatment of morphology within the GB syntactic framework. If one is working in the GB/Minimalist framework, Word Syntax might be the most natural choice (but compare DM §2.5 and MM §2.8 above). Recommended reading for Word Syntax Di Sciullo and Williams (1987) Fabb (1988a) Lieber (1981) Lieber (1992) Selkirk (1982) Toman (1998) Williams (1981). 2.16 O V E R V I E W Having sampled the tenor and key concerns of these frameworks for ‘doing morphology’, we have seen some points of convergence and some near-diametric oppositions. Table 2.27 compiles the continuum profiles from this chapter to offer an overhead
Autolexical Categorial DM Network_M PFM Word Syntax
DM Word Syntax
LM&P/StratalOT Prosodic
LM&P/StratalOT MM Word Syntax
Formalist
In-grammar
Phonological formalism
Incremental
Autolexical Categorial NM Prosodic
A-Morphous NM
NM Prosodic
A-Morphous LM&P/StratalOT Prosodic
LM&P/StratalOT NM
Net_Model
Construction LMBM MM Net_Model PFM Word-based
A-Morphous Categorial LMBM MM
LMBM MM Word-based
Autolexical Categorial MM
(Net_Model = Network Model; NM = Natural Morphology; Network_M = Network Morphology)
DM Prosodic Word Syntax
Morpheme-based
Table 2.27 Unified continuum table.
DM LMBM
Autolexical DM Network_M
Autolexical Construction LM&P/StratalOT
Construction
Construction LMBM Net_Model Network_M
A-Morphous Construction Network_M PFM Word-based
Categorial Word Syntax
Net_Model Network_M PFM Word-based
NM Net_Model
Word-based A-Morphous PFM
Realisational
Syntactic formalism
In-lexicon
Functionalist
Word/Lexeme- based
86 Contemporary Morphological Theories
erspective, as it were, on the lay of the land. From this vantage, choices among p the frameworks for applications to new phenomena may become easier to make. Compatibility with the reader’s own views on the architecture and operation of grammar more generally can likewise be assessed, so that the tool fits the hand, as well as the task. NOTES 1. Halle and Marantz (1993: 112–13) are baffled by this move and fundamentally misappraise the claims made by Anderson (1992). The confusion arises out of their own initial assumptions (DM, §2.5): namely, that morphosyntactic information enters into syntactic structures only through the insertion of meaningful pieces (i.e. morphemes). As a result, and contrary to Anderson’s proposal, they hold that morphemes must in fact be included in Anderson’s structural representations, and the corresponding phonological representations then removed, and later reinserted into the derivation. If this were the actual proposal of A-Morphous Morphology, it would indeed be inelegant. 2. The name Extended Word and Paradigm is perhaps at odds with the relatively limited usefulness Anderson (1992: 79) ascribes to the notion paradigm. Paradigmatic (contrastive) relationships, whether inflectional or derivational, lie in the background while stems, words, and formation rules occupy primary positions. 3. In this regard, the entries that are assumed in Automodular Grammar are largely homologous with the Attribute-Value Matrices (AVMs) used in Head-driven Phrase Structure Grammar (HPSG; Pollard and Sag 1994). Compare the notion of hierarchical lexicon developed there, and as it is deployed in Construction Morphology (§2.4). Both Sadock (2012: 6–7) and Booij (2010: 5–16) see for their respective morphological approaches an affinity with the modular framework known as the Parallel Architecture (Jackendoff 2002; Culicover and Jackendoff 2005), which all parties acknowledge as standing in need of a theory of morphology (Culicover and Jackendoff 2005: 545). Whereas Sadock (2012) implies that the Parallel Architecture, while ‘fully modular’ (distinguishing phonological, syntactic, and conceptual structures), does not subdivide as finely as it might, Booij (2010) draws the inference that morphology ‘is not a module of grammar on a par with the phonological or the syntactic module that [each] deal with one aspect of linguistic structure only’ (11). In Construction Grammar more broadly, however, it is indeed questionable whether any aspect of the grammar is to be considered autonomous. 4. This approach clearly implies at least one phonological module, but this is not the concern of Sadock (2012), in which there is no entry in the index for phonology. Its responsibilities, units, and rules would have to be worked out by analogy from the Automodular architecture and descriptive method. 5. The Interface subsystem of Autolexical Syntax was qualitatively different from the modules of grammar (Sadock 1991: 36), but it has been promoted to the status of fellow component in Automodular Grammar (Sadock 2012: 24). 6. The issue of productivity (see §4.2) is acknowledged here as ‘notoriously difficult’. Automodular Grammar is far from the only framework to establish a core and a margin of morphology, devoting attention to the former while minimising the tractability of the latter, and massaging intermediate cases with the promised existence of ‘clear cases of both’ productive and unproductive patterns (Sadock 2012: 148). 7. Compare the analysis in Prosodic Morphology (§2.13), in which both root and non-root are morphemes. 8. In fact, Embick and Noyer (2007: 296, fn. 8) equate ‘abstract morpheme’ and ‘functional head’.
Theory profiles 87 9. MS here is fully distinct from the Morphological Spelling (MS) component of LMBM (§2.6). 10. It would seem to be important in this enriched virtual workspace that processes manipulating structure be as limited as possible, in order to ensure that the theory not lose predictiveness: that is, not become vacuous in that there is nothing that it rules out. 11. The list would, not coincidentally, comprise the so-called ‘strong’ verbs. Aside from the lack of suffix – or ‘presence’ of zero – the list otherwise shows considerable diversity in the [+past] forms, ranging from identity with the presumed ‘root’ (beat) to vowel alternations of various sorts, which DM ascribes to the operation of readjustment rules. These rules will have their own lists of verbs to which they apply, partly overlapping with environment of the zero-suffix realisation of [+past], and partly overlapping with the verbs that receive the ‘context-free’ dental suffix, jointly defining the so-called ‘mixed’ verbs in English. 12. With regard to the absolute default zero that, as stated, should be insertable whenever no other affix is called for, the analytical problem of permitting inaudible affixes in representations (how many are there? where are they?) was checked by a stipulation discussed in American Structuralist morphology, ruling out putative zeroes that have no non-null allomorph: that is, that are always null (Bloch 1947: 402; Hockett 1947: 322–3, 340). Halle (1997: 426, fn. 1) names NULL as ‘the phonological marker for the singular of English nouns’, as had Harris (1942: 172) in the article to which Bloch and Hockett had reacted. Positing a context-free insertion of zero in this way amounts to the deliberate option to tolerate an untold (and perhaps untellable) distribution of such structural zeroes. The lack of a formal indicator of position, i.e. a leading or trailing hyphen, in the rule glosses over the question of placement. The additional detail that, according to the rule, the zero both implies and is implied by ‘nothing’, and therefore everything, is a further undesirable feature of this aspect of the analysis. 13. Earlier work on this theory indeed used ‘based’ in the name, but in subsequent work (e.g. Beard 1995), an increasing role for the base component in the architecture of the theory led Beard to alter the name, although the change has not been uniformly taken on board. 14. Beard (1995: 54–5) casts doubt on Bybee’s (1985) Relevance Hierarchy as a universal category order (see §2.10), but at the same time he has his own universal set of categories to propose; this can hardly be a coincidence. 15. See note 9 above regarding differing MSs. 16. Zwicky (1987a) distinguishes among abstract operation types that may be involved in morphological expression (e.g. prefixation, vowel substitution, reduplication), specific operations that a particular language may happen to make use of (e.g. the different uses of -s suffixation and of -er suffixation in English), and the morphological rules that may invoke one or more formal operations, potentially of different types. Thus the morphological rules in a language may separately make use of the same operation to realise distinct feature content. This separation specifically removes the expectation that homophonous operations are functionally tied together, and furthermore it renders less surprising acquisition facts like the asynchronous mastery of English [plural], [possessive], and [3sg Pres], despite the homophonous marking (including distribution of allomorphs). 17. Trisyllabically shortened, to be sure (grave → gravity). 18. Compare the similar use of stipulated lists of legal inputs in DM (§2.5). Such lists are regularly abbreviated to two or three sample roots for reasons of preserving a manageable exposition, of course, but the practice carries the added benefit of visually minimising the degree of stipulation implied by such a strategy. 19. In Wunderlich (1996: 97), the category Aspect linearly precedes Tense.
88 Contemporary Morphological Theories 20. Fabri et al. (1996) describe a computational implementation of MM paradigm construction as a recursive process in light of the architecture and minimal entity-types assumed in the framework, the hierarchy of inflectional features, radical under-specification of said features, and the constraining force of the listed principles. 21. Dressler follows Trubetzkoy in haplology, that is, merging two identical syllables in sequence, to compress mor[phopho]nology. Hap[lolo]gy itself remains, perhaps temptingly, as it is. 22. The reader is referred to Dressler (1985b: 265–79) for his fuller metatheoretical counter-arguments in this regard. 23. Parallels to HPSG are many here. The work of Riehemann (1998; see §2.4) is also compatible in its hierarchical approach to derivational patterns. 24. A Courier- style typeface is customary for DATR expressions used in Network Morphology. 25. Feature values stipulated at a node can also override default values, e.g. for N_IV (not shown), == (“” _a), which overrides the value (“” _i) it would otherwise inherit from NOUN (Corbett and Fraser 1993: 137). 26. This has the effect that IFD is not limited to any specific property specification, and therefore would apply only if there exists no applicable realisation rule keyed to a non- null property-set within the relevant rule block for the relevant lexeme-class. 27. If a stem is used in the realisation of all and only the occurrences of some morphosyntactic property, say {TNS:past}, it may be (mnemonically) useful to coin an index which reflects this use, i.e. identify a ‘past stem’. This does not, of course, entail that all indices for the particular stem set must bear functionally defined indices. From a realisational perspective, function-based names can give a (misleadingly) morphemic cast to an element of form. 28. The metatemplates, also known as meta-redundancy-rules in Janda and Joseph (1992a, 1992b, 1999) are rather parallel in function, stating formal similarities across exponents. They register the similarity (capturing a generalisation) without entailing the collapsing of distinct exponents or the rules that introduce them. 29. The format and operation of such a lexical entry is striking in its resemblance to the corresponding entries put forward in Automodular Grammar (§2.2). 30. This is undesirable from a functional perspective, and as a purely hypothetical exercise might be better termed a ‘doppelgänger’ or ‘evil twin’ lexeme, but it is common sense, rather than the logic of the typology per se, that rules such entities out of consideration. 31. This foreshadows the faithfulness and opacity concerns evident in his later Optimality Theory work (McCarthy 1982: 204–5, 213–14, 221). 32. Goldsmith (1976: 267ff.) characterises the task of the child during phonological acquisition as involving de-autosegmentalisation, transforming and interpreting autosegmentally structured representations into unidimensional linear representations. 33. The continued presence of WP is evident at several junctures in recent decades. Anderson’s (1977) ‘Extended Word and Paradigm’ (EWP) model is the predecessor of his (1992) A-Morphous Morphology (§2.1). Aronoff (1994) seeks to redirect attention to the necessity of acknowledging purely formal elements, stems in particular, that underlie morphosyntactically mixed collections of forms, promoting what Aronoff calls a morphomic level (‘Parasitic’ or ‘Priscianic’ formations in Matthews (1972: 86)), intermediate in the mapping between morphosyntax and phonological realisation, thus a level of relevance to ‘morphology by itself’ (25). In PFM (§2.12), Stump has emphasised the centrality of the paradigm in the organisation of inflectional morphology, drawing as well on some key WP methods. 34. In this vein, Finkel and Stump (2009), and Stump and Finkel (2013) strive to measure predictiveness, and from there to evaluate relative complexity of inflection class systems within and across languages. Stump and Finkel (2013: 6) allude to both the
Theory profiles 89 e xponence-based definitions of PFM (§2.12) and to the implicational approach found in Blevins (2006), suggesting that a hybridisation between the two could be profitable. 35. That is not to say that suppletion, or lack of formal correspondence among paradigmatically related words, is predicted in WP to be just as likely as consistent signification, but rather that the source of consistency in form-meaning association is diachronic descent from a common lexical source, whether continuing as a root or grammaticalised into a dependent element, clitic, or affix. It is not, on this view, the business of the grammar to create or maintain morphemes qua morphemes. 36. This atom-oriented lexicon stands in contrast to Lieber (1981) and its decidedly more organised contents (complete with stems as well as roots). The agenda there was to move all morphology into the lexicon, and although the tree structures of Word Syntax may be taken more benignly as generalisations about lexical structures, those practitioners taking their cue from Fabb or the functional head movement (no pun intended) are taking a more literally syntactic view. 37. Somewhat ironically, the strongest form of Word Syntax implies that lexicalisation is not real, since only single morphemes are inserted at terminal nodes, in keeping with proposals dating back at least to the Sound Pattern of English (Chomsky and Halle 1968).
Chapter 3
Time for a test drive: putting descriptive frameworks through their paces
3.0 P U R P O S E O F T H I S C H A P T E R In Chapter 2, a summary characterisation was offered for each of the fifteen theoretical frameworks. The numerous cross-references in the running text were intended to flag points of convergence, divergence, and subtle distinction that the respective literatures could serve (or fail) to bring out. When one is working within a specific theoretical framework, it can undoubtedly be distracting – to author and reader alike – to balance one’s exposition with giving fair play to ‘the competition’, on the one hand, and with scrupulous bibliographical credit and annotation, on the other. In addition to the cross-referencing technique, the present chapter places on the table for discussion three distinct morphological phenomena: Scottish Gaelic noun inflection, Georgian verb agreement marking, and Sanskrit gerund formation. A basic analysis for said phenomena is outlined in the idiom of each of the frameworks. Subsection designations correlate with those in Chapter 2, and along the way a number of sources beyond the respective recommended reading lists are introduced. Each of the languages addressed in this chapter possesses many more topics of interest for morphology, but in order to create multiple opportunities to ‘touch the elephant’ of morphology in a relatively brief space, these diverse but circumscribed phenomena may demonstrate where a given framework’s metatheoretical choices serve to direct its approach to the ‘same’ data. 3.1 S C O T T I S H G A E L I C N O U N S : I N I T I A L C O N S O N A N T MUTATION The initial consonant mutations found in Scottish Gaelic nouns, verbs, and adjectives are related diachronically to phonological rules for which the phonetic motivation was subsequently lost. The literature concerning the initial mutations (and the corresponding alternations across all the Celtic languages: Irish, Welsh, Breton, Manx, and Cornish) is substantial, and accounts are grounded in widely varying phonological, morphological, and syntactic theoretical perspectives (see references in Stewart 2004). The following examples demonstrate the distribution
Time for a test drive 91
of the initial consonant mutation traditionally called ‘lenition’ in the inflected forms of four common Gaelic nouns (data from Stewart 2004, 2013; MacLaren 1935). Table 3.1 Example paradigms for Scottish Gaelic nouns. (Source: Stewart 2004, 2013; MacLaren 1935) I.
doras (m.) ‘door’ Singular
Plural
Nom.
doras /t ɾas/
Gen.
dorais /t ɾiˉʃ/
dorais /t ɾiˉʃ/ dhoras /ɣ ɾas/
Dat.
doras /t ɾas/
Voc.
a dhorais! / ɣ ɾiˉʃ/
dorais /t ɾiˉʃ/ a dhorais! / ɣ ɾiˉʃ/
III.
sgoil (f.) ‘school’ Singular
Plural
Nom.
sgoil /sk /
sgoiltean /sk tʃan̪ /
Gen.
sgoile /sk /
Dat. Voc.
II.
balach (m.) ‘boy, lad’ Singular
Plural
Nom.
balach /pal~ax/
balaich /pal~eç/
Gen.
balaich /pal~eç/
bhalach /val~ax/
Dat.
balach /pal~ax/
balaich /pal~eç/
Voc.
a bhalaich! / val~eç/
a bhalaich! / val~eç/
IV.
clach (f.) ‘stone’ Singular
Plural
Nom.
clach /kh~l ax/
clachan /kh~l axan̪ /
sgoiltean /sk tʃan̪ /
Gen.
cloiche /kh~l jç /
chlach /xl~ax/
sgoil /sk /
sgoiltean /sk tʃan̪ /
Dat.
cloich /kh~l jç/
clachan /kh~l axan̪ /
a sgoil! / sk /
a sgoiltean! / sk tʃan̪ /
Voc.
a chlach! / xl~ax/
a chlachan! / xl~axan̪ /
The basic alternation patterns of initial lenition in Scottish Gaelic (Stewart 2013) are shown in Table 3.2. For each orthographic consonant in the first column, there exist contrasting ‘broad’ (velarised, or at least non-palatal(ised)) and ‘slender’ (palatal(ised)) phonemic correspondents, presented in said order within the cells of the second column. The third column provides the orthographic representation of the corresponding lenited consonant phonemes, and the fourth column gives the phonemic alternants that match with the radical (unmutated) phonemes in the same row.1 3.1.1 A-Morphous Morphology Inflectional rules take as input pairs {S, M} consisting of a phonologically specified stem and a (contextually appropriate) morphosyntactic representation (MSR) (Anderson 1992: 122). Since the lenition mutation is regular and productive in terms of its mapping between alternant pairs, it is preferable to capture the mutation
92 Contemporary Morphological Theories Table 3.2 Initial lenition alternant sets in modern Scottish Gaelic. (Source: Stewart 2013) Radical Grapheme
Phoneme(s) j
b c d f g l m n p r s
/p, p / /kh, ch/ /t̪ , ʤ/ /f, fj/ /k, c/ /l~̪ , / /m, mj/ /n̪ ˠ, ɲ/ /ph, phj/ /rˠ, ɾ̥/ /s, ʃ/
t
/t̪ h, ʧ/
Lenited Grapheme(s) a
bh ch dh fh gh l mh n ph r sh sd th
Phoneme(s) /v, vj/ /x, ç/ /ɣ, j/ Ø /ɣ, j/ /l~̪ , l/b /ṽ, ṽj/ /n/c /f, fj/ /ɾ, ɾ̥/ /h, ɕ/ /s, ʃ/ /h, ɕ /
a Lenition is conventionally indicated in writing by a following for all obstruents. b Broad and slender are not subject to alternation in lenition contexts. c Both broad and slender phonemes alternate with a plain alveolar nasal under lenition. d Before obstruents: and, surprisingly, (for which see Rogers 1980).
generalisations at the level of inflectional word-formation rules. For at least the doras–balach class (henceforth class Na), i-Ablaut will similarly be (part of) a WFR. Stem sets are as follows: Class Na Doras: {/tɔras/} Balach: {/paɫax/} Class Nb Sgoil: {/skɔʎ/} Class Ng Clach: {/kɫojç/ [gen./dat., sg.]; /kɫax/} (lexically specific alternation) Since none of the forms in the set has multiple specifications for the same feature(s), there is no call for layering in the MSRs. WFRs (all are +N) are as follows: 1. [{+nom., +dat.}, +sg.] /X/ → /X/ 2. [+gen., +sg.] /YVC/ → /YV [+high] C/ (Na) /X/ → /Xə/ Rule (1) states that the bare stem will be used in the nominative and dative singular. In the case of clach, the lexically specified [+dat.] stem will be selected, owing to its greater specificity, and will be used as is for the dative. In (2), disjunctivity is to be invoked twice:
Time for a test drive 93
a. the more specific clause will apply to Na nouns only, and the second clause will apply elsewhere, and b. the lexically specified [+gen.] stem will be selected for clach. (3) [+gen., +pl.] /CY/ → /C’Y/ (4) [+pl.] /YVC/ → /YV [+high] C/ (Na) /X/ → /Xtʃən/ (Nb) /X/ → /Xən/ (Ng) The Elsewhere Condition is in play here as well, since rule (3) will precede and preempt rule (4). Within rule (4), where it does apply, the different clauses are indexed to the lexical class of the input stem, and thus apply disjunctively. (5) [+voc.] /CYVC/ → /əC’YV [+high] C/ /X/ → /əX/
(Na)
In (5) the first clause precedes and preempts the second clause. C’ is used here to indicate the mutated alternant of the corresponding C in the input stem. 3.1.2 Autolexical Syntax/Automodular Grammar Mutation and i-Ablaut are consigned to the principles of Prosodic Phonology (see §2.13; McCarthy 1981; Marantz 1982), as was proposed in Sadock (1991: 26). The remaining few ‘lexemes’ have the following lexical representations: Table 3.3 Lexical entries for inflectional affixes. -e Syntax Semantics Morphology
-tean
nil nil nil nil N[fem]\N[gen.,sg.] N[Nb]\N[pl.]
-an
a-
nil nil nil nil N[Ng]\N[pl.,{nom.,dat.}] N[voc.,sg.]/N
The Morphology describes appropriate insertion contexts, using Categorial Grammar formalism (see §2.3). In Autolexical Syntax, stems are considered to be the head of inflected words. Inflections (Y) are introduced by the following general rule (X = N, for the present data set), and then placed with respect to the stem (X[−0]) depending on whether they are prefixes or suffixes: X[−1]
→
X[−0], Y
The Case and Number properties would be assigned based on context, whereas declension class would be a lexical property of the noun. All four example lexemes
94 Contemporary Morphological Theories
are simple nouns (N[0]), and therefore semantically intransitive predicates (F[−1]) (Sadock 1991: 31). Table 3.4 Lexical entries for noun stems.
Syntax Semantics Morphology
doras
balach
sgoil
clach
N[0] F[−1] ‘door’ N[−0]
N[0] F[−1] ‘boy’ N[−0]
N[0] F[−1] ‘school’ N[−0]
N[0] F[−1] ‘stone’ N[−0]
The combination of the affixes and the stems give N[−1], i.e. inflected words in the morphology, once all appropriate inflections are introduced. These are N[0] elements in the syntax, and examples of such inflected words would be the following: Table 3.5 Lexical entries for noun words.
Syntax Morphology
doras
bhalach
sgoile
cloich
N[1, nom., sg.] N[−1]
N[1, gen., pl.] N[−1]
N[1, gen., sg.] N[−1]
N[1, dat., sg.] N[−1]
Semantics are assumed to be unchanged under inflection. 3.1.3 Categorial Morphology Whereas affixation is accounted for in Categorial Morphology by addition operations, non-concatenative morphology is effected by means of substitution operations (Hoeksema and Janda 1988). First the two-place operations, definable in terms of lexical entry triples on the morpholexically context-sensitive affixes: -tean -an -e
Where x H {b, g}
The Vocative prefix applies in all classes, and so does not require the subcategory specification in its input requirements: a
These affixes will be added via a cancellation operation – left-cancellation for the suffixes, right-cancellation for the prefix. Initial lenition would have a lexical entry , and its effect, i.e. the operation fmut, should be treated with a rule of replacement.2
Time for a test drive 95
fmut (C[–strident, –continuant, a spread glottis]X) = C[+continuant, −a voice]X The i-Ablaut would parallel mutation to some degree, with an entry , where application is limited to Na (the class of doras and balach), and the operation defined as follows: fablaut (XVC) = XV[+high]C The alternation a ~ oi in clach seems to be separate from this,3 and so should probably be handled in the lexicon, rather than with a rule that would imply more general applicability. 3.1.4 Construction Morphology Much of the focus to date in this framework has been on derivational morphology and compounding, but the choice is by no means a necessary one. The explicit consideration of inflection in Booij (2010: 22–3), although brief, outlines the need to consider inflected forms as wholes to which sets of morphosyntactic and morphosemantic properties pertain, that is, constructions with holistic properties, rather than as assemblies of independent meaningful units. Likewise, form-based generalisations are readily handled in the framework, so that potentially morphomic entities such as syncretism or the use of the form of a word as a stem for inflection or base for derivation are describable without entailing the carryover of other aspects (e.g. semantics, gender) of the word-form so used. In the case of Scottish Gaelic noun inflection, the schemas that relate distinct Case and Number combinations will effect little change overall, since the semantic element is unchanged in inflection. As for the initial mutations, Booij proposes the following paradigmatically related schemas to show the link between English sing and sang: [XiY]V, [–past] ≈
[XaY]V, [+past]
The method is rather blunt, in that it declares the correspondence at the full segment level, but since the pattern is neither productive nor part of the synchronic phonology of English, this type of representation is defensible. Although the prospect of setting up the analogous schema pairs for every Radical ≈ Lenited correspondence might give the impression of one’s having missed a sound structural generalisation, there is apparently no phonological rule that can be abstracted from all the alternant pairs (Green 2007). This strategy for representing non-concatenative relationships within the framework would imply the following example schema pairs: [pʰY][Radical] ≈ [fY][Lenited] [mY][Radical] ≈ [ṽY][Lenited] [sY][Radical] ≈ [hY][Lenited] Each of these lenition schema-pairings would depend hierarchically from a higher organising node, say, [lenition], to capture their parallel status and function. In order to ensure that -initial words do not lenite as /h/ when followed by an obstruent (e.g. sgoil), a more specific schema-pair, something like:
96 Contemporary Morphological Theories
[skY][Radical] ≈ [skY][Lenited] could appear hierarchically below the [s] ≈ [h] correspondence, and thus would override the more general one above when appropriate. In light of the fact that, distinct from the Germanic ablaut residue in English, the initial mutations in Scottish Gaelic are both productive and multifunctional, the alternations must be made available as a form-only level for morphological construction schemas to draw upon, that is, as a morphomic entity, and the chain of instantiation would be a lenited form instantiating a lenited stem, which in turn is instantiating a lenited initial schema. The default constructional schema corresponding to Noun would be the basis for other more elaborated schemas. The prevalence of initial mutation in the language would make it reasonable to distinguish the initial segment in line with the mutation schemas above (X = initial segment, Y = rest of form), and acknowledge inflectional class (as i) in light of mutation’s distribution: [XY]Ni
↔
[Case { }, Num { }]
↔
[nom. sg.]
The apparent default status of [nom. sg.] would require a minimally specified subschema: [XY]Ni
In order to indicate the lenited requirement on the initial segment for [gen. pl.] forms, for example, an association in the subschema with the aforementioned [lenited] system can be indicated by means of an association with the initial segment slot in the construction. [XY]Ni
[Gen. pl.]
[lenited] Figure 3.1 Genitive plural constructional schema.
In the case of paradigm IV, the shape of the [voc. sg.] could be said to be based on the shape of the [gen. pl.], but with the added prefix [ə]. Since the relationship between the two subschemas, however, consists in the initial lenition only, and because the formal generalisation does not extend to paradigms I–III, acknowledging the formal similarity is likely better left implicit, in that both would instantiate a (morphomic) lenited stem (Figure 3.2), rather than setting up a multi-layered structure suggesting that the [gen. pl.] is transformed into the [voc. sg.], an incremental, rather than realisational analysis. The facts concerning final palatalisation of the stem are split between categorical (phonologically distinct for lexemes with broad final-syllable vowels, vacuous for those with slender vowels) in paradigms I and II, versus an idiosyncratic and
Time for a test drive 97
[əXY]N4
[Voc. sg.]
[lenited] Figure 3.2 Vocative singular constructional schema for paradigm IV.
lexeme-specific residue, e.g. the case of clach (paradigm IV). The former type of palatalisation should be set up as a correspondence hierarchy [palatalisation], analogous to that suggested for [lenition] above, while the latter type will be much more specific, and hardly schematic at all, effectively a lexical stipulation. The advantage in the Construction Morphology approach here is that the productive nature of initial lenition (despite being non-concatenative) is captured in its schematisation, but the truly irregular pattern seen in cloiche ‘rock [gen. sg.]’ is not forced to be factored out in the same way, and thus misrepresented. 3.1.5 Distributed Morphology For the derivation of each inflected word-form, the terminal nodes of S-Structure are taken into Morphological Structure, where the morphosyntactic feature nodes associated with the 0-level node for the root in question are assessed, in order to identify any needed morphological operations, so as to ensure the appropriate final configuration for Vocabulary insertion and PF interpretation.4 The analysis entails the following set of listed affixes among the Vocabulary, including here associated diacritic MP rules to trigger mutations as readjustment rules: Affix MP rules Ø-X [+leniting] ə-X [+leniting] ə-X X-ə [+i-Ablaut] X-Ø [+i-Ablaut] X-Ø [+i-Ablaut] X-ən X-tʃən
Meaning [+gen., +pl.] [+voc.] [+voc.] [+gen., +sg.] [+gen., +sg.] [+pl.] [+pl.] [+pl.]
Restrictions Where X = doras, balach, clach . . . Where X = doras, balach, clach . . . Where X = sgoil . . . Where X = clach, sgoil . . . Where X = doras, balach . . . Where X = doras, balach . . . Where X = clach . . . Where X = sgoil . . .
The patterning in the plural forms can be mediated by the Elsewhere Condition, such that, for example, a Vocabulary item specified for [+gen. +pl.] (e.g. the first in the list above) will preempt a Vocabulary item specified only for [+pl.]. DM’s assumptions require that all features that syntax standardly assigns to a word must be part of the structural representation at some point, so despite the fact that forms such as [+nom. +sg.] correlate with no overt affixation, and no MP readjustment diacritic need be invoked in their realisation, the c orresponding
98 Contemporary Morphological Theories
f eatures must nevertheless be represented and dealt with in the derivation (Halle and Marantz 1993: 137). One might continue coining as many phonetically null Case markers as necessary, or one might establish impoverishment(-style) rules deleting the unmarked values ([+nom.] with respect to Case and/or [+sg.] with respect to Number). For doras and balach, [+nom. +pl.] is distinguished formally from [+nom. +sg.] through the insertion of a phonetically null trigger of the MP readjustment of i-Ablaut (final palatalisation). In the case of sgoil, the under-specified [+pl.] exponent -tean applies in the derivation of any Case form, and thus it also serves to contrast the otherwise unmarked [+nom.] forms from each other. This analysis assumes the insertion of formally unified root Vocabulary items in the respective derivations. Readjustment rules are ‘morphologically-conditioned phonological rules . . . that have the form of phonological rules and apply to morphemes after Vocabulary Insertion’ (Halle and Marantz 1993: 124, 128), and it would be these which effect the mutations in specified contexts. The readjustment rule analysis consigns mutations to an analogous place in the Scottish Gaelic grammar as that in which English grammar houses the putative synchronic rule that targets the verb do (and only that verb), readjusting the /u/ to /ɪ/ in the context [+past, -participle] (and only in that context), while the Vocabulary item /-d/ does the entire work of exponence (Halle and Marantz 1993: 128). From a DM perspective, then, the correlation between particular sets of morphosyntactic features and the initial mutations cannot be one of exponence at all, but rather it is a stipulated by-product of the derivation.5 The assumption within DM that morphological structure is generated by the syntax entails that the ‘exponent proper’ be affixal in nature, even if it is phonetically null. This privileging of pieces (IA) over processes (IP) is a choice that leaves the regularity and productivity of the initial mutations without even the potential of assuming a direct morphological function. 3.1.6 Lexeme–Morpheme Base Morphology The analysis here needs to consider only I[nflectional]-derivation and Morphological Spelling, i.e. the realisation of the inflectional categories Case, Number, and the inherent category of inflectional class (which may or may not correspond one to one with Gender). The grammatical functions for which the various inflected forms may be used seem not to be critical here. The Separation Hypothesis permits the treatment of the relationship between inflectional categories and their exponents as a mapping. The evidence given supports treating I and II as instances of the same lexeme-class (call it Na), and III and IV should provisionally be assigned to distinct further classes unto themselves (Nb and Ng, respectively). Let us assume that the initial mutations are formally parallel,6 despite some divergence in phonetic detail. All operations on the stem, whether affixations or alternations, are to be considered elements of Morphological Spelling. The lexeme contributes its phonetic representation as an input to MS, and depending on inflectional class, Case, and Number, different MS operations are selected.
Time for a test drive 99 Table 3.6 Morphological Spelling rule application patterns. Class
Case
Number
Na
Nom.
Sg. Pl. Sg. Pl. Sg. Pl. Sg. Sg. Pl. Sg. Pl. Sg. Pl. Sg. Sg. Pl. Sg. Pl. Sg. Pl. Sg.
Gen. Dat. Nb
Voc. Nom. Gen. Dat.
Ng
Voc. Nom. Gen. Dat. Voc.
Mutation
@ @
i-Ablaut
-
- n
-tʃ n
@ @ @ @ @
@
@ @
@
@
@
@
@
Two dimensions are unable to capture the complex mapping fully, but the matrix above does make clear the usefulness of a separation between inflectional categories and their exponents, in combination with lexical declension class. 3.1.7 Lexical Morphology and Phonology/Stratal Optimal Theory Mutation and i-Ablaut in LM&P are level-one phenomena, despite their regularity and productivity, by virtue of the locus of their effects, i.e. the stem. It is difficult to say whether the Vocative prefix triggers initial mutation or not, since the mutation is motivated independently for the Genitive Plural. The plural suffixes do not interact (synchronically) with i-Ablaut, and they do not pile up in the data here, so a precise level assignment for the suffixes is not possible. It would seem only that they must apply after the level 1 rules for mutation and Ablaut. The apparent Blocking phenomenon has some interesting implications here, especially in the sgoil class, since neither initial mutation nor i-Ablaut is evident. It cannot reliably be determined whether the Plural suffix and the Genitive Singular suffix are applied in addition to level 1 inflection, or whether they apply as a back-up to the non-application of the level 1 inflection. The identical Genitive Singular suffix is used in addition to a stem alternation in the case of clach, i.e. cloiche, and so this appears simply to be the Genitive Singular suffix used with Feminine nouns. If, on the other hand, the Plural marker in sgoiltean is a backup to initial mutation, the prediction would be that plural nouns which begin with /sp/, /st/, or /sk/ should mark the plural categorically with some
100 Contemporary Morphological Theories
affix, -tean or otherwise. To verify this prediction, it is necessary to go beyond the given data, and even then the facts are unclear. If the suffix is motivated by blocking, however, one is hard-pressed to explain the suffix used in addition to mutation, in attested cases like ghillean ‘of young men’ (cf. gille ‘young man’). The Elsewhere Condition and the related Blocking effect would not predict this multiple marking. 3.1.8 Minimalist Morphology To account for the distributions here, non-monotonic stem hierarchies and a list of affixes are appropriate. The noun sgoil uses a single stem throughout its paradigm, working only with affixes.7
Figure 3.3 Stem hierarchy for doras ‘door’.
[ paɫax ]+N, +Masc
[ V …. ]+Gen, +Pl
[ … eç ]+Gen +Pl
[ əV …. ]+Voc Figure 3.4 Stem hierarchy for balach ‘boy’.
[ kʰɫax ]+N, +Fem
[ x … ](+Gen, +Pl) ˅ +Voc Figure 3.5 Stem hierarchy for clach ‘stone’.
[ .. ɔjç ]+Gen
+Dat
Time for a test drive 101
The affixes are as follows: ə- +voc. -ə +gen. -an +pl. -ʧan +pl.
/ +fem. / −gen., +fem.
Although affixes in MM must be positively (+) specified, both stems and input conditions for affixes may have (−) specifications (Fabri et al. 1996: 236–7; Wunderlich and Fabri 1995: 264, 280). MM’s hard line against inflection classes that do not follow from substantive or otherwise independently motivated features such as Gender makes it difficult to restrict a general affix such as [-ʧan] to plural forms of sgoil, but not apply to the (equally [+fem.]) plural forms of clach. 3.1.9 Natural Morphology In every case, there is no word-form less ‘markered’ (merkiert, or ‘featured’) than the Nominative Singular, which represents the unmarked value for both Case and Number. In I, II, and III, the Dative is syncretic with the Nominative, and syncretism (many-to-one mapping) is considered to be semiotically suboptimal in NM. If one were obliged to choose a form that was next in line in markedness to the Nominative, however, it would have to be the Dative, so the syncretism could in fact be worse than it is. For the masculine nouns (I and II), the Plural is more markered than the corresponding singular, and that is in line with constructional iconicity. Also in I and II, the Genitive is more markered than the nom./dat., but less markered than the Vocative, which is surely the most marked case of all in these paradigms. In III, Case is neutralised within Number, excepting the gen. sg. This is unusual in comparison to the other three examples, but chances are that sgoil in particular may have been influenced by the cognate word in English. The presence of some morphological idiosyncrasy is perhaps less worrisome on that assumption.8 That the gen. pl. is not distinguished formally from the other plural forms is particularly unusual, however, given the other three examples. As for IV, the syncretism between nom. and dat. sg. is lost, which is good from a biuniqueness standpoint, but there is a new (near-)syncretism of the gen. pl. with the voc. sg. (very unusual on the markedness/iconicity dimension). The gen. sg. in IV is just like the dat. sg., but with a final /ə/, and the same is true in III, although less strikingly than the facts in IV. Although there are some affixes in use here, these paradigms rely to a remarkable degree on so-called modulator featured symbolising, the least optimal symbol type, short of no marker at all. The fact that there is at least one syncretic pair in each number column of each paradigm here would suggest that the case system is under pressure to collapse or to attract a new marker morpheme in one Case (more likely the dat. for I and II, on markedness grounds). It seems the nom./dat. distinction is being kept alive by patterns like IV. If IV is a(n unproductive) minority pattern in the language,
102 Contemporary Morphological Theories
the pressure to regularise forms like cloich to clach is quite high, and the Natural Morphology prediction would be that dat. will collapse in time, all else being equal. 3.1.10 The network model Since these are nouns rather than verbs, the Relevance Hierarchy is not directly, shall we say, relevant here. The first thing to do, then, is to draw networks for the related word-forms and to see how they compare: I.
t̪ ɔ ɾ ə s
In 3-D, we would also see:
t̪ ɔ ɾ ɨ ʃ ɣɔɾəs
t̪ ɔ ɾ ɨ ʃ
ɣɔ ɾ ə s ɣɔɾɨʃ Figure 3.6 Network for doras ‘door’. II.
paɫax
In 3-D, we would also see:
pa ɫ eç p aɫ e ç
vaɫ ax v a ɫ ax vaɫeç
Figure 3.7 Network for balach ‘boy’.
III.
s kɔʎə
s kɔʎ
s k ɔ ʎ tʃ ə n Figure 3.8 Network for sgoil ‘school’.
Time for a test drive 103
IV.
kʰ ɫ a x
In 3-D, we would also see:
xɫax kʰ ɫ ɔj ç
xɫax kʰ ɫ ɔj ç ə
kʰ ɫ ɔj ç ə
kʰ ɫ a x ə n
Figure 3.9 Network for clach ‘stone’.
I and II show nicely closed networks, indicating that there is an element of regularity in these related forms. The fact that the corresponding forms also fill parallel grammatical functions is a sure sign of a paradigmatic pattern. The stronger versus weaker links are even in the same positions with respect to the phonemic sequences. This pattern is predicted to be stable and should be relatively productive. The alternation of the initial consonant in I (changing place, manner, and voicing) is more distant in phonetic terms than in II (manner and voicing), which is still greater than in IV (manner only). The less the phonetic distance between alternants, the more recoverable the correspondence, and the easier lexical access will be. Class I, therefore, stands out in language-independent terms, although if the alternation is productive, that fact may support the pattern’s continued existence. III shows a simpler pattern of identity of the stem across the board with suffixal inflection. The regularity here makes the morphological boundary and constituency even more readily detectable than the patterns observed in I and II, but the one-to-many form–meaning mappings undercut the value of the stem’s consistency. As sets of related forms go, the pattern in IV is quite remarkable. There should be a lot of pressure on this paradigm to regularise at least the vowel quality of the stem. The lexical strength of the word for ‘stone’, however, might be quite high for Gaelic speakers, given the frequency of occurrence of physical stones in the relevant parts of the world. Such an extralinguistic circumstance might provide some support for this irregular pattern’s survival. It is also important to find out just how productive the vowel quality alternation is in Gaelic nouns more generally, since that may affect the degree to which the alternation may be considered an irregularity. In this limited data set, IV stands out. It would be premature to assume that this sample was representative of the language as a whole or that type and token frequencies can be reliably projected without more evidence.
104 Contemporary Morphological Theories
3.1.11 Network Morphology The situation here is remarkably similar to the Russian example discussed in the earlier presentation of Network Morphology (§2.11). The four paradigms under discussion here may be seen as belonging to two or three declension classes. NOUN N_F N_α
N_β
N_γ
Doras, Balach
Sgoil
Clach
Figure 3.10 Inflectional class hierarchy for Scottish Gaelic nouns.
Since the phonology of the mutated and/or ablauted stem clearly depends on the phonology of the root, lexical items will be assumed to have up to four formally distinct yet relatable stems for use in the statement of particular morphological facts. There is more redundancy in the stem set at the phonological level, but this follows from a limitation in the formalism. The formal correspondences among stems might be formalised via an analogue of the First/Last/Rest convention (as used in Hoeksema and Janda (1988; §2.3) and as used for argument structure in Evans and Gazdar (1996)). Brown (1998: 216–17) offers a tentative hierarchical representation of morphophonological selection, but the system is not readily transferable to this case. This will not be pursued here. Nodes required for the basic inheritance hierarchy are as follows: Doras: < > == N_a
Balach: < > == N_a
== tɔrəs
== palax
== gɔrəs == tɔrɨʃ
== valax == paleç
== gɔrɨʃ.
== valeç.
Sgoil: < > == N_b
Clach: < > == N_g
== skɔl
== klax
== “” == xlax == klɔjç. NOUN: == “” == “” == “” == “”. N_a: < > == NOUN % masculine nouns
Time for a test drive 105 N_F: N_b: N_g:
== “” == “a_” == “”. < > == NOUN % feminine nouns == “_e” == “a_”. < > == N_F == “_tean”. < > == N_F == “” == “_an”.
Inheritance principles together with the default/override relation and rules of referral will map the above lexical entries into the paradigms in question. In this way, generalisations between and across declensions are captured, and the fact that Nb is more similar to Ng than either of them is to Na is captured without making sameness or difference a simple binary choice. Carstairs’s (1987) notion of a macro- paradigm might therefore cover the relationship between Nb and Ng. 3.1.12 Paradigm Function Morphology The given data show eight paradigm cells for Gaelic nouns. We are dealing, therefore, with two morphosyntactic features, {CASE} and {NUM}. The former is an n-ary feature with four permissible values: nom., gen., dat., and voc. The latter, with only two permissible values, is effectively binary. There are no co-occurrence restrictions to mention here, and thus all eight cells of the paradigm are defined (exhaustivity; see Spencer 2003). Regular and productive stem-internal alternations are to be described in PFM as stem formation-rules, and since the formally most differentiated paradigms, doras and balach (class Na), show four distinct but related stems, four stems are posited for the class N in general. Since initial mutation is unified from a conditioning perspective but not from a form perspective, it is misleading to render mutation as a quasi-phonological rule. The stated alternations as given in Table 3.2 are adequate for the present purpose. I-Ablaut can be simply formulated as a feature-changing rule, but even this must be clearly recognised as a morphologically conditioned rule. The alternation patterns, therefore, are assumed to be static relationships between alternants, Basic-C and Mutant-C for initial consonants, and Basic-V and Ablaut-V for stem-final vowels. Stem-formation rules will be as follows: Where L is a masculine (=Na) noun with root C1YVnC, each of (a)–(d) implies the other three: (a) (b) (c) (d)
The Basic stem is identical to the root. The Mutant stem has Mutant-C for C1. The Ablaut stem has Ablaut-V for Vn. The Combo stem has Mutant-C for C1 and Ablaut-V for Vn.
106 Contemporary Morphological Theories
Where L is a feminine (=Nb or Ng) noun with root C1Y, each of (e)–(g) implies the other two: (e) The Basic stem is identical to the root. (f) The Mutant stem has Mutant-C for C1. (g) Refer other stems to Basic stem. Lexically specified stems such as cloich for clach override the application of more generally applicable stem formation and selection rules. Given the limited data set, there is distributional evidence for exactly three rule blocks: a stem selection block (Block 0), a suffixing block (Block 1), and a prefixing block (Block 2). A general paradigm function for Gaelic nouns can be posited as follows: Where s is a complete set of morphosyntactic properties for lexemes of category N, (i) PF () =def Nar2(Nar1(Nar0())) The rule blocks are the following: Block 0: (ii) RR0, {CASE:voc.}, [N] () =def , where Y is X’s Combo stem (iii) RR0, {CASE:gen., NUM:pl.}, [N] () =def , where Y is X’s Mutant stem (iv) RR0, {CASE:gen.}, [N] () =def , where Y is X’s Ablaut stem (v) RR0, {NUM:pl.}, [N] () =def , where Y is X’s Ablaut stem (vi) RR0, { }, [N] () =def , where Y is X’s Basic stem Block 1: (vii) RR1, {NUM:pl.}, [Nb] () =def (viii) RR1, {CASE:gen., NUM:pl.}, [Ng] () =def (ix) RR1, {CASE:gen., NUM:sg.}, [Nb] () =def (x) RR1, {NUM:pl.}, [Ng] () =def Block 2: (xi) RR2, {CASE:voc.}, [N] () = def According to PFM’s paradigmatic interpretation of the Pān·inian Principle, as represented in the formalisation of the paradigm function (PF) above, the narrowest applicable rule in each block will apply in defining the evaluation of the PF for any given pair (the Pān·inian Determinism Hypothesis). No rules in block 1 are applicable to lexemes of class Na; inflection in that class is accomplished without suffixation. The distinct stem formation rules for masculine versus feminine noun lexemes allow the rules of stem selection to be stated generally across the category N. The following proofs exemplify the preceding analysis: Where s = {CASE:nom., NUM:pl.}, PF () = Nar2(Nar1(Nar0()))
[by (i)]
Time for a test drive 107
= RR2,{ }, [N](RR1, {NUM:pl.}, [Ng](RR0, {NUM:pl.}, [N]())) [by Narn notation] = [by IFD, (x), and (v)]
Where s = {CASE:gen., NUM:sg.}, PF () [by (i)] = Nar2(Nar1(Nar0())) = RR2, { }, [N](RR1, {CASE:gen., NUM:sg.}, [Nb](RR0, {CASE:gen.}, [N] ())) [by Narn notation] = [by IFD, (ix), and (iv)] Where s = {CASE:voc.}, PF () = Nar2(Nar1(Nar0())) [by (i)] = RR2, {CASE:voc.}, [N](RR1, { }, [N](RR0, {CASE:voc.}, [N] ())) [by Narn notation] = [by (xi), IFD, and (ii)] Recall that the IFD serves, where no more specific rule is applicable within a rule block, to map the input to itself. Thus the block is evaluated, the form is definable, and no formal change to the input is effected, i.e. there are no zero-morphs involved in this analysis. Note that rule (viii) above is an identity function, but it is a separate stipulated override, not a default, partially realising the properties {CASE:gen., NUM:pl.} on lexemes of class Ng. Note also that (full or partial) syncretism in these paradigms is handled through the application of defaults, rather than through special rules of referral in the rule blocks.9 3.1.13 Prosodic Morphology Assuming that part of the lexical entry for any root is a segmental tier, the mutation effects can be represented as features which are associated to the initial C position in the skeleton, adding or altering features so as to convert the initial C to its mutated counterpart. The same morpheme does not condition a uniform phonological effect on the initial C of the stem, so the Structural Description and Structural Change must be somewhat complex (cf. Lieber 1983, 1992). The morpheme contains at least the feature specification [+continuant], which overrides the lexical specification for the C1 slot (vacuously where the stem is continuant-initial). Since the stop contrast is one of aspiration rather than voicing,
108 Contemporary Morphological Theories
but the fricative contrast is one of voicing, [–a voice] could be a part of the morpheme, sensitive to the setting of [spread glottis] in the root. The mutation never results in a change from [+voice] to [–voice], and therefore an analysis in which [Voice] is a privative feature is also possible. The i-Ablaut seems to be assigned right to left, given its effect on the last consonant (Cn), so morphemes triggering i-Ablaut can be formalised so as to attach to the last vowel of the stem (Vn), rather than to a particular vowel numbered left to right. Such morphemes will consist of a feature [+high], which will override the lexical specification for Vn’s height. This could also be done as a spreading of palatality from Cn of the root, but palatalisation of C next to front vowels is general enough in Scottish Gaelic that it need not be handled in the morphology, separate from phonology. Since mutation is here taken to be a matter of changing specifications in roots, rather than filling empty slots in the C–V skeleton, the Prosodic Morphology analysis of mutation is different from Arabic interdigitation, or from the spreading and prespecification in reduplication. This is a more powerful sort of operation than Prosodic Morphology was originally designed to handle, but the tools required are implied in the model. Using an OT approach here (Prince and Smolensky 2004), however, three constraints could be posited: ALIGN (Mutation-L, Stem-L) A mutated segment must be at the left edge of a stem. *hC[+stop] The sequence /h/ followed by a stop consonant is ill-formed. PARSE An element in the underlying representation must appear in the surface form. With the constraint ranking ALIGN (Mutation-L, Stem-L) >> *hC[+stop] >> PARSE, ALIGN keeps the mutation at the left edge of the stem. If the co-occurrence constraint were ranked higher than ALIGN, the mutation would be allowed to move in from the left edge just in case it would violate *hC[+stop]. Underlying /s/- stop clusters do not license mutation, and because this co- occurrence constraint outranks PARSE, it is better to leave mutation unparsed than to force the /s/ to mutate before a stop. Even though the mutated alternants are not phonetically parallel, and even though the conditioning for mutation in the data is completely morphological in the synchronic grammar (Stewart 2004; cf. Green 2007), this formulation within OT makes it seem as though it were driven primarily (if not purely) by segmental and prosodic phonology. Since i-Ablaut is more restricted in its application than initial mutation, the restriction to the doras–balach class might have to be a condition on the Parse constraint, i.e.: PARSE [dat., pl.]Class1
Time for a test drive 109
This mixes general morphological conditioning and particular lexical-class conditioning, but the OT formalism could accommodate it. As for the ‘normal affixation’ cases, Prosodic Morphology does not differ fundamentally from a concatenation account, except that the morphemes are represented as belonging to distinct morpheme-tiers. 3.1.14 Word-based Morphology Word-based morphology focuses on predictability among forms as a key analytical strategy (Blevins 2006: 537). In order to assess the implicational structure of the paradigms in question, the first level to inspect is the set of inflected forms as wholes with respect to their cells. Patterns of complete syncretism are as follows: I and II: III: IV:
[nom. sg.] [voc. sg.] [gen. sg.] [nom. sg.] [nom. pl.] [nom. pl.]
= = = = = =
[dat. sg.] [voc. pl.] [nom. pl.] [dat. sg.] [gen. pl.] [dat. pl.]
=
[dat. pl.]
=
[dat. pl.]
The formal correlation of [nom.] with [dat.] is clearly strong in Gaelic, since it would seem that it is only the distinction in paradigms like IV (and there only in the singular) that keep the distinction alive at the morphological level. The [voc.] syncretism across Number in paradigms I and II likewise contrasts with both III and IV. In comparing the full forms in paradigm III, one may abstract the respective [nom.] forms10 as stems in the [voc.] forms, and thus the prefixed exponent may also be abstracted. In paradigm IV, the relation is similar, but the formal match is between the [gen. pl.] and the [voc. sg.] for purposes of stem and exponent abstraction in the latter word-form. The exponent of [pl.] in the [voc. pl.] forms of both III and IV is the analogical (or categorical) extension of the [sg.]/[pl.] contrast that is established in the [nom.] forms. One may reliably use the [dat. sg.] form to abstract stem and exponent in the [gen. sg.] in paradigms III and IV. Despite the different distribution of lenition (and palatalisation) across the paradigms, the phonemic identity of the lenition alternant pairs (Table 3.2) constitutes a fully regular and productive pattern in the synchronic grammar of Scottish Gaelic. Therefore, in contrast with the treatment of Germanic ablaut in Blevins (2003: 756) in terms of series, that is, recognisable alternation sets ‘resident in a finite – in fact quite small – number of high-frequency forms’, the lenition patterns have apparently been ‘encapsulated in a separate system of rules, templates, or schemas’ (756). This status as a system beyond particular paradigms means that instead of stipulating occasions of lenition form by form or lexeme by lexeme, as if it were an irregularity (not rule- governed), and thereby missing some of the largest generalisations in the language, a Word-based analysis places the lenition pattern at the right level of generality, permitting the prediction of the morphomic ‘lenited stem’ (final palatalisation aside). In
110 Contemporary Morphological Theories
terms of predictiveness, the reliable direction for initial segments is only from radical alternant to lenited alternant (a function from the former set to the latter), owing to the few ambiguities such as and both alternating with the /ɣ, j/ pair, and and each alternating with /h/ in certain phonologically specified environments. The key dimensions of a Word-based analysis of the Gaelic paradigms would include paradigm patterns, leading (or kenn-)forms, and the encapsulated lenition schema to ensure that all needed forms are predictable via form-based analogies (Blevins 2005: 7). Table 3.7 Paradigm patterns for Scottish Gaelic nouns.
[Nom. sg.] [Gen. sg.] [Dat. sg.] [Voc. sg.] [Nom. pl.] [Gen. pl.] [Dat. pl.] [Voc. pl.]
I and II
III
IV
R RP R a LP RP L RP a LP
R Re R aR R tean R tean R tean a R tean
R RP e RP aL R an L R an a L an
R = Radical (word-form built on unmutated root); L = Lenited initial segment (see Table 3.2 for correspondences); P = Final palatalization
With regard to leading forms, paradigms I and II can be predicted from the [nom. sg.] and one other cell: either of the phonologically identical [gen. sg.] / [nom. pl.] forms. To account for paradigms III and IV, one needs to know the [nom. sg.], [gen. sg.], and [nom. pl.] forms, since the [dat. sg.] can be predicted from the [gen. sg.], less the exponent. On the one hand, an analogy that deletes is more tractable than one that adds, especially when no other form shows the -e exponent, but on the other hand, as mentioned above, comparison with the [dat. sg.] form permits the abstractive analysis of the [gen. sg.]. Prediction and abstraction do not necessarily run in parallel, and the assumption of morphologically simple leading forms is often not the most helpful choice for the purposes of inter-form prediction. Given the availability of the lenition alternant system, the lenited initial is determined for any word-form for which the paradigmatic cell is specified as ‘L’. In the case of final palatisation, the example of cloich for clach shows that the precise shape of the final vowel-consonant sequence may need to be lexically stipulated, at least in some cases (cf. balaich for balach). In other words, since it is only the palatalisation pattern’s distribution within paradigms that is predictable, not its phonemic content, final palatalisation in Scottish Gaelic may qualify for treatment as one or more series, not unlike the Germanic ablaut series referenced above. 3.1.15 Word Syntax Lieber (e.g. 1992: 165–71) has dealt most directly with mutation and Umlaut in the Word Syntactic framework. Lieber’s examples of mutation involve a complex
Time for a test drive 111
affixation whereby an overt affix (a ‘mutation trigger’) attaches to the stem at one point and an empty timing slot is attached adjacent to the segment to be mutated. On analogy with the Fula analysis in Lieber (1992: 167–9), the empty timing slot attaches autosegmentally to the stem’s initial segment, forming a geminate. The resulting initial geminate is assumed to meet the structural description of a phonological process of ‘lenition’ which produces the observed mutation effects. The fact that no overt affix correlates with the mutation in gen. pl. forms in Scottish Gaelic means simply that there is a zero-affix meaning [+gen., +pl.] which associates the empty timing slot in initial position. Perhaps both could be handled at once if one were to assume that the empty timing slot ‘is’ the [+gen., +pl.] affix, a prefix, although this move is an innovation here, not suggested in Lieber (1992) or elsewhere. If the voc. prefix a-similarly contributes an empty timing slot just after it, this could add some indirect support for the empty gen. pl. prefix. In classes Na and Ng (but not Nb), a null [+gen., +pl.] affix could explain the failure of additional [+pl.] marking, since that would be featurally redundant. If one assumes further that the null [+gen., +pl.] does not apply to Nb instead of applying with no perceptible effect on the initial, this could explain the application of the [+pl.] suffix in sgoiltean [+gen., +pl.].11 The analysis of Umlaut is similar to that of initial mutation, since Umlaut, strictly speaking, is triggered by a vowel in a following morpheme. Lieber (1992: 170) appeals to a floating feature ([–Back], for German), which is part of the lexical entry of triggering suffixes. Stems, on this analysis, are under-specified, with only marked values present underlyingly. The floating feature, once associated to the last vowel in the stem, preempts the later association of the unmarked value ([+Back], for German). To accommodate the productive Gaelic i-Ablaut facts (e.g. dorais and balaich), however, the triggering suffix must be null itself, but carry a floating [+High], since the central vowels /a/ and /ə/ both raise (but do not front) to /ɨ/. Other affixes in the data contribute inflectional features to the stems they attach to by means of the unexceptional application of affixation and percolation. 3.1.16 What is at stake here? The analyses of Scottish Gaelic nominal case formation have brought out several important challenges to conventional morphological practice: 1. There exist languages for which affixation is not the majority morphological operation. 2. Non-concatenative processes need not be branded automatically as irregular. 3. Adopting the affix as norm and as the sole means of marking in a theory is an option for constructing a simpler theory, but it can lead to a loss of empirical coverage. For further consideration, and before imagining that the initial mutation facts are a relatively dismissible quirk that a comprehensive morphological theory need not take account of, there are several points to consider beyond the four sample paradigms in the present data:
112 Contemporary Morphological Theories
1. Initial mutation phenomena are a familial feature across all the Celtic languages (see, for example, Hamp 1951; Pyatt 1997; Green 2007). 2. Within Scottish Gaelic, initial mutation patterns including lenition are found not only in noun inflection but also in verb and adjective inflection (Stewart 2004). 3. Lenition alternations are also found (albeit word-internally) in both derivation and compounding in Scottish Gaelic, as well as at phrase-edges in certain syntactic constructions (Stewart 2004; cf. Borsley et al. 2007). Walking the line between ‘same’ and ‘different’ in the description of initial consonant mutation phenomena presents a considerable challenge to providing a simple and/or unified characterisation, even in a single language. A broader data set, beyond the scope of the present section, would show that just in Scottish Gaelic there seem to be several overlapping but not collapsible alternation patterns that have traditionally been treated together under the rubric ‘lenition’, despite distinct and identifiable grammatical conditioning for each (Stewart 2013; cf. Janda 1982). 3.2 GEORGIAN VERBS: AGREEMENT MARKER DISJUNCTIVITY Georgian verb agreement has provoked much discussion in both morphological and syntactic theory. Co-occurrence facts have resisted principled explanation in just those cases where multiple arguments are present and (apparently) compete for control of agreement marking. The individual markers used with underived transitive verbs in Georgian are the following: Table 3.8 The (overt) agreement affixes of Georgian. ‘Subject’ ‘Dir.Obj.’
1sg.
2sg.
3sg.
1pl.
2pl.
3pl.
v- m-
– g-
-s –
v-. . .-t gv-
–t g-. . .-t
-en –
In combination, however, the facts are as follows (Stewart 2001, adapted from Cherchi 1999: 42): Table 3.9 The present tense of xedav, ‘see’. Shaded cells would be reflexives, expressed periphrastically in Georgian. DO 1sg. Subj.
2sg.
3sg.
1sg. 2sg. 3sg. 1pl. 2pl. 3pl.
gxedav
vxedav xedav xedavs vxedavt xedavt xedaven
mxedav mxedavs
gxedavs gxedavt
mxedavt mxedaven gxedaven
1pl.
gvxedav gvxedavs
2pl.
3pl.
gxedavt
vxedav xedav xedavs vxedavt xedavt xedaven
gxedavt gxedavt
gvxedavt gvxedaven gxedaven
Time for a test drive 113
To see which forms really require explanation, it is helpful to consider a ‘what if’ paradigm based on the above, but ignoring the apparent cases of disjunctive application or insertion. All else being equal, and assuming somewhat arbitrarily that subject markers would appear outside of object markers, one would expect the following affixes to appear (Ø stands as a place-holder; predicted but non-appearing affixes are given in capital letters): Table 3.10 Hypothetical paradigm without disjunctivity for the present tense of xedav, ‘see’. Subj 1sg. 2sg. 3sg. 1pl. 2pl. 3pl.
DO 1sg.
2sg.
V-g-xedav† Ø-m-xedav m-xedav-s g-xedav-s V-g-xedav-t† Ø-m-xedav-t m-xedav-en g-xedav-en
3sg. v-xedav Ø-xedav xedav-s v-xedav-t Ø-xedav-t xedav-en
1pl.
2pl.
3pl.
V-g-xedav-t†
v-xedav-Ø Ø-xedav-Ø g-xedav-t-S† xedav-s-Ø V-g-xedav-t-T† v-xedav-t Ø-gv-xedav-t Ø-xedav-t-Ø gv-xedav-en g-xedav-en-T† xedav-en
Ø-gv-xedav gv-xedav-s
Thus there are six forms (marked † in the table) out of twenty-eight which are demonstrably not as expected. Every one of the six would otherwise have two consecutive overt prefixes or two consecutive overt suffixes. In the form *V-g-xedav- t-T, it would obviously be questionable which of two consecutive /t/ segments is deleted. Geminates are outlawed generally in Georgian, so the point is moot in this case. The disjunctivity otherwise, however, is not a matter of phonotactic violations (Anderson 1992: 87, fn. 13), but is rather entirely a matter of morphological distribution. 3.2.1 A-Morphous Morphology Anderson (1984, 1986, 1992) has written extensively about Georgian agreement, and he considers it strong support for the positing of disjunctive rule blocks. Rules which apply disjunctively by definition belong to the same rule block. The rule which actually gets to apply precedes the others in an ordered block, subject to the Elsewhere Condition. That is, an MSR that consists of a proper subset of the features contained in the MSR of the others is realised first and preempts the application of less specific rules. If this subset relation does not hold, however, then the appeal is to extrinsic ordering, a stipulated preferential application of that rule which is needed to match the surface facts. In the Georgian case, the 2nd person object prefix preempts the 1st-person subject marker where both are applicable, e.g. g-xedav, and not v-xedav, v-g-xedav, or g-v-xedav. There is no attempt to motivate a principled precedence relation of g- over v-; rather, the rule introducing g-applies to the exclusion of the rule introducing v-because it has to, in view of the data.
114 Contemporary Morphological Theories
Anderson (1984) also casts the -t suffixes as a unified non-3rd-person marker, and claims that the fact that a 1st-person plural object is realised by gv- but not -t is the result of disjunctive ordering and the prefix’s precedence in their common rule block. Thus blocks are tied to MSRs, and not to linear position classes (cf. PFM §2.12). Since a subject marker and an object marker are keyed to different MSR layers, according to A-Morphous Morphology, the disjunctivity cannot even be explained in terms of an MSR conflict. While A-Morphous Morphology’s tolerance of extrinsic ordering within rule blocks permits the desired result, by choosing this method the predictive value of the theory in this area is compromised. 3.2.2 Autolexical Syntax/Automodular Grammar As far as the Syntax module knows, so to speak, the word-forms given in Table 3.9 are fully specified verbs. Any problematic aspects are to be dealt with entirely within the morphological component, but Autolexical Syntax finds itself in a difficult position if it is to account for disjunctivity between apparently compatible inflectional affixes. Table 3.11 Autolexical classification of Georgian lexemes.
Syntax Semantics Morphology
v-
g-
-xedav-
nil nil V[−1]/V[−0]
nil nil V[−1]/V[−0]
V[0] F[−2] V[−0]
Since vxedav and gxedav are both V[−1] forms, i.e. acceptable as fully inflected words, the autolexical specification of the prefixes puts them into competition. This gains the disjunctive application, but it does not explain the precedence of g- insertion over v-insertion where both are equally motivated in the sentence. Although word order at the sentence level in Georgian is claimed to be free, the canonical order is S-O-V. Under the assumptions of Autolexical Syntax, since there is a default mapping between abstract syntactic structure and morphological structure, it is possible to derive the needed S-O-V prefix ordering ‘for free’ from the syntax, with no need to appeal to separate linear precedence rules specific to the morphology. Thus it is possible to come close to capturing both the paradigmatic and syntagmatic aspects of the Georgian agreement prefix disjunction without extraordinary manœuvres.12 3.2.3 Categorial Morphology If both v- and g-take ‘verb stems’ as input, and if a ‘verb stem’ as a formal unit crucially has no agreement markers already in place, then the competition between v- and g-is predicted.
Time for a test drive 115
v-
g-
The dominance of {OB:2} over {SU:1} is not explained, however, and therefore must be stipulated. The analogous analysis is available for the suffixes – each one . This assumes, somewhat arbitrarily, that agreement suffixes are applied ‘before’ agreement prefixes. The order could be reversed, it would seem, but with no net improvement in explaining the disjunction. 3.2.4 Construction Morphology The constructional approach to this phenomenon would tie the surface forms, including their morphotactic profiles, to the morphosyntactic feature sets that they realise, making matters more about which forms map to which property sets, and then devising the inheritance tree(s) to reflect the distribution. Treating each form as a construct, and keeping in mind that ‘each expression may simultaneously instantiate more than one construction’ (Gurevich 2006: 62), it is relevant to note that each of the basic (non-reflexive) forms of the transitive verb lexeme is responsible for agreeing with both of its arguments. Since the morphological template for these constructs apparently conforms to a root plus a maximum of one agreement affix in each direction,13 as a first approximation, Figure 3.11 presents a hierarchical representation of the (sub)schemas that the inventory of individual morphs used for agreement marking would seem to predict, all else being equal. The two possible formal combinations that are not used are those in which both markers would realise subject arguments, and so the combinations are excluded due to semantic incompatibility, in contrast with the g-/v-disjunctivity under consideration. Considered from another angle, a form realising a two-argument person-number combination may involve 0 ([b]), 1 ([a[b]] or [[b]c]), or 2 ([a[b]c]) agreement morphs, and whenever possible, a form will bear both one prefix and one suffix. In framing constructions as multi-level constraints on expressions, the morphological template imposes an output constraint on forms. From among the logically possible combinations of markers, that combination of markers best satisfies the output constraint which (1) leaves no agreement slot (a or c) unfilled if, among the set of arguments to be realised, one argument has an exponent that may fill that slot, (2) realises each argument overtly in at least one of the slots (if possible, given the available array of exponents), and (3) realises the more marked value (DO > SU; pl. > sg.) in cases of conflict within slots (cf. Stewart 2001). Consulting the hypothetical paradigm in Table 3.10, it is clear that of the twenty-eight non-reflexive forms, every observed inflectional construct instantiates one or another of the construction schemas in the hierarchy above, and in fact no semantically well-formed combination pattern goes unused. Table 3.12 recasts and annotates information from Tables 3.9 and 3.10, and Figure 3.11 in a manner that assesses the constructional source of each form in light of how well each grammatical form in the paradigm maximises the potential informational content of the
116 Contemporary Morphological Theories [b]
[[b]en]
[gv[b]t]
[gv[b]s]
[[b]t]
[g[b]en]
[g[b]t]
[[b]s]
[g[b]s]
[m[b]en]
[gv[b]]
[m[b]t]
[g[b]]
[m[b]s]
[v[b]en]
[m[b]]
[v[b]t]
[v[b]s]
[v[b]]
[[b]c]
[gv[b]en]
[a[b]]
[a[b]c]
Figure 3.11 Construction schema hierarchy for potential agreement marker combinations. Unattested combinations are indicated by dashed association lines.
exponents, given the arguments to be indicated. Aside from [SU: 2sg.] and [DO: 3], which are never overtly marked, other arguments are recoverable when their usual exponent is unexpressed owing to the following constellation of distributional facts: Whereas: 1. among the prefixes, only v-realises a subject argument; 2. all subject values other than 1st and 2nd singular have a suffix exponent; 3. among the suffixes, only -t realises an object argument, specifically 2nd person plural; 4. 3rd-person objects have no overt exponent; and 5. 1st-and 2nd-person subjects cannot co-occur with objects of the same person value, regardless of number (analytic-only reflexives). Therefore, the presence of g-[DO: 2sg.] and the absence of any suffix permit the inference of [SU: 1sg.] without its usual exponent. The syncretism found with gxedaven [SU: 3pl., DO: 2] may be unproblematic if the constraint favouring realisation of each argument can at least partially be upheld.
Time for a test drive 117 Table 3.12 Schemas and instantiating constructs. Form
Root only
Prefixed
xedav
Both arguments realised overtly?
Slots a and c filled?
2Sg.
3Sg. 3Pl.
(null)
both inferred
neither
(null)
both inferred
neither
3Sg.
(null)
a only (SU)
3Pl.
(null)
SU directly, DO inferred SU directly, DO inferred DO directly, SU inferred DO directly, SU inferred DO directly, SU inferred SU directly, DO inferred SU directly, DO inferred SU directly, DO inferred SU directly, DO inferred SU directly, DO inferred SU directly, DO inferred
c only (SU)
(ill-formed)
(both)
SU directly, DO inferred SU directly, DO inferred (ill-formed)
both (SU)
Pre- emption slot?
(null)
[a[b]] 1Sg.
v-
gxedav
1Sg.
v-
2Sg.
g-
mxedav
2Sg.
(null)
1Sg.
m-
gvxedav
2Sg.
(null)
1Pl.
gv-
3Sg.
-s
3Sg.
(null)
3Pl.
(null)
3Sg.
(null)
3Pl.
(null)
3Sg.
(null)
3Pl.
(null)
a only (SU) a only (DO)
a
a only (DO) a only (DO)
[[b]c] xedavs
xedavt
xedaven
Twice- affixed
Direct Object
[b]
vxedav
Suffixed
Subject
2Pl.
3Pl.
-t
-en
c only (SU) c only (SU) c only (SU) c only (SU) c only (SU)
[a[b]c] *vxedavs vxedavt
1Sg. 3Sg. 1Pl.
v- (none) -s v-. . .-t 3Sg. (null) 3Pl.
*vxedaven
(null)
1Sg. 3Pl. 3Sg. 1Pl.
v- (none) -en -s 2Sg. g- v-. . .-t 2Sg. g-
1Sg.
v-
3Sg.
-s
1Pl.
v-. . .-t
gxedaven
3Pl.
-en
2Sg. 2Pl.
mxedavs mxedavt mxedaven gvxedavs gvxedavt gvxedaven
3Sg. 2Pl. 3Pl. 3Sg. 2Pl. 3Pl.
-s -t -en -s -t -en
1Sg. 1Sg. 1Sg. 1Pl. 1Pl. 1Pl.
gxedavs gxedavt
2Pl.
both directly DO directly, SU partially DO directly, SU inferred DO directly, SU g-. . .-t inferred DO directly, SU inferred g- both directly g-. . .-t SU directly, DO partially m- both directly m- both directly m- both directly gv- both directly gv- both directly gv- both directly
both (SU) (both) a (DO), c (SU) a (DO), c (SU)
a
both (DO)
a
both (DO)
c
both (DO)
a, c
a (DO), c (SU) a (DO), c (SU)
c
a (DO), c (SU) a (DO), c (SU) a (DO), c (SU) a (DO), c (SU) a (DO), c (SU) a (DO), c (SU)
118 Contemporary Morphological Theories
Of the four distinctions realised syncretically by gxedavt, [SU: 1pl., DO: 2sg.] is like gxedaven for [SU: 3pl., DO: 2] in that the prefix is an exponent of the object argument and the suffix, the subject. In the cases of [SU: 1sg., DO: 2pl.] and [SU: 3sg., DO: 2pl.], the exponents of the more marked values preempt those of the less marked values in the same slot. Finally, in the realisation of [SU: 1pl., DO: 2pl.] also as gxedavt, the prefix reflects the object’s relative markedness, but while the suffix clearly realises plural number, it is ambiguous. Is it realising the subject at least once, or is the suffix, like the prefix, giving preference to realising the object’s markedness value in competition with a subject marker for the same slot. In a way, one cannot lose, since the -t suffix has a double warrant to appear. Although each of the constraints as proposed has independent motivation, they do at times conflict, and so some ranking may be required. The general Construction Grammar approach seeks to identify the constraints from various sources that bear on construction schemas and the constructs that instantiate them. In the present case, the morphotactic system of Georgian constrains agreement marking to up to one agreement prefix and up to one agreement suffix, and this interacts with an information-based (markedness and faithfulness) constraint hierarchy that governs the realisation pairings (slots evaluated in tandem), resulting in the most informative, well-formed candidate structure in cases where one or both slots are subject to competition. 3.2.5 Distributed Morphology In Halle and Marantz (1993: 117ff.) the Georgian agreement affixes are explicitly assumed to be clitics, rather than prefixes and suffixes per se, and so their morphosyntactic properties are fused into one proclitic, with the possibility of [+pl.] fission, allowing the -t to be inserted at the right edge of the stem. The v-/g-issue is handled as a fusional clitic, but the competition of -t with -s and -en is ignored. It seems safe to assume that a more complete analysis would handle all three suffixes as part of the clitic cluster, subject to fission as the -t is, but there may come a point when the morphological operations would be fewer if separate proclitic and enclitic clusters were generated, fused, and then supplied with phonological features. The clitic analysis allows DM a space apart from the rest of inflection to carry out clearly morphological operations without reference to the host. In a language like Georgian, which has little to no stem allomorphy conditioned by particular affixes, the clitic analysis is not demonstrably in error. In other languages, however, where more morphophonological operations accompany affixation, a comparable appeal to cliticisation might be subject to the possibility of falsification. Independent evidence of clitic status for the agreement markers (distribution, status of bound elements further out from the root, etc.) could serve to better demonstrate empirical, rather than theory-internal, motivations behind the choice of a clitic-based analysis. An alternative mechanism for accounting for disjunctivity among affixes is the operation of an impoverishment rule that specifies the deletion of particular morphemes (feature values) in the context of other specified values (Harley and Noyer
Time for a test drive 119
[1999] 2003: 477–9; Halle 1997: 431). A possible impoverishment rule that would stipulate the observed pattern would be the following: [+1]
→
Ø in the environment [ __, +2, –subject]
The prefix v-realises [+1] subjects, whether singular or plural, so there must be a crucial sequencing of this impoverishment rule to account for a form like g-xedav-t in which the subject’s [+plural] morpheme is realised as -t but the [+1] person specification is deleted via impoverishment, and thus the prefix v-is not insertable. It is not the lexical item g-that trumps the insertion of v-on this analysis, but rather the satisfaction of the conditioning environment for [+1] deletion. This proposed analysis does reflect the stated definition and operation of impoverishment rules, but this particular rule lacks independent motivation: why should there be an impoverishment rule at all, and why this one, given that there is no semantic conflict between [+1 subject] and [+2 object]? 3.2.6 Lexeme–Morpheme Base Morphology Verbs receive agreement specification in the syntax in LMBM, and the abstractly inflected verb is submitted to the Morphological Spelling component for the (incremental) spell-out of features, operating ‘outward’ from the phonological representation of the verb lexeme. The operation of the MS component in LMBM is described as a spelling mechanism which interprets inflectional features individually or in small groups (in the case of fusional exponents), executes the modification of the stem appropriate to the feature (set) in question – informed by the inflectional class of the lexeme at hand – and then immediately erasing the working read-only memory, beginning the process again with the next feature (set) as yet uninterpreted. This mechanism iterates exhaustively, and so disjunction between independently motivated affixes is not immediately predicted. Where g- precludes v-in a surface form, the most natural analysis is that the spelling mechanism has interpreted both arguments together and has spelled them as the canonical exponent of the object, presumably the first argument encountered in the set of inflectional features. The fact that the putative fused morpheme is phonologically identical to the 2nd-person object marker, while not exactly portrayed as an accident, does not follow from anything else in the grammar. Georgian is otherwise quite agglutinative – why this particular formal economising in the agreement system? This solution describes the fusion without really explaining it. An alternative view, also permitted within LMBM, is the MS component’s ability to selectively ‘erase’ features it finds in the output of inflection (cf. DM’s ‘impoverishment’ rules, §3.2.5). Inflectional features are ordered by syntactic structure, but no internal sub-bracketing of the features is available to the spelling mechanism. In other words, the MS component can ‘see’ the full set of features that an inflected lexeme bears, and it can act on those features in groups of up to five at a time (a hypothesised constraint on working memory, modelled as part of a finite-state transducer with a push-down stack; Beard 1995: 55–72). From this perspective,
120 Contemporary Morphological Theories
it is certainly possible to conceive of a language-specific spelling rule that says: ‘on encountering both {SU:1st} and {OB:2nd} in the same feature set, erase {SU:1st} (i.e. perform no modification to the stem) and spell {OB:2nd} as usual.’ The power of such a rule, and of this broad perspective for the spelling mechanism, is open to criticism, of course, but again it points up the ability of a theory which assumes the Separation Hypothesis to allow features to go unexpressed in the phonology, yet still be present in the representation, in a way that ‘morpheme-as-sign’ theories cannot. 3.2.7 Lexical Morphology and Phonology/Stratal Optimality Theory The v-morpheme can attach to a verb root, but just not to a 2nd person object prefix. This could be taken to suggest that the 2nd person g-belongs to a later stratum than the v-, and so the presence of the g-precludes the further addition of the v-from a previous stratum. The question is, if g-belongs to a later stratum, why should it get to apply first? On the different-strata analysis, we would be forced to say that g-must apply first for a language-specific and non-phonological reason, perhaps that object markers must be inserted first. In this case, then, g-would apply and preclude the insertion of v-. This is simply begging the question. An appeal to a template [subj | obj | root . . .] simply restates the distribution, that moving out from the root, the object marker is inserted closer to the stem. The stratum explanation is still required in order to effect disjunction. Since there is no co-occurrence of affixes from this group at the same end of any verb stem, and since there is no apparent morphophonology to account for in these data, it is questionable whether any independent justification would be forthcoming for the different-strata account. 3.2.8 Minimalist Morphology The data under consideration require no stem hierarchy, and so the work is entirely embedded in specifying the output (realised features) and the input (conditions for insertion) for the affixes. Respecting the principles given in Wunderlich and Fabri (1995: 262–3) of non-redundancy (no output information in the input conditions), markedness (only ‘+’ values in the output, unspecified features default to ‘−’), and input specificity (affixes apply to the most specific input forms available), the formulation for the Georgian affixes would be as follows, assuming set-valued features AgrS and AgrO to clarify parallel values for person and number: 0. Feature co-occurrence restriction: AgrS [person] ≠ AgrO [person] for [person] = +1 ˅ +2 1. /t/ AgrO: [+2, +pl.] / AgrS: [-pl.] 2. /gv/ AgrO: [+1, +pl.] 3. /en/ AgrS: [+pl.] / AgrS: [-1, -2] 4. /v/ AgrS: [+1] / AgrO: [-2]
Time for a test drive 121
5. 6. 7. 8.
/m/ /g/ /t/ /s/
AgrO: [+1] AgrO: [+2] AgrS: [+pl.] [ ]
/
AgrO: [-pl.]
/ /
AgrS: [+1] ˅ [+2] AgrS: [-pl., -1, -2]
MM rejects zero-affixes, so such elements cannot be used to introduce output content, even where this would potentially simplify the statement of generalisations (Wunderlich and Fabri 1995: 264). The device of feature disjunction can be powerful because it makes unnatural classes statable, but at a descriptive level it does permit the -t suffix (rule no. 7) to realise [+pl.] for non-3rd person subjects without either coining a ±3 person value (on top of ±1 and ±2) or placing illegal ‘−’ specifications in the output. The rules are in principle not extrinsically ordered, but they are presented here in descending order of specificity, and respecting a general feature hierarchy of Person > Number (Wunderlich and Fabri 1995: 264). The paradigm space in MM is ‘opened up’ starting from the most specific affix, and so if one applies these rules in the order above, the correct distribution is achieved. 3.2.9 Natural Morphology It seems that there is a certain irreducible amount of un-sign-like behaviour in the Georgian facts, whether the analysis involves zero-morphemes or syncretism between unitary and fused morphemes. Zero-marking of 3rd person is not exceptional on general markedness grounds, but for 1st person in the case of v-preemption, this is less expected. The approach of Mayerthaler ([1981] 1988: 8ff.), however, allows for a more sophisticated picture of markedness calculation. Typical attributes of the speaker are to be taken as background in a discourse context, not requiring especially salient marking in contrast with non-speaker attributes, which are to be interpreted as ‘figures’ in the foreground. With these two perspectives in place, then, the motivation for maintaining a marker for 2nd person, even at the expense of a 1st-or 3rd-person marker, is clear. Second person is more marked than 3rd, since there are indefinitely many 3rd person referents available in any given discourse situation. Second person is more marked than 1st, as well, since 2nd person is part of the non-speaker class. First and 3rd person do not conflict in the Georgian system, so no further hierarchical relationship is determinable. With an explanation for the dominance of 2nd person over 1st in hand, however, the motivation for preemption, as opposed to closer linear proximity to the stem (i.e. gxedav rather than *vgxedav), is not evident. Natural Morphology, therefore, would seem to have a piece of the puzzle that more formalist theories are forced simply to stipulate. NM is not, however, in a position to account for the disjunction, which would seem to be a purely formal matter.
122 Contemporary Morphological Theories
3.2.10 The Network Model The idea of a schema in the Network Model is thought to represent the connections which exist between words in the mental lexicon. Schemas can be defined phonologically, morphologically, syntactically, semantically, and in other ways as well. Prototypes or ‘best exemplars’ are thought, therefore, to serve as an organising principle for lexical categories. Based on the observed Georgian forms in Table 3.9, one could posit an abstract schema for inflected verbs: Pref – Stem – Suff This schema may be fully instantiated even when a particular argument lacks an overt exponent, since some arguments are realised partially by a prefix and partially by a suffix, e.g. {SU:1st, pl.} corresponds to a v-prefix and a -t suffix, all on its own.14 It could be assumed, therefore, that the simple schema is used as a guide in determining ‘acceptable Georgian verbs’, and hypothetical verbs which bear more than one agreement prefix or more than one agreement suffix will be judged as unacceptable. While this may work as a synchronic generalisation, of course, it offers no insight into how such an arbitrary limitation could come into being. Iconicity would predict at least one marker per argument, but this does not always happen, e.g. g-xedav ‘I see you’ (cf. §3.2.9). At the same time, patterns as regular and productive as this, i.e. agreement marking on verbs of all sorts, are predicted (albeit with some reservation in Bybee (1988)) to have a degree of independence from the words which instantiate them, since they are used freely in neologistic formations, etc. The extremely high frequency of the marking system, however, may yet have explanatory value if it is recalled that frequent forms more readily sustain idiosyncrasies, whereas rarer forms are subject to regularisation (e.g. replacing zero-expression with something overt). There is no competing system of markers for transitive verbs in Georgian, however, so perhaps at present there is no viable regularising pressure exerted on the observed pattern, and thus it remains firmly and indefinitely entrenched, despite its (regular) quirks. 3.2.11 Network Morphology In order to get the paradigmatic facts right, it seems that all subject–object combinations would have to be treated as units, sometimes realised as two bound morphs, sometimes one (and in the case of reflexives, partially periphrastically). When a combination of arguments is realised by a single morph, the marker would be a de facto fusional morpheme. == g_ “”
An unfortunate side-effect of this analysis is that the homophony of the putative fusional morphemes with the canonical exponents of one of the arguments so combined is portrayed as an accident. The possibility of casting the above fact (in the technical sense) as a rule of referral is but a slight improvement:
Time for a test drive 123 == “”
for there is no a priori reason to expect that any correspondence would exist, let alone such a close one, between a fusional and a non-fusional exponent. The rules of inference in the DATR format state facts about the realisation of morphosyntactic features without reference to the broader context of other rules applying in a given form. There is no ready way to capture disjunctivity here without a notion of competition between applicable rules. Since Network Morphology does not formalise a notion of slots or position classes, there is no natural way of inducing competition, or of predicting (rather than declaring) a ‘winner’, should such competition occur. 3.2.12 Paradigm Function Morphology Georgian agreement is taken up in Stump (2001: Ch. 3), partly in response to Anderson’s (1986, 1992) analysis, and in particular because of the challenge the facts pose for the Pān·inian Determinism Hypothesis: the assumption that for a given rule block, the narrowest applicable realisation rule is always uniquely identifiable, and said rule applies to the exclusion of all competitors in the same rule block. There are four agreement prefixes, v-, m-, gv-, and g-, and the realisation rules introducing these markers are the following: a. RRpref, {AGR(su):{PER:1}}, [V] () =def b. RRpref, {AGR(ob):{PER:1}}, [V] () =def c. RRpref, {AGR(ob):{PER:1, NUM:pl.}}, [V] () =def d. RRpref, {AGR(ob):{PER:2}}, [V] () =def These rules embody several assumptions: 1. they are all introduced by a single rule block (RRpref is not qualitatively different from RR2, or from any arbitrarily indexed rule block), and so constitute a position class; 2. there are distinct morphosyntactic features for subject and object a greement, identified diacritically, rather than structurally (cf. A-Morphous Morphology’s layered MSRs, §2.1); and 3. {AGR} features are set-valued features, rather than atomic-valued features (i.e. {AGR} takes feature-value pairs as its value). For evaluation purposes, it is important to compare the rules for narrowness and applicability. Rules (b), (c), and (d) are paradigmatically related, and so cannot co-occur for practical reasons. Rules (a), (b), and (c) all realise {PER:1}, and so, should the relevant properties co-occur in a particular context, the cell would be realised periphrastically as a reflexive construction, according to Georgian grammar. The only possible competition scenario, therefore, is between rules (a) and (d), the v-/g-conflict exactly. Rules (a) and (d) are apparently equally narrow, and both are applicable in extensions of {AGR(su):{PER:1}, AGR(ob):{PER:2}}. Stump’s response to this is
124 Contemporary Morphological Theories
to posit two modes of rule application, expanded and unexpanded. Rules generally apply in unexpanded mode, realising a particular morphosyntactic property set. Certain rules, however, are defined as applying in expanded mode, realising ‘every well-formed extension of a particular property set’ (Stump 2001: 72; emphasis in original). Rules applying in expanded mode are actually rule schemas, instantiated by each member of a class of rules applying in unexpanded mode. In the present case, the dominance and categorical applicability of rule (d) is assumed to be evidence of expanded application: d’. RRpref, ←{AGR(ob):{PER:2}}→, [V] () =def The arrows surrounding the second subscripted rule-index are the formal means of indicating expanded application. The effect of (d’) in Georgian will be to realise every well-formed extension of {AGR(ob):{PER:2}} with the g-prefix in the prefix slot, i.e. every inflected form which realises a 2nd-person object will have a g- prefix, and never any other agreement prefix. This approach is more restrictive than a theory which allows for the possibility of fully extrinsic rule ordering, because the constitution of rule blocks in PFM is fundamentally tied to distribution and position classes, whereas A-Morphous Morphology, for example, permits exponents which are realised in linearly distant positions to be part of the same rule block. PFM insists on localised competition. The rule schema approach also predicts that a schema cannot be preempted by another rule applying in expanded mode by definition. Schemas are therefore constrained, and can only be invoked where the ‘every well-formed extension’ criterion is met. The suffix -t which realises only – but not all – extensions of {NUM:pl.}, by contrast, cannot be handled with an expansion schema: e. Where a ≠ 3, RRsuff, {AGR(su):{PER:a, NUM:pl.}}, [V] () =def f. RRsuff, {AGR(ob):{PER:2, NUM:pl.}}, [V] () =def g. RRsuff, {AGR(su):{PER:3, NUM:pl.}}, [V] () =def h. RRsuff, {AGR(su):{PER:3}}, [V] () =def Again, to evaluate narrowness and applicability, rules (e), (g), and (h) are in a paradigmatic relation and cannot conflict. Conflicts between (e) and (f) are resolved in two ways: where a = 1, both subject and object are realised by a -t suffix, and so the resolution is vacuous (degemination or no); where a = 2, the combination entails a periphrastic reflexive construction. The remaining conflicts are between (f) and each of (g) and (h): (f) trumps (h) by narrowness, but this is not so for the relation between (f) and (g), which are apparently equally narrow. The effect of (g) is never overridden, and so the criterion of ‘every well-formed extension’ would seem to be met. Recasting (g) as an expansion schema: g’. RRsuff, ←{AGR(su):{PER:3, NUM:pl.}}→, [V] () =def
This analysis predicts very simply that every extension of {AGR(su):{PER:3, NUM:pl.}} will show the -en suffix, and this is indeed the case. The phonetic
Time for a test drive 125
r ealisation of (e) and (f) are distinct in some dialects, and so the splitting of the -t suffixes is diachronically and dialectologically supported, and not simply a theoretical expedient. The expansion schema as a theoretical construct preserves the Pān·inian Determinism Hypothesis, but there is no formally based explanation for why only those rules defined as expansion schemas are so defined. That part of the explanation may well be extra-linguistic. Stewart (2001) focuses attention on lexeme class, rather than property set, as an alternative in the evaluation of the relative narrowness of competing rules in a block. Since the set of verb lexemes that may have two arguments is a subset of the set of verbs, a rule that is defined as applying to two-argument verbs is narrower than one defined as applying to the category of verbs as a whole. Thus it would be the domain of applicability, not the range of properties realised, that is decisive in the case of Georgian, and the Pān·inian Determinism Hypothesis could be upheld without an appeal to a second mode of application. 3.2.13 Prosodic Morphology The mechanisms of prosodic morphology are actually superfluous here, since there is no morphophonological alternation, copying, or other non-concatenative operation to perform. Since there is no apparent synchronic phonological motivation for the g-/v-disjunction (both /gv/ and /vg/ are permitted in word-initial onsets in Georgian), there is no advantage here over what any purely concatenating (IA) analysis would offer. 3.2.14 Word-based Morphology The question of disjunctive ordering as it is usually discussed with reference to mechanisms of blocking and grammar-internal interactions is analysed in Word- based Morphology as an artefact of the constructive approach (see §2.14), which treats grammatical rules as ‘assembly instructions’ (Blevins 2006: 533–5). The dynamic of disjunction is reframed as ‘ultimately a property of the search process, in which more specific patterns must be checked first or else the default will always apply’ (Blevins 2003: 758). Rather than making overrides something that rules must do in a generative grammar that not only produces the acceptable but also rules out the unacceptable, from a WP perspective, it is the responsibility of cognitive appraisal of the established systems to sort out the preferred patterns on the basis of absolute and relative frequency. Thus the problem with forms such as *v-g-xedav-t-t for ‘we see you [pl.]’ is not that it cannot be formed in principle, but rather that it cannot gather any momentum for use in the context of a speech community that has already established an alternative full word g-xedav-t (or any other form) for the purpose. Furthermore, the recurrent upper limit of one agreement prefix and one agreement suffix is so frequent (more than 50% in this small portion of a much larger paradigm system15) that it may reasonably be hypothesised as abstractable
126 Contemporary Morphological Theories
from existing forms and usable in the making of form-based analogies to produce and parse new forms. The WP approach, once again, is in the business of predicting forms, not constructing them, and so even if this response of a ‘rigged system’ bleeding fuel from completely reasonable innovations offers little to no insight into how earlier generations of speakers came to use the same form g-xedav-t for three out of the four synthetic forms realising object agreement for [Per:2, Num:pl.] arguments, the cause of the syncretism was not to be found in the synchronic grammar to begin with. 3.2.15 Word Syntax Overt and phonetically null affixes are no problem for Word Syntax, so long as the surface form is consistent, e.g. 2nd-person singular subjects are consistently marked with a null prefix, 2nd-person objects by (at least) a g-prefix. Since words are built up exhaustively from morphemes in Word Syntax, every element of content (grammatical or lexical/derivational) associated with the inflected word is attributable to at least one constituent morpheme. This entails that in cases like [1pl.-subj., 2pl.-obj.] gxedavt there is some element in the morphological structure which contributes the [1pl.-subj.] specification. Since there is no such readily identifiable overt element, the simplest answer is to posit a phonetically null allomorph of the v-prefix (the -t suffixes are again a moot issue, owing to the general anti-gemination constraint in the language). The problem is that the distribution of this null allomorph is suspiciously specific, i.e. it ‘appears’ just in case the g-prefix is also called for (recall, however, that /gv/ and /vg/ are both attested word-initial onsets in Georgian). Another option here is to claim (as in §3.2.11 above) that a class of fusional morphemes serves to introduce specifications for subject and object at the same time. This would mean that the markers in Table 3.9 are not to be analysed as combinations of two independent morphemes, but rather are complex but unitary affixes that fuse multiple argument specifications. The result, however, would be a rate and degree of homophony with respect to the simple (single-argument) agreement marker system used with intransitives, e.g., that is not encouraging. 3.2.16 What is at stake here? The three Person values and two Number values that underlie the Georgian agreement system are conceptually transparent and cross-linguistically unexceptional as morphosyntactic categories go. The distribution of the formal markers of these categories, at least in the very limited data set considered here, is likewise not decidedly aberrant when surveyed holistically – of the twenty-eight forms in the paradigm, twenty-two show all warranted (non-null) agreement markers. No allomorphy as such affects these markers, and they do not change their position with respect to the verb root. The descriptive issue boils down to six forms, four of which show the same pattern of g- trumping v-, and the remaining two suggest a hierarchy of -en > -t > -s. At a semantic level, the combinations are all well formed, so the deviations from
Time for a test drive 127
full and explicit exponence are unexpected. Any proposed resolution of the conflict or competition must characterise (and manage) conflicts among features and/or affixes in a minimally stipulative manner, and without generating in the process any adverse implications for the formation of the remaining members of the paradigm. The key concept in the present section has been disjunctivity, and different frameworks have sought to conjure the effect within their existing assumptions. The facts show that in every case, the competition arises between affixes of the same positional sort: prefixes with prefixes and suffixes with suffixes. In the terms of frameworks that make use of phonetically null (zero-) affixes, there are no cases found in the Georgian paradigm in which a semantically motivated overt affix is blocked in favour of a zero. One cannot be certain, however, whether zeroes might (exceptionally) be permitted to ‘stack up’ with overt affixes, or whether zeroes are subject to preemption just as the non-null affixes are. A theory that is comfortable with structural zeroes in this setting, and furthermore that remains conscientious about their distribution, would want zeroes to be as much like other affixes as possible, rather than standing as special entities with their own rules. The cost of assuming unexceptional zero-affixes here, however, is that the number of cases of preemption in the paradigm rises from six to twelve (see Table 3.10); one would have to assume, for consistency’s sake, that an inaudible element (the zero-morph) has been prevented from ‘appearing’ in such cases on purely theory-internal grounds. Other means for accomplishing disjunctivity effects have included various takes on fixed structural templates consisting of slots, thus invoking a metaphor of scarce ‘real estate’ into which affixes may potentially be placed. The related notions of position classes, lexical strata, or rule blocks must likewise appeal to a metric for determining relative ordering for insertion. Blocking in these cases has generally been attributed to the functioning of some version of the Elsewhere Condition or the Pān·inian Principle, referring the question of ordering to relative specificity of content and/or conditions for (item) insertion or (rule) application. In incremental theories, unspecified roots must be given their feature values by insertions or applications, but when logically or even demonstrably present features that, all else being equal, would show an overt exponent, but in particular environments do not, there is the further obligation to delete or prevent their usual exponence. Special context-sensitive rules can often serve to make the facts come out right, but such rules often lack other justification. Making reference to relevant markedness hierarchies has the advantage of representing independently recognised cross-linguistic tendencies, and so when the facts at hand align with established markedness rankings (plural > singular number, non-3rd > 3rd person), it adds a measure of plausibility to theory-based claims of asymmetrical behaviour or treatment.16 By the same token, however, such evidence is ancillary rather than decisive, and can hardly be treated as causal within any particular language.
128 Contemporary Morphological Theories
3.3 S A N S K R I T G E R U N D S : P R E F I X – S U F F I X I N T E R A C T I O N The conjugation system of Classical Sanskrit verbs is notable for its extensive range of morphological expression: prodigious and ubiquitous compounding, meaning- and category-changing derivation, as well as a broad repertoire of inflectional forms (e.g. Whitney [1889] 1960). The full catalogue of roots, stems, theme vowels, and affixes (including distinct reduplication operations (Janda and Joseph 1986, 1992a)) – not to mention both internal and external sandhi processes – makes Sanskrit a descriptive project that is as exciting as it can be daunting. For the present purposes, the focus is limited to a single non-finite form that every verb may have, the gerund, also known as the indeclinable past participle (Rajguru 1938: 270–2) or the absolutive (Whitney [1889] 1960: 203). The gerund form is created in the general case by suffixing -tvā to the so-called ‘weak-grade’ root (roots are indicated by a preceding radical symbol ‘√’), e.g.: √bhū-‘be’: ger. bhūtvā ‘[after] having been’ or ‘[when X] had been’ (MacDonell [1927] 1986: 137) √jñā-‘know’: ger. jñātvā ‘[after] having known’ or ‘[when X] had known’ (Whitney [1885] 1945: 56) √vac-‘speak’: ger. uktvā ‘[after] having spoken’ or ‘[when X] had spoken’ (Gonda 1966: 78) The subject of this non-finite verb-form (‘X’ above) is interpreted as controlled by the grammatical (or on occasion, logical) subject of the main clause within which the gerund predication appears as an adverbial (Gonda 1966: 94). There also exists a derivational process in Sanskrit whereby an element with directional semantics called a preverb is prefixed to a verb root.17 Papke (2010: 2) presents the following set of contrasting derivatives based on the verb root √gam- ‘go’: ati-√gam- adhi-√gam- apa-√gam- abhi-√gam- ā-√gam- ni-√gam- vi-√gam- sam-√gam-
‘to pass by or over’ ‘to go up to’ ‘to go away’ ‘to go near to, approach’ ‘to return’ ‘to settle down upon’ ‘to go asunder, separate’ ‘to meet’
(over-go) (above-go) (off-go) (to-go) (towards speaker-go) (down-go) (apart-go) (together-go)
These derivatives are especially common with high-frequency verbs, with certain preverbs appearing only with semantically light √bhū-‘be’, √gam-‘go’, or √kr ̣- ‘make’. Whereas the inflectional paradigms of these derived verbs consistently follow the inflectional pattern of the verb root, the formation of gerunds for these derived verbs is markedly different. In lieu of the -tvā seen above, gerunds for preverb- derived verbs are formed by suffixing -ya (Burrow 1955: 371; Papke 2010), as in the following:
Time for a test drive 129
ger. nipatya ‘having fallen down’ (ni-‘down, into’; compare √pat-‘fall, fly’: ger. patitvā) (Mayrhofer [1964] 1972: 103; Whitney [1885] 1945: 94) ger. vimucya ‘having freed’ (vi-‘apart’; compare √muc-‘release’: ger. muktvā) (Gonda 1966: 78; Whitney [1885] 1945: 122) Combinations of more than one preverb on the same root are well attested, with linear order ‘in general determined only by the requirements of the meaning, each added prefix bringing a further modification to the combination before which it is set’ (Whitney [1889] 1960: 397), e.g.: ger. pratyāgatya ‘having returned’ (prati-‘reverse, back’; ā-‘(un)to, at’; √gam- ‘go’: ger. gatvā) (Deshpande 2003: 122, 428; Whitney [1885] 1945: 34) The issue for a morphological theory attempting to account for this phenomenon at a distributional level would seem to reside in the fact that the choice of suffix correlates specifically with the presence or absence of at least one category-preserving prefixed element. Not all Sanskrit preverbs are equally frequently attested in the formation of PV verb lexemes18, but independent of semantic or phonological aspects of the preverb chosen, the appearance of –ya in the corresponding gerund formation is effectively categorical. So as not to over-simplify, however, ‘preverb’ is not coextensive with ‘prefix applied to a verb root’, since the negative prefix a-/an-does not correlate with the shift to the -ya suffix: ger. akr ̣tvā ‘having not made’ (from √kr ̣-‘make’; cf. kr ̣tvā) (Whitney [1889] 1960: 356) Preverbs must be made a distinct class of some sort, therefore, in any analysis. If there is a question as to how closely bound the constituent units in these PV constructs are, Whitney ([1889] 1960: 398–9) notes as evidence of said closeness that although the P and the V elements would each have their own accent, in the paradigms of these derivatives the accent pattern, like the choice of inflection markers, follows the distribution of the base verb. Also at the morphophonological level, it is the rules of (word-)internal sandhi that apply to these derived verbs, rather than the partly distinct set of external sandhi rules that apply at word boundaries (399). 3.3.1 A-Morphous Morphology Anderson’s (1992) framework pursues an agenda of affording minimal significance to the role of visible or persistent word-internal structure in the interest of seeing the degree to which the impulse to parse out morpheme-like elements from complex words can be done without. The theory makes a fundamental distinction between compounding (based on two stems) and derivation (only one). The operation of compounding is assumed to create internal structure, in recognition of operations that need to pick out elements or boundaries inside the composite form in order to place and/or form their output correctly. The motivation for this structure
130 Contemporary Morphological Theories
visibility is particularly clear in cases of suppletive or inflection-class-specific forms, e.g. scrub-[men], not *[scrubman]-s. Anderson (1992: 259–60) initially predicts that the operation of derivation does not create visible internal structure, and any counter-examples that are raised are claimed to find alternative explanations in phonological attributes of bases or in paradigmatic support for the abstraction of patterns without necessitating directly ‘peeking’ inside an individual complex base. Carstairs-McCarthy (1993: 215–23) takes issue with Anderson’s defining distinction concerning the two types of word-creation, arguing and demonstrating that whatever need there is to allow visible structure in compounds as a class is equally required in derivational morphology; whether the internal boundaries can be done away with or not has a single answer for both domains. Carstairs-McCarthy (1993: 230) specifically holds up the case of Sanskrit PV gerund formation as a counter-example not simply to Anderson’s (1992: 303) opting to concede limited structure-building derivational operations (WFRs) on a purely stipulative basis, but also to a hypothesised locality requirement for such exceptions, namely that ‘a category-preserving WFR needs to “see” a word-internal pre-head or post-head boundary . . . only if the rule inserts material at that boundary’ (Carstairs-McCarthy 1993: 230). In order to select -ya instead of -tvā, the Sanskrit gerund-building rule must have access to the fact of the preverb’s presence (although the specific identity of the preverb is not relevant). As to locality, however, the placement of the gerund marker is not next to the visible boundary, but at the opposite end of the verb root. The problems raised by these structures might be defined out of existence by categorising PV lexemes as compounds rather than as prefixed derivatives, in which case only the locality condition is at issue. One might hope to identify a separate phonological condition for the choice of suffix (see §3.3.7 and §3.3.13 below), but that cannot be counted on a priori. This pattern in Sanskrit stands as an apparent challenge to the general devaluation of internal constituency with respect to subsequent derivational processes. Moreover, this should be much less possible for inflectional processes such as gerund formation, which A-Morphous Morphology assumes are carried out wholly in the syntax, and therefore quite unable to view structure internal to derivation. 3.3.2 Autolexical Syntax/Automodular Grammar From the Automodular perspective, components have access to category information as needed to do their respective work, so that the compositional semantic impact of the preverb and the adverbial interpretation of the gerund clause are worked out entirely within the semantic component, and any lexicalised idiosyncrasies of a particular PV are accounted for in the relevant lexical entry. The syntactic distribution is likewise determined in keeping with the adverbial function of the subordinate non-finite clause that the gerund heads. By the Feature Interface Principle, the feature [gerund] is available not only to the semantics and syntax, but also to the morphology (for marker placement) and furthermore to the
Time for a test drive 131
orphophonology (for marker shape). Each autonomous component, however, m derives a different significance from said feature in accordance with the localised concerns of its (modular) grammar. The relevant mismatch to be addressed in PV versus non-PV gerunds lies between the content of the feature [gerund] and the shape of the exponent to be selected. Assuming first two lexical entries, one for the ordinary verb root √gam-‘go’ and the other for the derived PV root abhi√gam-‘approach’, and distinguishing the two types of verbs by means of an arbitrary feature [P]: √gam- syntax: V morph: Stem [V] mphon: /ga/
abhi√gam- syntax: V morph: Stem [V, P] mphon: /abhiga/
the rule for forming gerunds could therefore be as follows: gerund syntax: [gerund] morph: Stem [V] mphon: 1. / . . . / 2. / . . . /
→ → →
Word [V, gerund] / . . .tvā / in [[V, −P], _ ] / . . .ya / in [[V, +P], _ ]
Such is the power of arbitrary feature designations that they can skirt issues like seeing inside complex stems. As a result of the PV derivation, the resulting complex verb bears a persistent tag that any component sensitive to it may detect and act upon accordingly. Other more familiar features such as conjugation class are handled in an analogous way (Sadock 2012: 151). Such also is the competence of the Interface component that it can virtually ‘see’ and simultaneously appraise multiple distinct structural representations of a given expression, e.g. a bracketing from the Morphological module and a selection between the allomorph options for gerunds on offer in the Morphophonological module. 3.3.3 Categorial Morphology In Schmerling (1983: 227), the definition of a morphological operation is sufficient to posit a set containing the outputs of said operation, and that set can be made subject to further operations, including independently defined affixation. In this spirit, the prefixation of one of the preverbs to a verb root is just such an operation, and the output of this operation can be designated as eligible for something as specifically targeted as ‘gerund formation by -ya suffixation’. The output of this latter operation would, by implication, constitute a further set for some hypothetical purpose, if needed, but so far this potential just so happens to remain unexploited. Once again, the locality constraint proposed by Hoeksema and Janda (1988: 218) – namely, that affixation ‘on level x is sensitive only to the properties of the [nearest] constituent on that level’ – runs into difficulty with the sensitivity of the -ya suffix to the presence of the preverb on the far side of the rightmost constituent of the base
132 Contemporary Morphological Theories
to which the -ya is attaching. If the PV derivation is construed as an attribute of the whole base, the constraint might be saved, but only at the cost of the constraint’s losing all predictiveness. 3.3.4 Construction Morphology In analysing the construction schemas that give rise to PV gerunds in -ya, both the derivation of the PV and the suffixation must be defined separately, and then brought into harmony, in a manner of speaking. The PV would prefix morphs of the class P to verb stems, resulting in a subclass of verbs: [P -Vstem]Vp-stem. The construction schema could remain at this rather abstract level, since it is well attested, if not fully productive. A further consideration, however, in light of the possibility of recursive prefixation (e.g. anuvyāhṛtya ‘having repeated (words)’ [anu-vi-ā-hṛ-(t)ya],19 from √hṛ-‘take, carry’), is that the prefixation must apply to verb stems, but these should not be specified in the construction schema as morphologically simple (unprefixed) stems. There are particular sequencing constraints that hold in cases of multiple preverb affixation (Papke 2010), but the conditions are more a matter of possible and/or useful spatial relations than of purely morphological compatibility, so the sequencing issues are better assigned to the semantic and/or pragmatic constraints on the relevant construction schema(s). Having defined a PV construction, its product is a category that the -ya suffixing construction can refer to as a morphological constraint: [VP-stem -ya]Vgerund. There is an additional level of significance that falls out from a constructional analysis as well, that is, the -ya not only realises [gerund] but also signals the PV identity of the base it attaches to, in concert with the unusual stress on the V root (whereas it is on the -tvā suffix in non-PV gerunds)20. The construction schema for PVs in general, and PV gerunds in particular, may make reference to one or more levels of structure as defining constraints:
[[[P–Vstem]Vp-stem]-ya]Vgerund …V(C) [accent] Figure 3.12 Construction schema for PV gerund formation in -ya.
3.3.5 Distributed Morphology A theory which places all morphological construction into a syntactic frame of reference will naturally emphasise the hierarchical elements of morphological
Time for a test drive 133
phenomena. The root for any verb used is present from the earliest stages of DM derivation – all else in the tree besides major category roots is abstract. The argument would be made that the preverb would have to be brought together with the verb head, because owing to the segmentability of the preverb with respect to the V root, an argument in support of a ‘complex root’ would be out of character for the framework. The preverb might be left-adjoined to the verb root, or the case might be made for merger (not fusion) under one node. Adjunction would seem to be sufficient overall, given that the vast majority of PV inflectional behaviour follows from the conjugation class of the base V. In order for the gerund morpheme to be realised, its node has to be unified with the root for purposes of Vocabulary insertion, and as a non-finite feature value, the association will be with a Tense node in the subordinate clause. Association of a Tense node with a V head is handled in the general case in the DM literature via morphological merger (Halle and Marantz 1993: 135–8), in keeping with the geometric position of Tense above and to the left of V in the syntactic tree (Kayne 1994), but recognising the need to end up linearly after the V root for purposes of Vocabulary insertion in the final Sanskrit word-form. Having assumed that ‘the [Vocabulary] insertion operation has available the entire syntactic tree so that insertion at a given node may make reference to features at other – primarily adjacent – nodes,’ Halle and Marantz (1993: 136) thereby laid the groundwork for handling phenomena such as this somewhat remote sensitivity within DM. The option of looking farther abroad to ensure correct insertion is sufficient for the derivation to cross the finishing line of PF interpretation with the right pieces in the right places. 3.3.6 Lexeme–Morpheme Base Morphology LMBM places lexical insertion and lexical derivation as feeding into D-structure, such that the syntactic structure that undergoes any movements on the way to S-structure is ‘populated’ with the major-category lexemes (N, V, and A). In the passage from D-structure to S-structure, inflectional categories are assigned (in tandem with movement rules) to the relevant positions in the syntactic structure. Then, and only then, do the morphological spelling (MS) operations come into play, providing phonological content to the as-yet-unrealised nodes in the tree. With this as prologue, the derivation of Sanskrit PV gerunds must touch on each stage of the way. As discussed in §2.6, LMBM maintains the existence of a universal set of primary and secondary functions (see Table 2.10), and these may be marked by (i.e. spelled out as) preposition-particles (free grammatical morphemes) or potentially as derivational prefixes. Among the grammatical function inventory, it is said to be only ‘the prepositions marking secondary but not primary functions [that] are available for L-derivations’ (Beard 1995: 248). In light of this prediction, a brief analysis of the preverbs with respect to the inventory is in order.
134 Contemporary Morphological Theories Table 3.13 Preverbs and the grammatical functions they realise. Preverb
Core meaning
Grammatical function(s)
ā- abhi- adhi- antar- anu- apa- api- ati- ava- ni- nis- (~nir-) para¯- pari- pra- prati- sam- ud- upa- vi-
‘to, unto, at’ ‘to, unto, against’ ‘above, over, on(to)’ ‘between, among, within’ ‘after, along, toward’ ‘away, forth, off’ ‘unto, close upon, on’ ‘across, beyond, past, over, to excess’ ‘down, off’ ‘down, into’ ‘out, forth’ ‘to a distance, away’ ‘around’ ‘forward, onward’ ‘reverse, back, against, in return’ ‘along, with, together’ ‘up(ward), out’ ‘to, unto, toward’ ‘apart, asunder, away, out’
[Goal], [Location] [Goal], [Opposition] [Superession ([Location])] [Intermediacy] [Posteriority], [Prolation], [Goal] [Origin] [Adession ([Goal] v [Location])] [Transession] [Subession [Origin]] [Subession [Goal]] [Goal] [Goal] [Circumession] [Goal] [Opposition] [Prolation], [Sociation] [Superession [Origin]] [Goal] [Origin]
Multiple preverbs are claimed to refine or otherwise modify the semantics of the (prefixed) base to which they attach,21 so the grammatical functions should be composed, sometimes acquiring figurative or lexicalised meanings in relation to certain roots. Table 3.14 Preverb combinations and their complex grammatical functions. Combinations
Grammatical functions
Example
vi-ava-
[Origin] + [Subession [Origin]] [Opposition] + ([Goal] v [Location]) [Origin] + ([Goal] v [Location]) [Posteriority] + [[Origin] + ([Goal] v [Location])] ([Goal] v [Opposition]) + [Subession [Goal]] [Opposition] + [Subession [Goal]] [Circumession] + [Goal]
vyavahr·tya ‘having behaved’ (√hr·-‘carry, take’) pratyaˉgaya ‘having returned’ (√gam- ‘go’) vyaˉhr·tya ‘having spoken’ (√hr·-‘carry, take’) anuvyaˉhr·tya ‘having repeated (words)’ (√hr·-‘carry, take’) abhinivisya ‘having resorted to’ (√vis- ‘enter’) pratinivr·tya ‘having returned’ (√vr·t- ‘turn’) paryupaˉsya ‘having attended upon’ (√aˉs-‘sit, remain’)
prati-aˉ- vi-aˉ- anu-vi-aˉ- abhi-ni- prati-ni- pari-upa-
These preverbs are not functioning as prepositions per se in their transitive (object- taking, NP case-governing) sense, in that often the preverb simply indicates the direction or path of the activity denoted by the verb, just as a free adverb might (e.g. forth, upward), and in some cases, the modifications of meaning seem to
Time for a test drive 135
militate against Beard’s claim that only secondary functions may be used in L-derivation, since although [Location], [Goal], and [Origin] are spatial functions, they are presented among the primary functions (Beard 1995: 303). Beard (1995: 364–9) hypothesises long-distance spelling as an alternative to movement operations to place clitics, for example, or to spell out only the phonological content of English auxiliaries22 directly in C(omplementiser) position within interrogative structures, while the auxiliary’s grammatical features remain in T(ense).23 On a broader level, MS is granted access to the total set of features associated with a word – namely, (1) those inherent in the base/root, (2) any features accreted through L-derivation operations, and (3) those corresponding to the syntactic configuration, including agreement – in order to ensure correct spelling. The use of the feature-set by MS is represented as an informal mapping from a feature matrix to exponents. The feature matrix furthermore includes both before-derivation and after-derivation values, so that the MS operations may pick out the information that is relevant for each operation’s application. The selected sets may be adjacent, nested, or overlapping (Beard 1995: 372–3), and as such, the mechanism is in place for the PV structure of the stem, in conjunction with the feature [gerund], to determine the selection of the -ya suffix in place of the default (and more locally determined) -tvā. 3.3.7 Lexical Morphology and Phonology/Stratal Optimality Theory The rule(s) implementing the attachment of a preverb to a verb stem should be a lexical rule, selecting the weak stem of the verb as base, and fitting into the otherwise quite general ‘stream’ of internal sandhi and other (morpho)phonological rules. The status of the individual sound-structural rules, whether they are more beholden to morphological structure or to phonological conditioning, is in the final analysis not problematic, provided that they do not show evidence of backsliding to earlier (already completed) levels. The dependency makes clear that the preverb must attach before the suffix but, as a phonologically oriented approach to morphology, opting to depend on an abstract, specifically morphotactic fact of [±complex base] is very uncomfortable in the determination of what should be a late, if not post-lexical, rule associated with the attachment of the word-closing suffix -tvā, or alternatively, -ya. An explicit tenet of LM&P is bracket erasure at the end of each level, which should irretrievably prevent the fact of a base’s morphological complexity from conditioning allomorph selection within a subsequent stratum. Even if the preverbs were part of the same level as -tvā, distant conditioning is problematic, all the more so if the conditioning is grounded in some ostensibly phonological attribute possessed by the class of preverbs, which upon further inspection form a rather diverse group at the segmental level at least. The phonological sensitivity of LM&P and Stratal OT (among others, especially Prosodic Morphology §2.13) permits more attention to be paid to the suprasegmental character of the verb stem, which provides some subtle (and, it so happens,
136 Contemporary Morphological Theories
linearly adjacent) clues that a phonologically grounded affixation rule could plausibly take as a relevant conditioning environment. The finite verbs of main clauses are, as a rule, unaccented unless they are clause- initial (Whitney [1889] 1960: 223). In a dependent clause, however, ‘the verb is accented, whatever its position’ (225), and although this stipulates the presence of an accent, it is not necessarily straightforward to predict exactly which syllable of a given verb will bear said accent without several further pieces of lexical and grammatical information. For the present purpose, it is sufficient to note that as a non-finite form, and given the adverbial function of any gerund-headed clause, a gerund will be accented. When forming the gerund of an underived (non-PV) verb, the simple root (or a slightly weakened version thereof) serves as the stem, and it is the -tvā suffix that has the accent (356). The stem of a PV that is used for gerund formation is likewise the weak (or weakened) form of the stem, but this stem nevertheless bears the accent, never the suffix -ya. The sensitivity of the -ya suffixation rule to the character of the stem is such that if that stem ends in a short vowel, an allomorph -tya is attached instead (356–7 and n. 15), closing and adding weight to the stem-final syllable, and incidentally better supporting the accent that must appear there. If the focus shifts from a templatic or hierarchical characterisation to a finer- grained description of the phonological attributes of the preverb prefixation rule – namely, that it selects for attachment an independently available stem that is recognisably distinct from the root, then the rules affixing the gerund endings may not need to ‘see’ the preverb, only its effect on the immediately adjacent stem-final syllable. Whereas staying resolutely at the level of unanalysed m orphemic pieces entails serious problems that require adding descriptive power and/or weakening constraints, the potentially neat alternative analysis above emerges only through digging deeper into the phonological material of morphological formatives. This said, however, the functional equivalence of -tvā and -ya (~-tya) remains unpredicted, and in fact constitutes a stealthy suppletion: there is no derivational relationship between the two markers, just complementary distribution (cf. Carstairs-McCarthy 1988, 1993). 3.3.8 Minimalist Morphology It is the custom in the MM literature to set up a hierarchy of the stems that are used in a particular lexical item’s paradigm, designating each with the grammatical properties for which it may be used. For the (mainly Germanic) languages described in the framework’s leading articles, a segmental phonological representation has generally proved sufficient.24 This may prove unwieldy in the case of Sanskrit, just on the grounds of the size of the paradigms alone, not to mention the many conjugation classes and subclasses,25 but a limited and slightly schematic representation could give a flavour of the analysis. MM is avowedly a theory of inflection (Wunderlich 1996: 93), and so it does not put forward an account of derivational processes. It would seem therefore to take the PV lexical items as verbs alongside all others in the category V. The problem
Time for a test drive 137
would seem to appear immediately – namely, that all verbs are potentially subject to gerund formation – but there is evidently a class of verbs that inherit the structure of their entire paradigm from the verb root (however one chooses to register the connection between source and derivative), except for this one inflectional form. Without access to the morphological complexity of the PV stems, the class would have to miss the structural generalisation – namely, the preverbs – and instead coin an ad hoc, poorly motivated inflectional class on the basis of the exceptional gerund affix that they take. If the presence or absence of stem accent can be represented in the stem hierarchy, and there seems no reason in principle that it cannot, the choice between gerund affixes could be determined on less tenuous grounds. The fact that all Sanskrit verbs – with or without preverbs – possess stems that are weaker and stronger, accented and unaccented, will be recognised in all the respective hierarchical lexical entries, but the key to pulling the analysis together is the designations that are attached to the different stems.26 For morphologically underived verbs, the unaccented weak stem will be indexed for use with [gerund] (among other uses), and so will take -tvā. PV lexemes, on the other hand, will designate their accented weak stem for the [gerund]. Again, the stem designated for this use in the hierarchical lexical entry of PVs will likely be associated with other purposes beyond the formation of the gerund, but it should be noted that – given the overriding congruity of PV paradigms with those of their source verbs – any other uses of the accented weak stem will systematically not be the ‘other uses’ that underived verbs assign to the unaccented weak stem. The label [gerund] will be part of a distinct disjunctive list of functions and located at a different node in the stem hierarchy for the two types of verbs. This detail, however, in no way solves the distribution of the -ya suffix, which, like -tvā, will be inserted via a rule sensitive to the [gerund] context. Some means of forcing the issue (such as sensitivity to derived status or the [+PV] diacritic alluded to elsewhere) remains a necessary deciding factor between the otherwise identically specified suffixes, but with arbitrary inflectional classes outlawed in MM, neither of these strategies is an obvious fit. 3.3.9 Natural Morphology On the derivational side, the semantic refinement of a verb’s meaning by the contribution of the preverb is performed in a constructionally iconic way: that is, more meaning ↔ more form. Morphotactically, the boundaries between root and affix and between pairs of affixes are mostly transparent, or at the very least recoverable, given the morphonological context of Sanskrit’s general rules of internal sandhi and stem-strength alterations. With respect to the inflectional suffixes, any state of allomorphy constitutes a deviation from biuniqueness, but -tvā and -ya are not related to one another, and the degree of alternation at the segmental level between the two compromises the affix’s effectiveness as a sign from the NM perspective. As a counterpoint, one might note that the choice between the two affixes is clearly rule-governed.
138 Contemporary Morphological Theories
There are in fact a number of formatives akin to -ya in Sanskrit including stem-forming suffixes: verb class IV present stem (in -ya), future passive participles (so-called gerundives, also in -ya), the passive conjugation (in accented -yá), a deponent intensive (also -yá), denominative verbs (also -yá), and the causative formation (in -aya). This multiple deployment of the same (or very similar) formal marks for diverse functions in the morphological system of a single language is of course not limited to Sanskrit (cf. Janda 1982; Janda and Joseph 1986: 104), but wherever such re-use occurs, it defies the biunique ideal. The -tvā suffix, by contrast, is tightly linked to the gerund function.27 The expectation of productivity in ‘natural’ morphological processes (Dressler 2005: 279; see §4.2, below) is certainly met by -tvā suffixation, since it applies to form the gerund of any basic verb. The use of preverbs is also quite free in principle, although there is a distinct attraction to higher-frequency basic verbs. Furthermore, the status of preverbs as optional modifiers (at least in unlexicalised uses) reduces their overall frequency if compared to the obligatory availability of inflectional categories. Even the -ya marker of the gerund is productive if it is considered in context: for any PV that one might construct, its gerund form would be marked by -ya (or -tya, depending on the root vowel). 3.3.10 Network Model In this framework, generalisations are detectable and thus representable via the technique of network drawing (Bybee 1988; see §3.1.10). Word-forms that largely coincide are deliberately selected for node status in constructing such networks, so that the relative strength of the linking branches and the density of connections between word-forms can indicate by contrast the nature, location, and degree of any differences. In order to represent the PV gerunds in Sanskrit (e.g. atigatya ‘having passed by’), a network that contrasts an exemplar of such a form is networked with a non-gerund PV (atigacchati ‘s/he passes by’), a non-PV gerund (gatvā ‘having gone’), and a non-gerund non-PV (gacchati ‘s/he goes’). In this way, the systematic relations among these constructs can be brought out in a concrete manner. atigát ya
a t i g á cch a t i
g a t v ā ́
g á cch a t i Figure 3.13 Network for PV gerund and contrasting forms.
Time for a test drive 139
Comparison with another PV in an analogous network, e.g. atikrāmya ‘having transgressed’, would reinforce the overall patterns seen with atigatya and further highlight the boundary after ati-, but this time overtly with contrasting bases, rather than by inferring this from the prefix’s absence in the non-PV forms alone. The next step after this would involve parallel contrasts with other PV gerund networks. The replication of the preverb identification-by-contrast process, together with the respective gerund suffix distributions, would in time make clear both the membership of the preverb set and the correlation of the preverbs with -ya suffixation in the gerund. Note also the slightly lighter link between the accented and unaccented stem vowels in Figure 3.13, constituting a further formal distinction in the patterning. 3.3.11 Network Morphology Network Morphology treats derivation as joint but asymmetrical inheritance from two orthogonal hierarchies (Brown and Hippisley 2012: 251). A derived lexeme inherits from its base, a node in the lexemic hierarchy, and from a lexeme- formation template (LFT; Brown and Hippisley 2012: 260; cf. WFR), a node in the derivational hierarchy. A derivative may differ from its base in potentially multiple aspects as a result of multiple inheritance, which may (1) specify information not present in the base, (2) elaborate on information that is less than fully specified in the base’s lexical entry, and/or (3) override facts inherited by default from the respective source nodes. At a more informal level, the following schema presents a comparison of a base verb and a derived PV: Table 3.15 Example of schema used for V to PV derivation. √muc-
vi√muc-
syntactic level syn cat = V args = 2 (NP_NP) semantic level ‘release’ phonological level stem 1 = /muʧ-/ stem 2 = /muk-/ morphological level mor_class = V_6
syntactic level syn cat = V args = 2 (NP_NP) semantic level ‘(set) free’ (lit. ‘release-away’) phonological level stem 1 = /vimuʧ-/ stem 2 = /vimuk-/ morphological level mor_class = V_6p
>
All four levels of information must be accounted for, because any given WFR may introduce one or more changes at any or all levels. In the present case, the derivation is category-preserving, but there is activity at all other levels. Translating this correspondence between base and derivative into DATR formalism produces the following (cf. Brown and Hippisley 2012: 253):
140 Contemporary Morphological Theories Muc: < > = VERB == V_6: == ‘release’ == muʧ == muk == 2.
By this entry, √muc-inherits its paradigm structure from the class VERB and its particular morphological exponence patterns from the sixth conjugation class. Its gloss, valence, and root shape are lexically specified, and the stem-final consonant alternation is registered.28 Vimuc: < > == LFT_PREVERB_VI == “Muc:< >”.
The entry above shows the lexemic inheritance of all features from the base verb by the derived verb, with the integration of features inherited from the LFT specific to the preverb vi-, defined as follows (cf. Brown and Hippisley 2012: 266): LFT_PREVERB_VI: < > == LFT_PREVERB == apart == vi.
This LFT itself inherits from the more general LFT below, which characterises preverb derivation generally. In this meta-LFT, the attributes that pertain to the entire class of preverbs are stated – including at last the generalisation concerning the -ya suffix for the gerund. LFT_PREVERB: < > == LFT_CAT_PRESERVING == direction == “” == “” == “” ya.
The preverb LFT in turn inherits its application dynamic from the still higher-level meta-LFT of category-preserving derivation via prefixation (Brown and Hippisley 2012: 266). LFT_CAT_PRESERVING: < > == LEXEME == “” == lx [“” (x) & “” (x)] == PREFIXATION.
Time for a test drive 141
As can be seen, the sources of inheritance at this level include rather abstract categories. Given this ascent of the lexical and derivational hierarchies, one may appreciate the role of default inheritance (as well as overrides) in capturing generalisations about the behaviour of lexemes, word-forms, and derivation in Network Morphology. The test of the system is a review of the theorems that are specified for the derived word, based on its place at the intersection of inheritance from the lexeme √muc-and LFT_PREVERB_VI. Vimuc: = v. Vimuc: = V_6. Vimuc: = release. Vimuc: = muʧ. Vimuc: = v. Vimuc: = V_6. Vimuc: = l(x) [ apart (x) & release (x) ]. Vimuc: = apart. Vimuc: = vimuʧ. Vimuc: = vimuʧya.
Given the corresponding paradigm of √muc-for reference, the full paradigm of vi√muc-would seem to be predictable, at least in principle, from this basis. 3.3.12 Paradigm Function Morphology Although PFM has devoted the majority of its attention specifically to inflectional processes and patterns (Stump 2001), the PV derivatives in Sanskrit have provoked discussion. In Stump (1993c), the gerund formation for PVs is analysed as involving the derived PV stem serving as input to gerund inflection, resulting in derivation being hierarchically inside inflection, as expected under the Split Morphology Hypothesis, and the very definition of the Affix Ordering Hypothesis. At the same time, it is noted that an inflectional prefix a-, used in the realisation of the imperfect, is systematically placed between the members of the PV, thus intercalating inflection within derivation, contrary to the Affix Ordering Hypothesis. Stump (1993c) treats the preverbs as non-head members of preposition–verb compounds, but more importantly, they are discussed under the rubric of category- preserving rules and the characteristics such rules have in general. Category- preserving rules allow considerable inheritance from a base by a derivative: not merely the syntactic category (thus the name), but also at least one morphosyntactic feature specification (19). Rejecting the hypothesis that affixes can be morphological heads (cf. Word Syntax §2.15), the claim instead in PFM is that category-preserving rules, be they derivational or compounding, produce headed structures – either a root from a headroot or a word-form from a headword-form, the former (derivative) corresponding to the latter (head) in the morphosyntactic property set realised by the head. This is in keeping with the inheritance among PVs of the conjugation class of
142 Contemporary Morphological Theories
their respective head verbs, as has been seen. The morpholexical rule (MLR; Stump 1993c: 29) for realising PVs is formalised as: MLRcompound([P x],[V z]) = def [V [P x][V z]] (Head: [V z]) Stump interprets the situation of the imperfect marker a-inside PV structures as evidence for the Head Application Default, whereby the inflectional marking of a headed structure that results from the application of a category-preserving rule will tend to be placed with respect to the head member of the complex rather than at the edge (Stump 1991, 1993c, 2001, 2005). In Spencer’s (2013) Generalised PFM framework, there is an integrated mechanism for approaching a broader range of lexical relations in a unified fashion. A GPF would take as arguments a verb base and a particular preverb to attach to it, returning a complex, headed verb structure: schematically GPF(), or as a specific example, GPF(), which would return ni√pat-‘fall down’. Paying out the fuller attribute structure proposed for GPFM: ≡ GPF() fform() = x |verb| fsyn() = Verb fsem() = [DIRECTION(x), [VERB(x,y)]] fli() = PVx(verb) GPF() ≡ fform() = ni pat- fsyn() = Verb fsem() = [DOWNWARD(x), [FALL(x)]] (intransitive) fli() = ni√pat GPF() ≡ fform() = ni pat ya fsyn, fsem, fli () = (identity function applies) The identity function is motivated under inflectional realisation: that is, within the paradigm of a single lexeme. While the representation of the lexeme as retaining the radical mark (√) may seem to signal the status of the PV as a derived entity, it is a trick of the formalism in the absence of a more elaborated theory of the feature structure. Spencer (2013: 184–9) provides such a device in the morpholexical signature, an attribute- value matrix familiar in general from theories of a hierarchical lexicon (e.g. HPSG, Riehemann’s version of Construction Morphology §2.4). If it is a matter of sensitivity to the prefix, as has been considered elsewhere above, the morpholexical signature does not in itself provide a persistent indication of derived status. The signature breaks out indexed stems (in the mode of ‘classic’ PFM) within the fuller AVMs of lexemes, but the LI for the whole is established once and remains stable across all stems.29 The morphological category and inflectional class are also shared by default. If neither the LI nor the stem indices can trigger the special gerund exponent for PVs, then a targeted realisation rule may have to be coined within the form component of the PVx family
Time for a test drive 143
of derivational relationships, such that the lexical entry for any verb derived by a PV relationship will, as a concomitant effect, be specified for a -ya exponent of {gerund}. This is perhaps a comparatively blunt instrument, but something rather similar will need to be invoked for suppletion in general, and where suppletive affixes correlate with independent derivational relationships, the conditioning of the suppletion has to be lodged within that derivation, rather than separately in each derivative. 3.3.13 Prosodic Morphology Again in a phonologically oriented framework, the approach of choice will most likely focus on the prosodic characterisation of the stems in PV versus non-PV gerunds, and the method for doing so here is straightforward. The P prefix and the V have their own lexical accent associated to one syllable each. When the two morphemes are concatenated, one of the accents, that of the non-head member (or alternatively, the word-initial or leftmost accent) is delinked from the accent-bearing unit. One might invoke the Obligatory Contour Principle (OCP; Leben 1973) to disallow two consecutive accented morphemes within a word, or in OT terms, a markedness constraint *Complex(accent) or *Non-headAccent could provisionally serve to cancel the P’s accent. [accent]
[accent]
[accent]
[sam]
[gam]
[sam
[accent]
+
gam]
Figure 3.14 Preverb loses its accent (by OCP) when combined with accented verb stem.
When the default gerund marker -tvā is attached to the stem, it effectively ‘pulls’ the accent from the root, or alternatively, -tvā has its own lexical accent and the same OCP-based explanation as above deletes the accent from root. [accent]
[ga
[accent]
+
tvā]
Figure 3.15 Verb stem loses its accent (by OCP) when accented -tva¯ is suffixed.
144 Contemporary Morphological Theories
The alternative gerund suffix -ya, however, is unaccented and thus attaches freely to an accented stem without ‘pulling’ the accent to itself. [accent]
[ga
+
ya]
Figure 3.16 Verb stem retains its accent when unaccented -ya is suffixed.
If the stem to which the -ya suffix is to be added ends in an accented short vowel, the allomorph -tya is selected (Whitney [1889] 1960: 356–7), still without pulling the accent, and furthermore boosting the weight of the root by closing its final syllable. This is represented here as epenthesis of a (perhaps default) consonant / t /.30 [accent]
[accent]
[ … g a + y a]
[accent]
[…ga +
t y a]
[…ga +
t y a]
Figure 3.17 Suffix -ya becomes -tya when suffixed to accented stem ending in short vowel.
3.3.14 Word-based Morphology The paradigm of any given PV is predictable based on the conjugation of the base verb in all cells save one, the [gerund]. The case of the PV gerund constitutes an override of the formal analogy with the base verb, and this is sustainable in the system as a regularity associated with the PV formation in general because not only is it productive in terms of its own type frequency, but also any given verb has, in principle, as many options for combining with one or more preverbs as make sense semantically and/or pragmatically. Upon closer observation, the sequences and combinations of preverbs are not in fact free (Papke 2010), but with high- frequency, low-specificity ‘light’ verbs as bases, the options for PV combination increase along with the communicative motivation to modify the basic verb in order to indicate adverbial-type details of direction or path, either literally or figuratively.
Time for a test drive 145
Even when particular PV combinations become lexicalised and develop a measure of semantic opacity, the exceptional gerund suffix remains in place. Diachronically, one might reasonably expect that a derived PV, if bleached of its original directional semantics through conventional use, e.g. parinayati ‘s/he marries’ (lit. ‘leads around’),31 might, in the absence of the supporting cultural context, be reanalysed, leave the single exceptional form behind, and become aligned (under strengthened analogical pressure) with the full regular paradigm of the base verb √nī-‘take, lead, carry’ (conjugation class 1). 3.3.15 Word Syntax The verb must be the head of the structure on syntactic, semantic, and (quasi-) universal (RHR) grounds, as well as based on the (technically indirect) evidence of the paradigmatic parallels to the forms of the verb without the preverb. This evidence is to be considered indirect because each of the forms would be derived independently, using distinct morphemes. If the verb is the head, the gerund inflection cannot gain access to the presence of the non-head preverb, and so the rule to insert -ya should not be able to be correctly triggered. If the preverb were exceptionally promoted to head of the structure, its category would percolate to the higher node, and thus not permit the attachment of a verb-specified inflection, and so this boosting of preverb visibility by stipulation does not constitute an improvement. Percolation of head features is a standard feature of Word Syntax tree structures, and as such, this is consistent with the retention of all of the basic inflectional patterns associated with that head verb in the derived PV. If the subcategorisation of the gerund suffix -ya were specified not merely as attaching to verbs V]__V], but as being further sensitive to a diacritic [+PV] that stems so derived would bear (Lieber 1981: 35–6), then the triggering environment is in place as simply as that. There are two reasons why this is perhaps less than compelling, however. First, having met up with diacritics in the context of LM&P, level ordering, and Stratal OT (§2.7), the best-established lexical strata are tied to etymological sources [+Germanic], [+Latinate], and so on. A putative [+PV] diacritic would be based on derivational history, not etymological distinctions, and because PV derivation is productive, these formations constitute an open class, and thus the number of them can increase at any time. The more disconcerting aspect of this solution is that the diacritic is coined to solve exactly one problem, the distribution of a single suffix. Nothing else in the language needs the PV badge to be visible; the derivatives themselves, once derived, move on as more or less ordinary verbs, with all their paradigmatic forms otherwise neatly determined. Furthermore, if this diacritic is to be like the [+Latinate] diacritic, those roots not meeting the criterion are marked as ‘−’ for the feature (Lieber 1981: 37). In the present case, this entails that every other verb in the language is [–PV]. Reframing ‘−’ as ‘under-specified’ is perhaps a minimal improvement, but hardly a generalisation. In order to arrive at a solution that is less ad hoc, percolation of non-head features in addition to the more standard percolation of head features could allow the
146 Contemporary Morphological Theories
derived stem to bear the right mix of features: the functional equivalent of the [+PV] diacritic, but based on the concrete details of the specific construction. Lieber’s (1981: 49–50) example of category-preserving prefixation of counter-in English on N, V, and A bases gives rise to Feature Percolation Convention III: ‘If a branching node fails to obtain features by Convention II (i.e. RHR-style category determination by suffix), features from the next lowest labelled node are automatically percolated up to the unlabelled branching node’ (50). Since preverbs do not convert their verb bases into prepositions (or anything else), they are to be assumed to be inert for percolation purposes. In concluding the section on percolation, Lieber (1981: 61) revises the Adjacency Condition of Allen (1978: 49) in terms of subcategorisation frames and intervening brackets: namely, no more than one bracket between an affix and the conditions necessary for its insertion. Thus, if the morpheme -ya needs the preverb to ‘potentiate’ its insertion, it cannot be done. In Lieber (1992: 91–3), enough examples of category-preserving derivational suffixes had been brought forward that the innovation of Backup Percolation from a non-head was proposed in order to fill in values for features left unspecified after Head Percolation. Based on the example of Russian babushka ‘grandmother (diminutive)’, the intention is to allow the left-hand stem baba-to assign values for Case, Person, and Gender features when the under-specified diminutive suffix -ushka is unable to assign anything beyond the categorial signature ‘N’ to the derived word. Even with this added option in hand, what ‘feature’ specification does the preverb have that the PV derivative needs? None, unless it is the elusive [+PV]. Backup percolation does not achieve the desired result. 3.3.16 What is at stake here? In examining the Sanskrit PV derivatives and their exceptional gerunds, much has emerged on the issues of adjacency in structures, visibility of earlier operations or levels, and the technique known as bracket erasure, which may serve to clean up representations and thereby tighten potential conditions on rule application. The conditioning environment for the insertion of -ya as an exponent of [gerund] is apparently a peculiar one on purely structural grounds – so much so that some frameworks are not prepared to admit it, let alone describe it. If the conditioning environment is morphotactic in nature, then it constitutes sensitivity to an abstract construction: any preverb is both necessary and sufficient, in the context of [gerund] of course, for -ya to be inserted two linear positions away from said preverb. That is not a canonical distribution by any means. A- Morphous Morphology and PFM both allude to the Split Morphology Hypothesis here, suggesting that although defining the line between inflection and derivation can be difficult, and perhaps language-specific, there remain both functional distinctions and distributional patterns that lead to a segregation of the two types. On the other hand, there are frameworks that would want to assimilate morphological operations into one component, as it were, and others that would distribute it far more widely than into two categories.
Time for a test drive 147
The significance of what may seem a single hiccup in the otherwise smooth inheritance scheme from base V to derived PV is remarkable in view of the creative responses that the -tvā/-ya complementarity draws out of each framework in turn. The question is whether this apparently rare and localised puzzle calls for no more than a narrowly defined ‘patch’, or whether the fact that the unexpected gerund formation in -ya is categorical within the open class of PV derivatives in Sanskrit demands a more principled theoretical account than an ad hoc escape hatch can supply. QUESTIONS FOR CONSIDERATION 1. Now is a good moment to recall the elephant parable. Where were the moments when the various theoretical frameworks were most diverse in what they were seeking to accomplish in their analyses? How would each define a ‘successful analysis’, and by which criteria? Were they looking at, and looking for, the same thing at some discernible level? What might that be? 2. Where do the boundaries lie between morphology and so-called neighbouring components? Is such compartmentalising helpful, constructive, and perhaps inevitable? Is it distracting, confusing, and divisive? How has your linguistic worldview shaped your reactions to the respective treatments of the three phenomena in this chapter? 3. Some call descriptive frameworks ‘lenses’, while others, ‘blinders’. What is your impression to this point? Do you have a metaphor for such frameworks that you find more compelling? NOTES 1. See Green (2007) for the obstacles that prevent a plausible synchronic phonological analysis of Celtic initial mutations. 2. Because of the multifunctionality of mutation and i-Ablaut, the entries contain purely formal second members, Nmut and Nablaut. Categorial Morphology would typically give more content-specific second members, such as N[+Nom] or the like, and so these lexical entry formulations are rather more like schemas, containing a variable as the second member, thereby abbreviating (part or all of) several distinct morphological rules. The operations fmut and fablaut, however, are defined over strings, and so they are phrased appropriately without reference to input and output categories. These affixes and operations may be applied singly or jointly to bases, according to the rules of Categorial Grammar. 3. More data would make clear the lack of motivation for a separate synchronic ablauting rule. 4. In more recent DM, features which have a bearing on morphological realisation, but which are ‘syntactico-semantically unmotivated’, that is, which are ‘never essential to semantic interpretation, since the derivation diverges onto PF and LF branches prior to the insertion of these morphemes’, are referred to as ‘ornamental’ morphemes, inserted only at PF (Embick and Noyer 2007: 305). Counted among these are agreement features such as Case, although the status of Number may be more than ornamental (307–8). This still places all nodes and morphemes before Vocabulary insertion, and consequently
148 Contemporary Morphological Theories before the application of readjustment rules, if any. This move does seem to depend on morphological case being reliably determined by syntactic configuration, but that issue remains an open one. 5. Cf. Pyatt (1997) for an extended DM analysis of initial mutation across the Celtic languages. 6. More data would confirm this assumption. 7. It should be noted that, in the context of the rest of Scottish Gaelic inflection, stem uniformity makes this word quite irregular. An interrogation of the often-encountered conflation of the concepts ‘regularity’ and/or ‘productivity’ and/or ‘naturalness’ with ‘affixation’ is long overdue. Regularity, for example, is in principle language-and system-dependent. 8. The imperviousness of the initial cluster to alternation in lenition contexts and the fact that the root already ends in a slender-compatible VC sequence (and thus would vacuously palatalise at the most) means that the morphological ‘irregularity’ of sgoil is more apparent than actual. In addition, borrowed words in Gaelic are readily nativised and do, as a rule, participate in any initial mutation patterns, so the hypothesis of loanword exemption would ultimately have to be discarded. 9. See Stump (1993a) for a discussion of criteria related to the decision ‘to refer or not to refer’. 10. The predictive relations are stated over forms, not lexical entries. The features realised by the forms related in this way are not an obstacle to predictions in WP. 11. See LM&P’s response (§3.1.7), however, for discouraging counter-evidence to this hypothesis from data beyond the set specifically under discussion in §3.1. 12. Cf. Singer’s (1999) dissertation, which relies on Optimality Theory constraints on top of Autolexical theory in the analysis of what seems like every morphological phenomenon in Georgian except the present question. 13. The template slots are traditionally referred to as PRON1 and PRON2, respectively, although as it turns out, they do not necessarily realise subject and object in that order. In the construction schemas in this section, these slots are represented by the variables a and c, respectively. 14. Note that the case cannot be made for so-called circumfix status, since both v-. . .-t and g-. . .-t consist of morphs that appear individually in the realisation of related properties. Their co-operation would seem to be in principle one of compositionality rather than extended exponence. 15. Both Blevins (2006: 561, fn. 24) and Gurevich (2006: 81–2, 109–10) remark on how relatively well-behaved the Georgian agreement system is in comparison to the baroque patterns that the rest of the language’s morphological system presents – patterns that would be especially resistant, both authors claim, to description via approaches that do not focus on whole words and paradigm structure. The implication would seem to be that the tail is wagging the dog, as it were, in that so much attention has been placed (without clear resolution) on what is potentially a minor feature of a remarkably complex system. Alternatively, this might be seen as an invigorating challenge to morphologists generally, that we extend our range of vision and our ambition. 16. A language that does not conform to even a broadly attested tendency, however, is not rendered invalid by that fact. The -s affix specifically on English 3rd-person (unmarked) singular (unmarked) present tense (unmarked) indicative (unmarked) verbs would be a source of great consternation if it were. 17. The semantics of these prefixed elements correspond to locational, directional, or positional designations realised as prepositions in many other languages. The discourse surrounding the status of these elements, both when functioning as preverbs (Sanskrit upasarga) and in their respective syntactically free versions (karmapravacanīya), is decidedly muddled. This can be seen when MacDonnell ([1927] 1968) states – with no appar-
Time for a test drive 149 ent expectation of confusion on the part of his readers – that the prepositions of Sanskrit ‘are nearly all postpositions’ (144). In contrast, although Whitney ([1889] 1960) argues that Sanskrit has ‘no proper class of prepositions . . . , no body of words having for their primary office the “government” of nouns’ (414), he nevertheless discusses their preverb behaviour and status under the heading ‘Composition with Prepositional Prefixes’, there characterising them as ‘certain words of direction, elements of adverbial character . . . , the so-called prepositions (according to the original use of that term), or the verbal prefixes’ (395). Whitney attempts to clarify that these (preverb) elements are ‘properly adverbs, having a special office and mode of use in connection with verbal roots and their more immediate derivatives’ (410). Deshpande’s Primer (2003), on the other hand, disagrees with Whitney on the issue of noun-case government for karmapravacanīya functions, and furthermore directly lists the elements in question thus: ‘Sanskrit grammarians enumerate the following prepositions which are prefixed to verbs . . .’ (121). The varying descriptions can be reconciled if it is considered that (1) MacDonnell is alluding to a diachronic development from Vedic prepositions to Sanskrit postpositions, (2) Whitney is privileging the free versions over the bound versions on etymological grounds, while troubling the form/function discrepancies associated with specialist versus lay uses of ‘preposition’, but without displacing the term he finds doubly inadequate, and (3) Deshpande is claiming traditional authority, and at the same time he is assigning a pedagogically accessible category to them. A synchronic analysis in whichever framework one chooses would do well first to break free, at least provisionally, of traditional and teaching-centred nomenclatures (cf. Haspelmath 2010, 2011). 18. See Mayrhofer ([1964] 1972: 103–4), Deshpande (2003: 121–5), or Whitney ([1889] 1960: 395–400) for lists of the preverbs most frequently compounded with verbs. See Stump (1993c: 13–17) as well, for further data and discussion. 19. The -tya allomorph of -ya appears when the verb stem contains a short vowel, adding prosodic weight to that preceding light root-syllable in the process. This -tya is not to be derived from the general gerund-forming suffix -tvā, neither phonologically nor etymologically. 20. Booij (2005: 13) describes a prosodically conditioned [pl.] allomorphy in Dutch between -en and -s (not a simple ‘reduction’ to -n) that is sensitive to an output condition on plural nouns that they should end in a trochee (metrically [ʹ ˘]): -en adds an unstressed syllable to a stress-final stem (creating a trochee), while non-syllabic -s is added to unstressed final syllables (preserving an existing trochee). For a contrasting take on the same distribution, see Harley and Noyer ([1999] 2003: 474). See also Carstairs- McCarthy (1988) in connection with issues of phonologically conditioned (suppletive) allomorphy more broadly. 21. Recall Whitney’s ([1889] 1960: 397) description mentioned in §3.3. 22. Auxiliaries are not lexical items in LMBM, but rather they are realised via the MS component. 23. This directly capitalises on the assumption of the Separation Hypothesis, which keeps form and content fundamentally distinct. 24. See, however, Clahsen’s (1999) target article using MM for the discussion of access and processing of regular and irregular past tense forms in German, and in particular the several essays from respondents published in that same volume. 25. MM rejects arbitrary class features altogether, assuming that the same patterning can be cast with reference to ‘inherent (grammatical or phonological) properties of the items involved’ (Wunderlich 1996: 94). This strong position is ruggedly optimistic, and its full implications for the traditional ten present-stem classes of Sanskrit verbs, mixed and matched in only partially predictable ways with the seven aorist-stem classes, await another forum. 26. These weak stems are morphomic in nature: that is, they are distinguished by the forms
150 Contemporary Morphological Theories that they are used to create, not by what (if anything) they might ‘mean’ in themselves, as it were (Aronoff 1994). In fact, it is potentially counter-productive to name a stem after even a characteristic use (e.g. the ‘present’ and ‘aorist’ stem systems) if the stems are in fact used in the formation of mixed groups of properties. It is less confounded at a practical level to coin language-specific or otherwise conventional names for stems (like ‘weak’ and ‘strong’, suitably defined, in the case in point). 27. Since vowel length is contrastive in Sanskrit, the several morphs in -yā ought not to be lumped in nonchalantly with the members of the -(a)ya group. 28. The stem-final alternation is actually predictable under the aforementioned sandhi system of sound-structural rules in Sanskrit (Gonda 1966: 12), but for simpler exposition in the present context, the alternation is directly stipulated in the lexical entry as if it were idiosyncratic. The muk-stem is used with the gerund in -tvā, but in the derived PV, the corresponding vimuk-stem duly inherited from the base nevertheless has a use, just not for the gerund in -ya. Instead, the PV uses the vimuk-stem in the derivation of the past active and passive participles. 29. The formal characterisation of stems in GPFM may also ultimately be amenable to a [±accent] representation, but although the phon attribute of each stem is given, the framework foregrounds stem indices (0–n), with no obvious way to refer to aspects of phonological composition. A morphophonological metageneralisation (Stump 2001: 47–50) might prove useful in addressing the pattern at the stem level as a possible future investigation. 30. An alternative analysis could try to make -tya the underlying shape of the suffix, but with the initial / t / regularly deleted after segments other than short vowels. This is clearly not tenable, however, on the broader phonological evidence, e.g. √muc-+ -tvā → muktvā ‘having released’ (*mukvā, *mucvā). 31. Derived from the saptapadi, or ‘seven steps’, part of a Hindu marriage ritual during which the couple walk together in stages around a fire.
Chapter 4
Broadening the discussion
4.0 T H E S E A R C H F O R D E F I N I T I O N The multiple areas of convergence that have emerged among the theoretical frameworks may be a consequence of their common object of description: namely. the internal structure and horizontal relationships of the words of natural language. One might expect, then, that disagreements among theoretical frameworks would not be substantial, but rather would amount to so much detritus, random by-products of the hard work of investigation, deposited as a matter of course. The degree to which there fails to be agreement on what ‘counts as’ morphology in the first place, whether morphology is potentially autonomous or is reducible to the functioning of other structural concerns (suitably enriched), or whether word structure is better assessed as built up or factored out, makes clear that simple theoretical equanimity is unlikely to prevail as even a social strategy among morphologists, let alone as a source of new linguistic knowledge. The ‘meta-metalanguage’ of morphology is harried by homonymy, pestered by polysemy, and side-tracked by synonymy.1 The terminological battles are indices of the deeper distinctions that the continua are intended to bring out, since they touch on defining assumptions. The arc of this book has developed from a preliminary consideration of morphology as an object of study to a ‘tour’ of sorts, introducing a number of frameworks that have made forays into more detailed description of how one might look into, across, and/or out from the words of natural language. In the preceding chapter, many of the varied concepts and methods for morphological analysis were demonstrated with controlled data sets (inasmuch as this is possible in the face of such diverse assumptions), in order to bring out the sorts of analyses that each new viewpoint predicts, or at least would seem to suggest. This chapter selects and unpacks just a few domains of interest to which morphological research has contributed, topics which will certainly continue both to drive theory-building, and to inspire ever more effective and reliable applications in computation and education, among other fields.
152 Contemporary Morphological Theories
4.1 T Y P O L O G Y When discussing the different criteria that can be used to define ‘types’ of languages, comparing languages specifically with regard to structure, rather than genetics, there is a small selection of often-repeated illustrations that comes from m orphology. The primary dimension of distinction is defined between analytic expression and synthetic expression of grammatical information. Analytic expression ideally matches single elements of content with single elements of form, each syntactically distinct from the next – in other words, no morphology to speak of. Synthetic expression represents the inverse tendency, that of compilation and combination of multiple units of content into single units of form. In general terms, this contrast can be made through the contrast of periphrastic versus inflectional expression of a category, e.g. the respective formations of French passé composé (analytic: ils ont mangé ‘they ate’, lit. ‘they have eaten’) as opposed to the passé simple (synthetic: ils mangèrent ‘they ate’). Regardless of distributional distinctions according to formality of register, differences in frequency, or special connotations attending one or the other of these expressions of ‘past-ness’, in the case of the former two formal units are used, whereas in the latter, just one. A comparable example in a different domain could be the use of case marking (synthetic) as opposed to marking via adposition (analytic) to indicate location, direction, or more abstract grammatical functions.2 A secondary dimension presupposes some degree of synthesis and has to do with the complexity of content expressed in individual formal pieces, or morphs. At one extreme is the biunique classical morpheme, such that morphs and content-units match up one to one, with the result that each piece of meaning may be segmented out as a concrete subunit of the whole word. The counter-posed condition is a deviation from one to one, specifically in the direction of single morphs corresponding to two or more units of content, e.g. the Scottish Gaelic verb suffix -(e)amaid which simultaneously marks 1st person, plural number, and conditional mood, but which cannot be segmented synchronically into meaningful portions (compare -(a)inn, which marks 1st person, singular number, conditional mood). This inextricable combination is traditionally referred to as fusion (cf. the different use of this term in DM §2.5). The most widely circulated morphological typology traces to Humboldt (1836, cited in Comrie 1989: 55), proposing a four-part categorisation of languages (not merely processes) emerging from the confluence of the dimensions outlined above. The types outlined are: 1. isolating – little to no use of morphology, 2. agglutinating – use of morphology with (optimally) biunique morphemes, 3. fusional – noticeable reliance on markers that express multiple features at once, and 4. polysynthetic – formation of structurally complex words with full propositional significance. These types are fairly straightforward at one level, in that each can be exemplified by one or more languages (traditionally, Classical Chinese and Vietnamese; Turkish
Broadening the discussion 153
and Hungarian; Latin, Greek, and Russian; and West Greenlandic and Chukchi, respectively). Comrie (1989) notes that polysynthesis can be recast as more elaborate agglutination, and thus as the ‘odd man out’ in an otherwise coherent set (45). Matthews (1991: 21) does not include polysynthesis in the scheme and characterises the remaining three as ‘a typology of extremes’ that leaves many languages as poor fits, mixed types, and even showing distinct types in different areas of their grammar (e.g. simple noun structure, but complex verb inflection). The broad dissatisfaction with, and primarily historiographical significance of, the traditional typological designations (Comrie 1989: 52) may be attributed to a perception that, based on the lack of predictions that these types make regarding the rest of the grammar, ‘nothing of any interest follows from classifying languages in this way’ (Spencer 1991: 38). In the opinion of Anderson (1992: 322), the said typology seems more a taxonomy of convenient labels than a system of categories determined by the intersections of logically possible alternatives, a more usual definition of a typology per se. Bloomfield (1933: 207–8) believed that the enormous variety of morphological systems across languages was ultimately beyond the ability of any ‘simple scheme’ to classify. Sapir (1921) likewise saw all preceding efforts to arrive at a morphological typology as unsatisfactory: not respectful of particular languages3 but rather ‘prone to force them down into narrow, straight-backed seats’ (122). Sapir’s own alternative proposal seeks to distinguish among parameters of degree of synthesis, formal technique of combination, and broad tendencies in expressing simple or complex semantic and/or relational concepts (101–2, 142–3).4 In attempting his more articulated typology, Sapir cautions readers concerning leakage between types, mixed patterning within individual languages, identifying diachronic residue as opposed to active morphological patterns, the likelihood or even inevitability of diachronic change in type, and the potential for cultural bias in assessing grammatical and/or conceptual complexity within any scheme that has putatively universal implications. In contrast with the frustrations that have emerged with the sometimes ad hoc categories of traditional morphological typology, as well as with ‘differences in terminology and underlying logic’ (‘Canonical Typology’; cf. Haspelmath 2007, 2010), the canonical approach to typology (Corbett 2009, 2010) has developed in recent years, seeking to reground typological discussions of morphological systems in terms of the logical possibilities provided by the elements, relations, and patterns that are posited in linguistic descriptions. Baerman and Corbett (2007) note the subordinate status accorded to morphology in typological research, where it is found ‘serving as a diagnostic for the presence of morphosyntactic features or as the context for phonological operations’ (115). In lieu of acquiescing to this supporting role, the canonical approach adopts a method whereby, having established the needed features and permissible values for each feature with respect to a given language, these combinations are ‘multipl[ied] out’ to define the paradigmatic space. This space is evaluated in the light of the variety of lexemes that may populate the cells of the paradigm. These initial steps would seem to be standard for any word and paradigm (WP) approach, but this paradigm structure itself, especially as it relates
154 Contemporary Morphological Theories
to the constitution and distribution of distinct inflectional classes, is assessed with respect to two broad, defining principles for inflection, namely: 1. distinctiveness: ‘Canonical inflection classes are fully comparable and are distinguished as clearly as possible’ (Corbett 2009: 3; cf. biuniqueness and other ‘natural’ qualities in NM §2.9). 2. independence: ‘The distribution of lexical items over canonical inflectional classes is synchronically unmotivated’ (Corbett 2009: 5). With regard to the latter principle, an inflection class for which the constituent lexemes are predictable on independent grounds (phonological, semantic, etc.) is non-canonical. Not defective, not dysfunctional, certainly not impossible: simply not the ideal of the type. Languages whose inflectional systems show deviations from these patterns, large or small, can nevertheless be situated and described in a rigorous manner within the typological space. In the area of derivational morphology, there are again two principles of canonical behaviour, but they are not simply rewordings of those proposed for inflection. Derivational formations are not obligatory in the same way that inflectional morphology is, and thus particular derived forms appear in a much more ‘patchy’ manner, subject to extra-linguistic factors such as perceived utility of the potential word, based on the subjective ‘name-worthiness’ of the concept to be expressed (Corbett 2010: 146). Also exerting a more general pressure on a possible new derivation are already existing or otherwise competing formations for the same meaning (blocking). The principles proposed to characterise canonical derivation, each making crucial reference to constituent elements for comparison, are the following: 1. ‘Canonical derived words have clear indicators’ that permit their synchronic formation and parsing: a. many-to-many substitutability of both base and derivational marker; b. transparent semantics with respect to the semantics of the base and any predicate added in the derivation; and c. transparent form, such that the source material is readily recoverable on inspection of the derivative. 2. ‘Canonical derived words are fully distinct from their base’: a. bearing its own lexical index and projecting a distinct paradigm (where applicable); and b. differing from its base in a systematic way through the addition of a semantic predicate.5 In this way, by positing larger principles that define canonical states, and in the wider literature articulating these principles through more concrete criteria that pertain to each one, the canonical typology constitutes a coherent construct, but it does not correspond to what is cross-linguistically frequent,6 nor is it counter- exemplified when theoretically possible types are not found among actual languages. The typology lays out the space that the descriptive framework implicates as potentially ‘inhabitable’, and then attested patterns are ‘calibrated’ with respect to the
Broadening the discussion 155
canonical point fixed by the criteria (Corbett 2010: 141). Those patterns that best match the canonical qualities of the system are correspondingly the least controversial instances of such a system. What wards off circularity here is simply the fact that the ideal state as defined by the system may not in fact have any perfect exemplars; this is what distinguishes the canonical approach from the relativity sometimes encountered in prototype- based accounts. Corbett invokes the cardinal vowel system of Daniel Jones as sharing a similar goal, but in the physical rather than grammatical domain: the cardinal vowels map out extreme extensions away from a neutral mid-point, and thus describe the vowel space. Whether a given language has any such extreme vowels in its inventory is immaterial to the cardinal vowel system’s validity as ‘anchoring point for the vowel space’ (Corbett 2010: 142). To the degree that paradigm spaces are not physical entities to be measured, but rather a product of a linguist’s theoretical description, the analogy requires further justification. The outside-in orientation of the canonical approach has given rise to what might sound like a mission statement for rebooting the morphological typology enterprise: ‘[Offering] a clearer picture of what there is, within the space of possibilities of what there might be’ (Corbett 2009: 10). 4.2 P R O D U C T I V I T Y In a descriptive context, there can be a tendency to shy away from questions concerning the ‘why’ behind particular phenomena. Assuming a fundamental element of arbitrariness underlying the associations of form and meaning into linguistic signs, then given the potential stock of material that a language has at its disposal from which to make new words, there are multiple levels of differential prosperity in the distribution of morphological resources that one might wish to explain. The question specifically is one of relative productivity with respect to the extension of existing patterns to new contexts (analogy; see Blevins and Blevins 2009), global and case-specific preference patterns in the face of two or more (nearly) synonymous expressions (blocking; see Aronoff 1976), and the sources of apparent restrictions on combinations that, all else being equal, should be viable but yet are not attested and/or are judged to be unacceptable by speakers (etymological, register-based, or stipulated lexical strata).7 Productivity is cast by Aronoff (1976) as ‘one of the central mysteries of derivational morphology . . . though many things are possible in morphology, some are more possible than others’ (34). If there is a democratic element to rises and falls in morphological deployment, where might one take a poll? Before attempting explanations of patterns and/or trends, it is of primary importance to collect accurate data for the language(s) in question. This has traditionally been accomplished with more or less painstaking dictionary work, but of course this is not without its drawbacks. In their intended capacities as comprehensive, compiled documents, dictionaries are by their nature historical records of language and so do not reflect the use of the moment. Dictionaries are also subject to a certain degree of editorial telescoping, especially in the case
156 Contemporary Morphological Theories
of truly productive patterns. For example, English affixes such as un-and -ness are each notably frequent in terms of their deployment in derived lexemes, and each is freely available for the creation of untold further items. In the New Oxford American Dictionary (McKean 2005: 1825), distinct entries are given for un-: the first for use with (1) ‘adjectives, participles, and their derivatives’ (unacademic, unrepeatable), and (2) nouns (unrest, untruth), while the second is specialised for use with verbs, but in distinct senses: namely, (1) ‘reversal or cancellation’ (untie, unsettle) and (2a) ‘deprivation, separation, reduction to a lesser state’ (unmask, unman) or (b) ‘release’ (unburden, unhand). Not surprisingly, given these semantic and base-category ranges, this particular dictionary continues for twenty-four three-column, fine-print pages of words beginning with orthographic (1825–48), including two half-pages of words listed with only syntactic category information – that is, without definition – as an indication of just how potent un-prefixation is (1826–7). Even once one has subtracted the interloping lexemes that coincidentally begin with but not a un-prefix (understudy, undulate, UNESCO, unicorn), the type frequency is evident and the presumption of productivity follows unbidden. Depending on the language’s morphological tendencies (see §4.1), the frequencies generated at the morphological pattern level may be interpreted as morpheme- frequencies. Among the literature on productivity, there are pockets of discussion working on exactly this level, identifying more or less productive affixes and considering them in terms of the apparent limits on their respective bases (Fabb 1988b; Plag 1996) and accounting for the attested combinations and sequences (Aronoff and Fuhrhop 2002).8 Plag (2006) offers a compact yet critical introduction to methodological issues in the study of productivity, distinguishing (after Bauer 2001) qualitative as opposed to quantitative approaches. Bauer’s (2001) work can be placed more solidly in the qualitative camp, as a consequence of a stated preference for seeing morphological processes in regard to their availability for use in producing new words (a yes/no question and a property of the rule or affix itself). As a starting point, Bauer (2001: 75–84) mistrusts the basic unconstrainedness of analogy as an alternative to a more determinate rule-governed model. He grudgingly acknowledges, however, the need for a more system-driven role of analogy in the formation of new words, but this is cast as a back-up mechanism to account for the unavoidable sporadic examples of ‘unruly’ pattern extension, such as the common paradigmatic relation -ist ↔ -ism in English (see §2.4) which can link personal and abstract noun-pairs whether the relation of the derivatives to the common base is synchronically recognisable (social-), somewhat more opaque in sense (Method-), or not even an independent base at all (bapt-). Bauer also notes unexpected trends of minor pattern expansion (as opposed to the more usual regularisation, or assimilation into a dominant pattern) as well as exaptation, that is, a semantic or functional ‘reanimation’ of previously moribund or even ‘dead’ morphological patterns (Lass 1990). In order to keep at bay apparent counter-examples to rule- government from the areas of playful or otherwise disposable (nonce) formations,
Broadening the discussion 157
Bauer builds a suitable rider into his definition of the productivity of a process, referencing ‘its potential for repetitive non- creative morphological coining’ (2001: 98). Bauer (2001: 60) proposes a unidirectional flow chart of influence among different linguistic and extra-linguistic sources that can have an impact on the demonstrable productivity of a particular morphological process (or process type) in any given language. rule-governedness
transparency constructional iconicity
compositionality
fit universal preferences
naturalness base frequency productivity
majority process or typological consistency
type frequency
Figure 4.1 Flow of influence in productivity. (Source: Bauer 2001: 60, © Cambridge University Press)
It remains clear here that, from Bauer’s perspective, rule-governedness is not just important; it is undominated in the hierarchy. A de facto conspiracy of naturalness factors (see §2.9) gives rise to productivity in concert with language- specific patterns from morphological type and the available number of acceptable bases, resulting in type frequency as the observable output and index of productivity. Barðdal (2008) prefaces a detailed constructional analysis of Icelandic morphosyntax with a concept map (21) assessing the ‘highly structured and systematic metaconcept of productivity’ (53). An intuitive sense of a process’s status as ‘living’ or ‘dead’ can summarise the network at a metaphorical level, but internal to the metaconcept itself, three central subconcepts underlie the discourse of productivity: namely, generality, regularity, and extensibility (22).
158 Contemporary Morphological Theories Generality co-occurrence
entailment
co-occurrence Regularity
Extensibility
Figure 4.2 The three subconcepts of productivity. (Source: Barðdal 2008: 23, © John Benjamins Publishing Co.)
General processes are schematically open, widespread in a language, and although they may or may not be regular, they are necessarily extensible in principle. Regular processes are rule-based by definition, and they tend to be combinable, transparent, and/or compositional, but they need not be general in distribution nor extend to new contexts. Extensible processes are flexible, synchronically meaningful (not relics), occurring with new forms, and attractive to existing words of distinct patterns. These may be regular, and they must be general. Furthermore, all three subconcepts correlate with frequency in their own right. Thus even from qualitative perspectives, productivity has an inherent tie to the systemic issue of frequency. Returning to Plag’s (2006) discussion of quantitative methods, the history of research has been one of deciding precisely what to count, and how. To be more interpretable, there is also at least the question as to when. Quantitative studies must set temporal bounds on the sampling window in decades or at most a century or two, for although an unbounded, panchronic count may well result in higher raw numbers, this boost is a false virtue because it does not model the language as any individual could have learned or used it. At the level of method, the simplest is to perform type frequency tabulations centred on particular affixes or other morphological processes, but on this standard, irregular strong past forms of verbs could seem more vital than they are. Some contrasting measure of active formation must be brought into the equation for interpretive purposes. Following Aronoff (1976), one might attempt to contrast such a basic ‘actual word’ type frequency with a hypothetical number of ‘possible words’ that could reasonably be derived with that affix or process without violating existing conditions for use. Plag (2006) points out the practical difficulty in counting potential words that not only do not exist, but also in defining a set the membership of which must be left open, ‘because new potential base words (e.g. new adjectives as bases for -ness) may enter the language at any time’ (541). One alternative while remaining in the realm of the attested is to count, within the set of existing derivatives using a target affix or process, the set of neologisms created on that pattern within a given time period. With the help of dictionaries such as the Oxford English Dictionary that track attestation dates, or alternatively with historical corpora, one can tie instantiations of ‘productivity in action’ to approximate moments of coining. Technologically, computer-searchable dictionaries facilitate
Broadening the discussion 159
searching and compilation, but of course once again there arise the difficulties of lag before recording and of editorial selection, but at least for those examples where many neologisms are recorded for a given affix at a given time, productivity is (temporarily) assured. Related to the dictionary methods, but less subject to filtering and neatening, is corpus-based research on lexical distribution and morphological choices. There is a sense in which a dictionary distils a corpus’s database of attested word-form instances (tokens) to the set of potential headwords to be listed, given some minimal threshold of documented appearance within the selected corpus in order to merit being recorded for posterity in the dictionary. Looking bottom-up at the contents of the corpus, tokens may be associated (via tagging) on morphological grounds – that is, according to which patterns are instantiated in their formation, thus abstracting higher-level types for consideration. Although productivity is most often researched in the context of derivation (and sometimes compounding), the diversity of exponence patterns present in the inflectional morphology of the language under consideration will bear on whether it could be interesting and worthwhile to separate out the distribution(s) of one or more inflected forms within the lexeme-frequencies for analysis. A concordance will typically be exhaustive in listing all word-forms in a corpus, sometimes not even abstracting over spelling variations, but somewhere between excruciating ‘splitting’ and reckless ‘lumping’ should lie a suitably sensitive criterion for a relevant and reliable corpus-based count for the purpose of assessing productivity. The context of corpus studies in productivity adds to type frequency and neologism status the dimension of token frequency. While it might seem obvious that a high token frequency is relevant, there is no standard upper threshold to ‘establish’ productivity in some absolute sense, and so, as a check on ‘bigger is better’, there has come to be a focus on the nadir of attestation, a ‘one-off’ in a given corpus, a hapax legomenon. Since its status is tied to appearance in a particular corpus, the larger the corpus, the greater the significance of the hapax legomenon for drawing inferences about the productivity of any morphologically complex pattern it instantiates. Baayen (1993) proposes a probabilistic measure of productivity based directly on the division of the total of hapax legomena sharing a morphological element by the total number of tokens in the corpus that share that element. Plag (2006: 544–6) demonstrates some of the problems that ensue from reliance on any single quantitative measure of productivity, and so suggests reference to multiple measures and advises familiarity with the methodological assumptions (and potential hazards) of each. Questioning the issue from a different perspective, Blevins (2003: 758) offers the usage-based hypothesis that preemption of possible words formed by means of synchronically productive patterns results from the inertia that established word-forms carry in the face of the effort required to ‘sell’ an innovative alternative formation, no matter how regular, logical, transparent, or recognisable that alternative may be. In other words, the same systemic ‘weight’ that keeps suppletive went in place as the past form of go in English is active to different degrees in the maintenance of all
160 Contemporary Morphological Theories
apparent blocking.9 Blevins suggests that ‘asking why a given affix does not combine with every possible root in a language is rather like asking why every well-formed monosyllabic or bisyllabic form does not have a meaning’ associated with it (758). The resolution of the productivity question seems therefore to bear on both the cognitive and the social sides of language in use, but still its exploration may shed light, at least indirectly, on the internal structure of language. 4.3 F A I R C O M P A R I S O N The preceding sections in this chapter have considered contexts in which one can potentially get down quite close to ground-level, as it were, with the data to be described. When faced with a morphological problem, there is a palpable tension between considering the data first in their own right and referring (or deferring) to the authority of an external framework to decide what is or might be going on. It is possible to look at the theories discussed here as an assortment of tools, to be tried one after another on a task in order to determine which one or ones seem to clarify things. The trouble with process of elimination (according to alphabetical order by name, as here, or some other arbitrary sequence) as a method of problem-solving is that time or circumstance does not always permit every option an equal chance to take a turn. In lieu of relying on habit or trusting to fate, then, an assessment of the details of the task in as objective a manner as possible can at least narrow the selection of descriptive framework in a relevant and effective way. Haspelmath (2010) argues in favour of framework-free theorising in the interest of language-particular description of linguistic phenomena. This position is consistent with the ethic of descriptive linguistics in the first half of the twentieth century, that each language represents a particular system of systems that meets with misanalysis when a framework grown on other languages’ facts is brought to bear (cf. Boas 1911: 18, 35–43). Though this radical language-specificity may strike some as deliberately opting to forego any chance of identifying cross-linguistic parallels and higher-level generalisations, let alone anything even vaguely universal, it is Haspelmath’s contention that it is not simply the analyst’s biases that are problematic, rather that ‘any grammatical framework is precisely such a “prejudice” that we want to avoid’ (2010: 342). From this commitment, there is no metalanguage that one may adopt, because ‘a metalanguage by definition provides a pre-established set of expressions with a certain meaning’ (345), and it remains an open question as to whether grammatical categories and relations are indeed language-specific, as has been argued variously in the contemporary typological literature, but which was anticipated far earlier by Boas and also by Whorf (1945: 10–11; emphasis in original): Finally, in a still wider sense generic categories may be so formulated as to become equivalent to the concepts of a general science of grammar. Such categories are made by grouping what seem to us to be similar specific categories in d ifferent languages. Only in such a sense can we speak of a category of
Broadening the discussion 161
‘passive voice’ which would embrace the forms called by that name in English, Latin, Aztec, and other tongues. Such categories or concepts we may call taxonomic categories, as opposed to descriptive categories. In specifying that similarity of assigned name should not license ipso facto the assumption of identity for phenomena so named, the work of the linguist becomes a much more rigorous enterprise: not only must one argue for the validity of one’s account in connection with a specific language, but furthermore one must articulate exactly the significance of the categories and relations in question for the benefit of readers, but without easy recourse to more familiar terms coined elsewhere. Haspelmath (2010: 350–62) demonstrates analytic errors in morphosyntactic description that follow from a variety of universal assumptions on the part of both formalist and functionalist frameworks, as well as in the work of linguists adopting the more ‘stripped-down’ approach called Basic Linguistic Theory, which is often employed in, for example, the writing of descriptive grammars. The systematic counter-examples presented are encountered when the theories are extended beyond languages from among the ‘usual suspects’, leading Haspelmath to the general claim that ‘the framework is always biased against those languages that have not been studied yet’ (2010: 359). The framework- free, ‘take nothing for granted’ perspective is clearly not intended – far from it, in fact – as a space for drafting in safety before turning a preliminary, one-of-a-kind description into something more broadly recognisable or even ‘mainstream’. In the same spirit, but from another angle, when one charges headlong toward unfamiliar data with framework outstretched, one can hardly help but crop the research object. 4.4 O U T R O Having persevered to this point, one might well ask what the best way forward is. Neither this book nor its author operates under the misapprehension that everyone, after touring the morphological workshop, will affix themselves to a station, tools in hand, and labour away blissfully. The purpose of this book, as stated at the outset, is to open up for inspection some of the variety available in ‘doing morphology’, conceived specifically in light of how often students, scholars, and others find themselves intrigued, baffled, or frustrated by issues of word structure, and then find that their resources for discerning what is what are all too limited. It is hoped that the descriptions, demonstrations, and reading lists provided in this book may be useful for the reader when compiling a suitable, targeted dossier and when approaching both new data and the professional literature. Words are there – define them as you may – to be taken seriously, and if a descriptive challenge exceeds the given ‘off- the-shelf’ frameworks, or if it runs beyond the edges of the known territory in some way, then interpolating, extrapolating, reverse-engineering, or otherwise ‘hacking the system’ are distinct possibilities. At the very least, at this point the chances of reinventing the wheel are reduced.
162 Contemporary Morphological Theories
QUESTIONS FOR CONSIDERATION 1. Sapir lamented in the context of typological analysis that ‘the strong craving for a simple formula [if possible, a triune formula] has been the undoing of linguists’ (1921: 122). In particular, Sapir was responding to an observed tendency whereby a scholar establishes two extreme poles and then ‘throws everything else into a “transitional type”’ (123).10 Without even leaving morphology, how many examples of this ‘irresistible method’ in action can you identify? Are any of them ‘satisfactory’ as they stand? 2. Studies of productivity have historically proceeded in a more or less piecemeal fashion, considering limited sets of derivational affixes, in the usual case. It has proven difficult to achieve consensus on a definition of productivity that offers necessary and sufficient conditions, and indeed the concept seems to be a gradient one. With this in mind, what could be meant by the evaluation of a given affix or process as ‘surprisingly productive’? What sort of affix or process might be evaluated as ‘surprisingly unproductive’? NOTES 1. Some disagreements are (perhaps ironically) terminological in nature, ranging from (1) superficial alternative labels for a shared concept to (2) subtle misapprehensions based on a common name for distinct entities, states, or processes to (3) insidious partial coincidences that sabotage civil discourse, revealing category errors at inopportune moments. Indeed, it is for this reason that the present book does not attempt a glossary – it would be as long again as the text itself. The index as it stands is based on terms only, with full acknowledgement that each headword so listed does not necessarily constitute a coherent picture, and that not every possible cross-reference has been identified. 2. Particular languages may use both patterns simultaneously in a more or less redundant fashion – the case marked on an NP may be determined by an adposition that governs it (cf. Stewart and Joseph 2009). 3. In a related vein, Sapir (1921) also cautioned (with tongue in cheek, but with examples nevertheless) against thoughtless or sentimentalist conflation of morphological fusion and cultural sophistication, or of agglutination with a more primitive sensibility (124, fn. 2; cf. Anderson 1992: 321). 4. Anderson (1992: 334) does, however, see a strong affinity between his A-Morphous Morphology (§2.1) and Sapir’s (1921) ideas for a typological framework. 5. The proposal that inflection is distinct from derivation in that the former pertains to the realisation of morphosyntactic properties and the latter to added semantic predicates is attributed in Corbett (2010) to Spencer and Luís (2013). 6. Compare once again the naturalness desiderata of NM (§2.9). Whereas naturalness in that framework is claimed to correlate directly with cross-linguistic frequency and to attract diachronic developments, canonicity starts from the assumption that canonical classes and systems are ideals, unlikely to be observed ‘in the wild’, so to speak, and furthermore likely to be disrupted through certain processes of analogical change (Corbett 2009: 8). 7. See Haspelmath and Sims (2010: 114–31) for a concise introduction to these concepts. 8. The sequencing and allomorphic patterns that may be detected in productivity research can readily serve as empirical base for research in the level-ordering, Lexical Morphology and Phonology, and Stratal OT frameworks (§2.7; cf. Aronoff and Sridhar 1983).
Broadening the discussion 163 9. This is despite the regular, parsable, potential alternative *?goed that English acquirers and learners perpetually find tempting. In this connection, compare the conserving effect in Bybee (2007: 10–11), whereby high-token-frequency irregular forms are resistant to replacement by analogical means. Should the token frequency trend downward over time, the irregular form may become ‘vulnerable’ to the attraction of a more regular alternative. 10. This cautionary comment gives a measure of comfort in the context of the present text’s five dimensions with five-point continua, but then again, how much nuance is enough before sliding into unwieldiness?
References
Ackema, Peter, and Ad Neeleman (2004), Beyond Morphology: Interface Conditions on Word Formation [Oxford Studies in Theoretical Linguistics 6], Oxford: Oxford University Press. Akinlabi, Akinbiyi (1996), ‘Featural affixation’, Journal of Linguistics, 32.2: 239–89. Allen, Margaret R. (1978), Morphological Investigations, University of Connecticut dissertation. Anderson, Stephen R. (1969), West Scandinavian Vowel Systems and the Ordering of Phonological Rules, Massachusetts Institute of Technology dissertation. Anderson, Stephen R. (1977), ‘On the formal description of inflection’, Chicago Linguistic Society, 13: 15–44. Anderson, Stephen R. (1982), ‘Where’s morphology?’, Linguistic Inquiry, 13.4: 571–612. Anderson, Stephen R. (1984), ‘On representations in morphology: Case, agreement, and inversion in Georgian’, Natural Language and Linguistic Theory, 2.2: 157–218. Anderson, Stephen R. (1986), ‘Disjunctive ordering in inflectional morphology’, Natural Language and Linguistic Theory, 4.1: 1–31. Anderson, Stephen R. (1988a), ‘Inflection’, in Michael Hammond and Michael Noonan (eds), Theoretical Morphology, San Diego: Academic Press, pp. 23–43. Anderson, Stephen R. (1988b), ‘Morphological theory’, in Frederick J. Newmeyer (ed.), Linguistics: The Cambridge Survey, Cambridge: Cambridge University Press, vol. 1, pp. 146–91. Anderson, Stephen R. (1992), A-Morphous Morphology [Cambridge Studies in Linguistics 62], Cambridge: Cambridge University Press. Archangeli, Diana (1987), ‘An overview of the theory of Lexical Phonology and Morphology’, MIT Working Papers in Linguistics, 7: 1–14. Aronoff, Mark (1976), Word Formation in Generative Grammar, Cambridge, MA: MIT Press. Aronoff, Mark (1994), Morphology by Itself, Cambridge, MA: MIT Press. Aronoff, Mark, and Nanna Fuhrhop (2002), ‘Restricting suffix combinations in German and English: Closing suffixes and the monosuffix constraint’, Natural Language and Linguistic Theory, 20.3: 451–90. Aronoff, Mark, and S. N. Sridhar (1983), ‘Morphological levels in English and Kannada or Atarizing Reagan’, in John F. Richardson, Mitchell Marks, and Amy Chukerman (eds), Papers from the CLS Parasession on the Interplay of Phonology, Morphology, and Syntax, Chicago: Chicago Linguistic Society, pp. 3–16. Baayen, Harald (1993), ‘On frequency, transparency and productivity’, in Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 1992, Dordrecht: Kluwer, pp. 181–208. Bach, Emmon (1984), ‘Some generalizations of categorial grammars’, in Fred Landman and Frank Veltman (eds), Varieties of Formal Semantics, Dordrecht: Foris, pp. 1–24.
References 165 Baerman, Matthew (2007), ‘Morphological typology of deponency’, in Matthew Baerman, Greville G. Corbett, Dunstan Brown, and Andrew Hippisley (eds), Deponency and Morphological Mismatches, Oxford: Oxford University Press, pp. 1–19. Baerman, Matthew, and Greville G. Corbett (2007), ‘Linguistic typology: Morphology’, Linguistic Typology, 11.1: 115–17. Baker, Mark (1985), ‘The Mirror Principle and Morphosyntactic Explanation’, Linguistic Inquiry, 16.3: 373–415. Barðdal, Jóhanna (2008), Productivity: Evidence from Case and Argument Structure in Icelandic, Amsterdam/Philadelphia: John Benjamins. Bauer, Laurie (2001), Morphological Productivity [Cambridge Studies in Linguistics 95], Cambridge: Cambridge University Press. Beard, Robert (1986), ‘Neurological evidence for Lexeme–Morpheme Based Morphology’, Acta Linguistica Academiae Scientiarum Hungarica, 36.1–4: 3–23. Beard, Robert (1987), ‘Morpheme order in a Lexeme/Morpheme- Based Morphology’, Lingua, 72.1: 1–44. Beard, Robert (1988), ‘On the separation of derivation from morphology: Toward a Lexeme/ Morpheme-Based Morphology’, Quaderni di Semantica, 9.1: 3–59. Beard, Robert (1995), Lexeme–Morpheme Base Morphology, Albany, NY: SUNY Press. Beard, Robert, and Mark Volpe (2005), ‘Lexeme-Morpheme Base Morphology’, in Pavol Štekauer and Rochelle Lieber (eds), Handbook of Word-Formation, Dordrecht: Springer, pp. 189–205. Bermúdez-Otero, Ricardo (1999), Constraint Interaction in Language Change: Quantity in English and Germanic, University of Manchester dissertation. Bermúdez-Otero, Ricardo (2010/2014), ‘Stratal Optimality Theory: An Overview’, (last accessed 3 December 2014). Blake, Barry J. (1990), Relational Grammar, New York: Routledge. Blevins, James P. (2001), ‘Paradigmatic derivation’, Transactions of the Philological Society, 99.2: 211–22. Blevins, James P. (2003), ‘Stems and paradigms’, Language, 79.4: 737–67. Blevins, James P. (2004), ‘Inflection classes and economy’, in Gereon Müller, Lutz Gunkel, and Gisela Zifonun (eds), Explorations in Nominal Inflection, Berlin: Mouton de Gruyter, pp. 41–85. Blevins, James P. (2005), ‘Word-based declensions in Estonian’, in Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 2005, Dordrecht: Springer, pp. 1–25. Blevins, James P. (2006), ‘Word-based morphology’, Journal of Linguistics, 42.3: 531–73. Blevins, James P., and Juliette Blevins (2009), ‘Introduction’, in James P. Blevins and Juliette Blevins (eds), Analogy in Grammar: Form and Acquisition, Oxford: Oxford University Press, pp. 1–12. Bloch, Bernard (1947), ‘English verb inflection’, Language, 23.4: 399–418. Bloomfield, Leonard (1933), Language, New York: Holt. Boas, Franz (1911), ‘Introduction’, in Franz Boas (ed.), Handbook of American Indian Languages [Bureau of American Ethnology, Bulletin 40, Part 1], Washington, DC: Government Printing Office, pp. 1–83. Bobaljik, Jonathan D. (2012), Universals in Comparative Morphology: Suppletion, Superlatives, and the Structure of Words, Cambridge, MA: MIT Press. Bonet, Eulàlia (1991), Morphology after Syntax: Pronominal Clitics in Romance, Massachusetts Institute of Technology dissertation. Booij, Geert (1993), ‘Against split morphology’, in Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 1993, Dordrecht: Kluwer, pp. 27–49. Booij, Geert (2005), ‘Construction-dependent morphology’, Lingue e Linguaggio, 4.2: 1–16.
166 Contemporary Morphological Theories Booij, Geert (2007), ‘Construction Morphology and the lexicon’, in Fabio Montermini, Gilles Boyé, and Nabil Hathout (eds), Selected Proceedings of the 5th Décembrettes: Morphology in Toulouse, Somerville, MA: Cascadilla Proceedings Project, pp. 34–44. Booij, Geert (2010), Construction Morphology, Oxford: Oxford University Press. Booij, Geert (2013), ‘Morphology in Construction Grammar’, in Thomas Hoffmann and Graeme Trousdale (eds), The Oxford Handbook of Construction Grammar, Oxford/New York: Oxford University Press, pp. 255–73. Booij, Geert, and Jerzy Rubach (1985), ‘Postcyclic versus postlexical rules in Lexical Phonology’, Vrije Universiteit [Amsterdam] Working Papers in Linguistics 14. Borsley, Robert D., Maggie Tallerman, and David Willis (2007), The Syntax of Welsh, Cambridge: Cambridge University Press. Brown, Dunstan (1998), ‘Stem indexing and morphonological selection in the Russian verb: A Network Morphology account’, in Ray Fabri, Albert Ortmann, and Teresa Parodi (eds), Models of Inflection [Linguistische Arbeiten 388], Tübingen: Max Niemeyer, pp. 196–224. Brown, Dunstan, Greville Corbett, Norman Fraser, Andrew Hippisley, and Alan Timberlake (1996), ‘Russian noun stress and Network Morphology’, Linguistics, 34.1: 53–107. Brown, Dunstan, and Andrew Hippisley (2012), Network Morphology: A Defaults-based Theory of Word Structure, New York: Cambridge University Press. Burrow, Thomas (1955), The Sanskrit Language, London: Faber & Faber. Bybee, Joan L. (1985), Morphology: A Study of the Relation between Meaning and Form [Typological Studies in Language 9], Amsterdam/Philadelphia: John Benjamins. Bybee, Joan L. (1988), ‘Morphology as lexical organization’, in Michael Hammond and Michael Noonan (eds), Theoretical Morphology, San Diego: Academic Press, pp. 119–41. Bybee, Joan L. (1995), ‘Regular morphology and the lexicon’, Language and Cognitive Processes, 10.5: 425–55. Bybee, Joan L. (2007), Frequency of Use and the Organization of Language, New York: Oxford University Press. Bybee, Joan L., and Carol Lynn Moder (1983), ‘Morphological classes as natural categories’, Language, 59.2: 251–70. ‘Canonical Typology’, (last accessed 24 December 2014). Carstairs, Andrew (1983), ‘Paradigm economy’, Journal of Linguistics, 19.1: 115–28. Carstairs, Andrew (1987), Allomorphy in Inflexion, London: Croom Helm. Carstairs-McCarthy, Andrew (1988), ‘Some implications of phonologically conditioned suppletion’, in Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 1988, Dordrecht: Kluwer, pp. 67–94. Carstairs-McCarthy, Andrew (1993), ‘Morphology without word-internal constituents: A review of Stephen R. Anderson’s A-Morphous Morphology’, in Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 1992, Dordrecht: Kluwer, pp. 209–33. Cherchi, Marcello (1999), Georgian [Languages of the World: Materials 147], Munich: Lincom-Europa. Chomsky, Noam (1970), ‘Remarks on nominalization’, in R. Jacobs and P. Rosenbaum (eds), Readings in Transformational Grammar, Waltham, MA: Ginn, pp. 184–221. Chomsky, Noam (1981), Lectures on Government and Binding, Dordrecht: Foris. Chomsky, Noam (1993), ‘A minimalist program for linguistic theory’, in Kenneth Hale and Samuel Jay Keyser (eds), The View from Building 20: Linguistics Essays in Honor of Sylvain Bromberger, Cambridge, MA: MIT Press, pp. 1–52. Chomsky, Noam, and Morris Halle (1968), The Sound Pattern of English, New York: Harper & Row. Clahsen, Harald (1999), ‘Lexical entries and rules of language: A multidisciplinary study of German inflection’, Behavioral and Brain Research, 22.6: 991–1013; discussion 1014–60.
References 167 Comrie, Bernard (1989), Language Universals and Linguistic Typology: Syntax and Morphology, 2nd edn, Chicago: University of Chicago Press. Corbett, Greville G. (1982), ‘Gender in Russian: An account of gender specification and its relationship to declension’, Russian Linguistics, 6.2: 197–232. Corbett, Greville G. (2009), ‘Canonical inflectional classes’, in Fabio Montermini, Gilles Boyé, and Jesse Tseng (eds), Selected Proceedings of the 6th Décembrettes, Somerville, MA: Cascadilla Proceedings Project, pp. 1–11. Corbett, Greville G. (2010), ‘Canonical derivational morphology’, Word Structure, 3.2: 141–55. Corbett, Greville G., and Norman M. Fraser (1993), ‘Network Morphology: A DATR account of Russian nominal inflection’, Journal of Linguistics, 29: 113–42. Croft, William (2002), Radical Construction Grammar: Syntactic Theory in Typological Perspective, Oxford: Oxford University Press. Croft, William, and D. Alan Cruse (2004), Cognitive Linguistics, Cambridge: Cambridge University Press. Culicover, Peter W., and Ray Jackendoff (2005), Simpler Syntax, Oxford: Oxford University Press. Deshpande, Madhav M. (2003), A Sanskrit Primer [Michigan Papers on South and Southeast Asian Studies 47], Ann Arbor: University of Michigan Press. Di Sciullo, Anna Maria, and Edwin Williams (1987), On the Definition of Word [Linguistic Inquiry Monograph 14], Cambridge, MA: MIT Press. ‘Distributed Morphology: Frequently Asked Questions List’, (last accessed 24 December 2014). Dixon, R. M. W., and Alexandra Y. Aikhenvald (2002), ‘Word: A typological framework’, in R. M. W. Dixon and Alexandra Y. Aikhenvald (eds), Word: A Cross-linguistic Typology, Cambridge: Cambridge University Press, pp. 1–41. Donegan, Patricia, and David Stampe (1977), ‘On the description of phonological hierarchies’, Chicago Linguistic Society, 13: 35–38. Donegan, Patricia J., and David Stampe (1979), ‘The study of Natural Phonology’, in Daniel A. Dinnsen (ed.), Current Approaches to Phonological Theory, Bloomington: Indiana University Press, pp. 126–73. Dressler, Wolfgang U. (1985a), ‘Typological aspects of natural morphology’, Acta Linguistica Academiae Scientiarum Hungarica, 35.1–2: 51–70. Dressler, Wolfgang U. (1985b), Morphonology: The Dynamics of Derivation [Linguistica Extranea, Studia 12], Ann Arbor: Karoma. Dressler, Wolfgang U. (1985c), ‘On the predictiveness of Natural Morphology’, Journal of Linguistics, 21.2: 321–37. Dressler, Wolfgang U. (2005), ‘Word- formation in Natural Morphology’, in Pavol Štekauer and Rochelle Lieber (eds), Handbook of Word-Formation, Dordrecht: Springer, pp. 267–84. Dressler, Wolfgang U., Willi Mayerthaler, Oswald Panagl, and Wolfgang U. Wurzel (eds) (1987), Leitmotifs in Natural Morphology [Studies in Language Companion Series 10], Amsterdam/Philadelphia: John Benjamins. Embick, David, and Rolf Noyer (2007), ‘Distributed Morphology and the syntax- morphology interface’, in Gillian Ramchand and Charles Reiss (eds), The Oxford Handbook of Linguistic Interfaces, Oxford/New York: Oxford University Press, pp. 289–324. Evans, Roger, and Gerald Gazdar (1996), ‘DATR: A language for lexical knowledge representation’, Computational Linguistics, 22.2: 167–216. Evans, Vyvyan, and Melanie Green (2006), Cognitive Linguistics: An Introduction, Edinburgh: Edinburgh University Press. Fabb, Nigel (1988a), ‘Doing affixation in the GB syntax’, in Martin Everaert, Arnold Evers,
168 Contemporary Morphological Theories Riny Huybregts, and Mieke Trommelen (eds), Morphology and Modularity [Publications in Language Sciences 29], Dordrecht/Providence, RI: Foris, pp. 129–45. Fabb, Nigel (1988b), ‘English suffixation is constrained only by selectional restrictions’, Natural Language and Linguistic Theory, 6.4: 527–39. Fabri, Ray, Martina Urbas, and Markus Walther (1996), ‘A computational model of Minimalist Morphology’, Theoretical Linguistics, 22.3: 231–79. Fillmore, Charles J., Paul Kay, and Mary Catherine O’Connor (1988), ‘Regularity and idiomaticity in grammatical constructions: The case of let alone’, Language, 64.3: 501–38. Finkel, Raphael, and Gregory Stump (2009), ‘Principal parts and degrees of paradigmatic transparency’, in James P. Blevins and Juliette Blevins (eds), Analogy in Grammar: Form and Acquisition, Oxford: Oxford University Press, pp. 13–53. Flickinger, Daniel (1987), Lexical Rules in the Hierarchical Lexicon, Stanford University dissertation. Fodor, Jerry A. (1983), The Modularity of Mind, Cambridge, MA: MIT Press. Fraser, Norman M., and Greville G. Corbett (1995), ‘Gender, animacy, and declensional class assignment: A unified account for Russian’, in Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 1994, Dordrecht: Kluwer, pp. 123–50. Gazdar, Gerald, Ewan Klein, Geoffrey Pullum, and Ivan Sag (1985), Generalized Phrase Structure Grammar, Oxford: Blackwell. Goldberg, Adele (2006), Constructions at Work: The Nature of Generalization in Language, Oxford: Oxford University Press. Goldsmith, John (1976), ‘An overview of Autosegmental Phonology’, Linguistic Analysis, 2: 23–68. Gonda, Jan (1966), A Concise Elementary Grammar of the Sanskrit Language, trans. Gordon B. Ford, Jr, 2nd edn, Tuscaloosa/London: University of Alabama Press. Green, Antony D. (2007), Phonology Limited [Linguistics in Potsdam 27], Potsdam: Universitätsverlag Potsdam. Gurevich, Olga I. (2006), Constructional Morphology: The Georgian Version, University of California at Berkeley dissertation. Halle, Morris (1990), ‘An approach to morphology’, Proceedings of the 20th Annual Meeting of the North-East Linguistic Society, pp. 150–84. Halle, Morris (1997), ‘Distributed Morphology: Impoverishment and fission’, MIT Working Papers in Linguistics 30: 425–49. Halle, Morris, and Alec Marantz (1993), ‘Distributed Morphology and the pieces of inflection’, in Kenneth Hale and Samuel Jay Keyser (eds), The View from Building 20: Linguistics Essays in Honor of Sylvain Bromberger, Cambridge, MA: MIT Press, pp. 111–76. Halle, Morris, and K. P. Mohanan (1985), ‘Segmental phonology of Modern English’, Linguistic Inquiry, 16.1: 57–116. Halle, Morris, and Jean-Roger Vergnaud (1980), ‘Three dimensional phonology’, Journal of Linguistic Research, 1.1: 83–105. Hamp, Eric P. (1951), ‘Morphophonemes of the Keltic mutations’, Language, 27.3: 230–47. Harley, Heidi, and Rolf Noyer ([1999] 2003), ‘Distributed Morphology’, in Lisa Cheng and Rint Sybesma (eds), The Second GLOT International State-of-the-article Book: The Latest in Linguistics [Studies in Generative Grammar [SGG] 61], Berlin/New York: Mouton de Gruyter, pp. 463–96. Harris, Alice (1981), Georgian Syntax: A Study in Relational Grammar, Cambridge: Cambridge University Press. Harris, Alice (1984), ‘Inversion as a rule of Universal Grammar: Georgian evidence’, in David M. Perlmutter and Carol Rosen (eds), Studies in Relational Grammar 2, Chicago: University of Chicago Press, pp. 259–91. Harris, Zellig S. (1942), ‘Morpheme alternants in linguistic analysis’, Language, 18.3: 169–80.
References 169 Haspelmath, Martin (2007), ‘Pre-established categories don’t exist: Consequences for language description and typology’, Linguistic Typology, 11.1: 119–32. Haspelmath, Martin (2010), ‘Framework-free grammatical theory’, in Bernd Heine and Heike Narrog (eds), The Oxford Handbook of Linguistic Analysis, Oxford: Oxford University Press, pp. 341–65. Haspelmath, Martin (2011), ‘The indeterminacy of word segmentation and the nature of morphology and syntax’, Folia Linguistica, 45.1: 31–80. Haspelmath, Martin, and Andrea D. Sims (2010), Understanding Morphology, 2nd edn, London: Hodder Education. Heisenberg, Werner (1968), ‘Theory, criticism and a philosophy’, in From a Life in Physics: Evening Lectures at the International Centre for Theoretical Physics, Trieste, Italy, Vienna: International Atomic Energy Agency, pp. 31–46. Hockett, Charles F. (1947), ‘Problems of morphemic analysis’, Language, 23.4: 321–43. Hockett, Charles F. (1954), ‘Two models of grammatical description’, Word, 10.1: 210–34. Hoeksema, Jack (1985), Categorial Morphology, New York: Garland, [(1984), Rijksuniversiteit Groningen dissertation]. Hoeksema, Jack, and Richard D. Janda (1988), ‘Implications of process-morphology for Categorial Grammar’, in Richard T. Oehrle, Emmon Bach, and Deirdre Wheeler (eds), Categorial Grammars and Natural Language Structures, Dordrecht: Reidel, pp. 199–247. Hooper, Joan B. (1976), An Introduction to Natural Generative Phonology, New York: Academic Press. Hooper, Joan B. (1979), ‘Substantive principles in Natural Generative Phonology’, in Daniel A. Dinnsen (ed.), Current Approaches to Phonological Theory, Bloomington: Indiana University Press, pp. 106–25. Humboldt, Wilhelm von (1836), Über die Verschiedenheit des menschlichen Sprachbaues und ihren Einfluss auf die geistige Entwickelung des Menschengeschlechts, Berlin: Königliche Akademie der Wissenschaften. Jackendoff, Ray (1975), ‘Morphological and semantic regularities in the lexicon’, Language, 51.3: 639–71. Jackendoff, Ray (2002), Foundations of Language: Brain, Meaning, Grammar, Evolution, Oxford: Oxford University Press. Jaeggli, Osvaldo (1980), ‘Spanish diminutives’, in Frank H. Nuessel (ed.), Contemporary Studies in Romance Languages, Bloomington: Indiana University Linguistics Club, pp. 142–58. Janda, Richard D. (1982), ‘On formal identity and rule (un-)collapsibility: On lost and found generalizations in morphology’, WCCFL 1: 179–97. Janda, Richard D. (1986), On the Motivation for an Evolutionary Typology of Sound-structural Rules, University of California at Los Angeles dissertation. Janda, Richard D., and Brian D. Joseph (1986), ‘One rule or many? Sanskrit reduplication as fragmented affixation’, ESCOL ’85: 103–19. Janda, Richard D., and Brian D. Joseph (1992a), ‘Meta-templates and the underlying (dis)unity of Sanskrit reduplication’, ESCOL ’91: 160–73. Janda, Richard D., and Brian D. Joseph (1992b), ‘Pseudo-agglutinativity in Modern Greek verb-inflection and “Elsewhere”’, Chicago Linguistic Society, 28: 251–66. Janda, Richard D., and Brian D. Joseph (1999), ‘The Modern Greek negator µη(ν)(-) as a morphological constellation’, in Georgios Babiniótes and Amalia Mozer (eds), Proceedings of the 3rd International Conference on Greek Linguistics, pp. 341–51. Janda, Richard D., and M. Sandoval (1984), ‘Elsewhere’ in Morphology, Bloomington: Indiana University Linguistics Club. Jensen, J. T., and M. Stong-Jensen (1984), ‘Morphology is in the lexicon!’, Linguistic Inquiry 15: 474–98.
170 Contemporary Morphological Theories Karttunen, Lauri (1986), ‘Radical lexicalism’, in M. Baltin and A. Kroch (eds), Alternative Conceptions of Phrase Structure, Chicago: Chicago University Press, pp. 43–65. Kayne, Richard S. (1994), The Antisymmetry of Syntax, Cambridge, MA: MIT Press. Kiparsky, Paul (1973), ‘“Elsewhere” in phonology’, in Stephen R. Anderson and Paul Kiparsky (eds), A Festschrift for Morris Halle, New York: Holt, Rinehart, & Winston, pp. 93–106. Kiparsky, Paul (1982a), ‘Lexical Morphology and Phonology’, in Linguistic Society of Korea (ed.), Linguistics in the Morning Calm, Seoul: Hanshin, pp. 3–90. Kiparsky, Paul (1982b), ‘From cyclic to Lexical Phonology’, in Harry van der Hulst and Norval Smith (eds), The Structure of Phonological Representations, Dordrecht: Foris, part 1, pp. 131–75. Kiparsky, Paul (1985), ‘Some consequences of Lexical Phonology’, Phonology Yearbook, 2: 85–138. Kiparsky, Paul (2000), ‘Opacity and cyclicity’, Linguistic Review, 17.2–4: 351–65. Lasnik, Howard, and Juan Uriagereka (2005), A Course in Minimalist Syntax: Foundations and Prospects, Malden, MA: Blackwell. Lass, Roger (1990), ‘How to do things with junk: Exaptation in language evolution’, Journal of Linguistics, 26.1: 79–102. Leben, William (1973), Suprasegmental Phonology, Massachusetts Institute of Technology dissertation. Lieber, Rochelle (1981), On the Organization of the Lexicon, Bloomington: Indiana University Linguistics Club, [(1980), Massachusetts Institute of Technology dissertation]. Lieber, Rochelle (1983), ‘New developments in autosegmental morphology: Consonant mutation’, WCCFL 2: 165–75. Lieber, Rochelle (1992), Deconstructing Morphology: Word Formation in Syntactic Theory, Chicago: University of Chicago Press. ‘LMBM: Table of Contents’, (last accessed 24 December 2014). MacDonell, Arthur A. ([1927] 1986), A Sanskrit Grammar for Students, 3rd edn, Oxford: Oxford University Press. MacLaren, James (1935), MacLaren’s Gaelic Self- taught, 4th edn, rev., Glasgow: Alex. MacLaren & Sons. McCarthy, John J. (1979), Formal Problems in Semitic Phonology and Morphology, Massachusetts Institute of Technology dissertation. McCarthy, John J. (1981), ‘A prosodic theory of nonconcatenative morphology’, Linguistic Inquiry, 12.3: 373–418. McCarthy, John J. (1982), ‘Prosodic templates, morphemic templates, and morphemic tiers’, in Harry van der Hulst and Norval Smith (eds), The Structure of Phonological Representations, Dordrecht: Foris, part 1, pp. 191–223. McCarthy, John J., and Alan Prince (1993), ‘Generalized alignment’, in Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 1993, Dordrecht: Kluwer, pp. 79–153. McCarthy, John J., and Alan Prince ([1993] 2001), Prosodic Morphology I: Constraint Interaction and Satisfaction [RuCCS Technical Report 3], University of Massachusetts, Amherst, and Rutgers University, (last accessed 24 December 2014). McCarthy, John J., and Alan Prince (1994), ‘The emergence of the unmarked: Optimality in prosodic morphology’, in Merce Gonzales (ed.), Proceedings of the 24th Annual Meeting of the North-East Linguistic Society, pp. 333–79. McKean, Erin, ed. (2005), The New Oxford American Dictionary, 2nd edn, New York: Oxford University Press. Marantz, Alec (1982), ‘Re reduplication’, Linguistic Inquiry, 13.3: 435–82. Marantz, Alec (1988), ‘Clitics, morphological merger, and the mapping to phonological
References 171 structure’, in Michael Hammond and Michael Noonan (eds), Theoretical Morphology, San Diego: Academic Press, pp. 253–70. Marantz, Alec (1997), ‘No escape from syntax: Don’t try morphological analysis in the privacy of your own lexicon’, University of Pennsylvania Working Papers in Linguistics, 4: 201–25. Matthews, P. H. (1972), Inflectional Morphology: A Theoretical Study Based on Aspects of Latin Verb Conjugation [Cambridge Studies in Linguistics 6], Cambridge: Cambridge University Press. Matthews, P. H. (1991), Morphology, 2nd edn, Cambridge: Cambridge University Press. Mayerthaler, Willi ([1981] 1988), Morphologische Natürlichkeit/Morphological Naturalness [Linguistica Extranea, Studia 17]. Ann Arbor: Karoma. Mayrhofer, Manfred ([1964] 1972), A Sanskrit Grammar [Alabama Linguistic and Philological Series 20], trans. Gordon B. Ford, Jr, Tuscaloosa: University of Alabama Press. Michaelis, Laura A., and Knud Lambrecht (1996), ‘Toward a construction-based theory of language function: The case of nominal extraposition’, Language, 72.2: 215–47. Montague, Richard (1970), ‘Universal grammar’, Theoria, 36.3: 373–98. Montague, Richard (1973), ‘The proper treatment of quantification in ordinary English’, in J. Hintikka, J. Moravcsik, and P. Suppes (eds), Approaches to Natural Language, Dordrecht: Reidel, pp. 222–42. Nida, Eugene A. (1949), Morphology: The Descriptive Analysis of Words, 2nd edn, Ann Arbor: University of Michigan Press. Papke, Julia (2010), Classical Sanskrit Preverb Ordering: A Diachronic Study, Ohio State University dissertation. Partee, Barbara H. (2001), ‘Montague grammar’, in Neil J. Smelser and Paul B. Baltes (eds), International Encyclopedia of the Social and Behavioral Sciences, Oxford: Pergamon/Elsevier Science, vol. 15, pp. 9995–9. Peirce, Charles S. (1965), Collected Papers, ed. C. Hartshorne and P. Weiss, Cambridge, MA: Harvard University Press. Perlmutter, David M. (1988), ‘The Split Morphology Hypothesis: Evidence from Yiddish’, in Michael Hammond and Michael Noonan (eds), Theoretical Morphology, San Diego: Academic Press, pp. 79–100. Pinker, Steven (1999), Words and Rules: The Ingredients of Language, London: Weidenfeld & Nicolson. Pinker, Steven, and Alan Prince (1994), ‘Regular and irregular morphology and the psychological status of rules of grammar’, in Susan D. Lima, Roberta L. Corrigan, and Gregory K. Iverson (eds), The Reality of Linguistic Rules, Amsterdam/Philadelphia: John Benjamins, pp. 321–51. Plag, Ingo (1996), ‘Selectional restrictions in English suffixation: A reply to Fabb (1988)’, Linguistics, 34.4: 769–98. Plag, Ingo (2006), ‘Productivity’, in Bas Aarts and April M. S. McMahon (eds), The Handbook of English Linguistics, Malden, MA: Blackwell, pp. 537–56. Pollard, Carl J., and Ivan A. Sag (1994), Head-driven Phrase Structure Grammar, Chicago: University of Chicago Press. Prince, Alan, and Paul Smolensky (2004), Optimality Theory: Constraint Interaction in Generative Grammar, Malden, MA: Blackwell. Pullum, Geoffrey K. (2013), ‘The central question in comparative syntactic metatheory’, Mind and Language, 28.4: 492–521. Pullum, Geoffrey K., and Arnold M. Zwicky (1991), ‘A misconceived approach to morphology’, in Dawn Bates (ed.), Proceedings of the West Coast Conference on Formal Linguistics, pp. 387–98. Pyatt, Elizabeth J. (1997), An Integrated Model of the Syntax and Phonology of Celtic Mutation, Harvard University dissertation.
172 Contemporary Morphological Theories Rajguru, D. K. (1938), A Manual of Sanskrit Grammar, Bombay: K. B. Dhawale. Riehemann, Susanne (1998), ‘Type-based derivational morphology’, Journal of Comparative Germanic Linguistics, 2.1: 49–77. Riehemann, Susanne (2001), A Constructional Approach to Idioms and Word Formation, Stanford University dissertation. Robins, R. H. (1959), ‘In defence of WP’, Transactions of the Philological Society, 58.1: 116–44. Rogers, Henry (1980), ‘/m/ in Scots Gaelic’, Lingua, 52.2–3: 223–31. Sadock, Jerrold (1985), ‘Autolexical syntax: A theory of noun incorporation and similar phenomena’, Natural Language and Linguistic Theory, 3.4: 379–439. Sadock, Jerrold (1988), ‘The Autolexical classification of lexemes’, in Michael Hammond and Michael Noonan (eds), Theoretical Morphology, San Diego: Academic Press, pp. 271–90. Sadock, Jerrold (1991), Autolexical Syntax, Chicago: University of Chicago Press. Sadock, Jerrold (2012), The Modular Architecture of Grammar, New York: Cambridge University Press. Salam, Abdus (1990), Unification of Fundamental Forces, New York: Cambridge University Press. Sapir, Edward (1921), Language, New York: Harcourt, Brace, & World. Schmerling, Susan F. (1983), ‘Montague morphophonemics’, Chicago Linguistic Society, 19(P): 222–37. Selkirk, Elisabeth O. (1982), The Syntax of Words [Linguistic Inquiry Monograph 7], Cambridge, MA: MIT Press. Siegel, Dorothy L. (1979), Topics in English Morphology, New York: Garland. Singer, Kora (1999), Agreement: Cross-Linguistic Variation and Acquisition, University of Chicago dissertation. Spencer, Andrew (1991), Morphological Theory, Cambridge, MA: Blackwell. Spencer, Andrew (1993), ‘The morphology-syntax interface’ [Review of Sadock 1991], Journal of Linguistics, 29.1: 143–56. Spencer, Andrew (2003), ‘Periphrastic paradigms in Bulgarian’, in Uwe Junghanns and Luka Szucsich (eds), Syntactic Structures and Morphological Information [Interface Explorations 7], Berlin: Mouton de Gruyter, pp. 249–82. Spencer, Andrew (2013), Lexical Relatedness: A Paradigm-based Model, Oxford: Oxford University Press. Spencer, Andrew, and Ana Luís (2013), ‘The canonical clitic’, in Dunstan Brown, Marina Chumakina, and Greville G. Corbett (eds), Canonical Morphology and Syntax, Oxford: Oxford University Press, pp. 123–50. Spencer, Andrew, and Arnold M. Zwicky (1998), ‘Introduction’, in Andrew Spencer and Arnold M. Zwicky (eds), Handbook of Morphology, Oxford/Malden, MA: Blackwell, pp. 1–10. Stampe, David (1979), A Dissertation on Natural Phonology, New York: Garland. Steedman, Mark, and Jason Baldridge (2011), ‘Combinatory Categorial Grammar’, in Robert D. Borsley and Kersti Börjars (eds), Non-transformational Syntax: Formal and Explicit Models of Grammar, Malden, MA: Wiley–Blackwell, pp. 181–224. Stewart, Jr, Thomas W. (2000), ‘Apparent congruence and elusive bootstraps: Assessing systemic similarity in language contact’, Poster presented at the 9th International Morphology Meeting, Vienna. Stewart, Jr, Thomas W. (2001), ‘Georgian verb agreement without extrinsic ordering’, OSU Working Papers in Linguistics, 56: 107–33. Stewart, Jr, Thomas W. (2004), Mutation as Morphology: Bases, Stems, and Shapes in Scottish Gaelic, Ohio State University dissertation. Stewart, Jr, Thomas W. (2013), ‘The sub-types of initial lenition in Scottish Gaelic’, in Janet Cruickshank and Robert McColl Millar (eds), After the Storm: Papers from the Forum
References 173 for Research on the Languages of Scotland and Ulster triennial meeting, Aberdeen 2012, Aberdeen: Forum for Research on the Languages of Scotland and Ireland, pp. 100–16. Stewart, Jr, Thomas W., and Brian D. Joseph (2009), ‘How big can case systems get?: Evidence from Scottish Gaelic’, Word Structure, 2.1: 108–20. Stewart, Jr, Thomas W., and Gregory T. Stump (2007), ‘Paradigm Function Morphology and the morphology–syntax interface’, in Gillian Ramchand and Charles Reiss (eds), The Oxford Handbook of Linguistic Interfaces, Oxford/New York: Oxford University Press, pp. 383–421. Stump, Gregory T. (1990), ‘Breton inflection and the Split Morphology Hypothesis’, in Randall Hendrick (ed.), The Syntax of the Modern Celtic Languages [Syntax and Semantics 23], San Diego: Academic Press, pp. 97–119. Stump, Gregory T. (1991), ‘A paradigm-based theory of morphosemantic mismatches’, Language, 67.4: 675–725. Stump, Gregory T. (1992), ‘On the theoretical status of position class restrictions on inflectional affixes’, in Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 1991, Dordrecht: Kluwer, pp. 211–41. Stump, Gregory T. (1993a), ‘On rules of referral’, Language, 69.3: 449–79. Stump, Gregory T. (1993b), ‘Position classes and morphological theory’, in Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 1992, Dordrecht: Kluwer, pp. 129–80. Stump, Gregory T. (1993c), ‘How peculiar is evaluative morphology?’, Journal of Linguistics, 29.1: 1–36. Stump, Gregory T. (1995), ‘The uniformity of head marking in inflectional morphology’, in Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 1994, Dordrecht: Kluwer, pp. 245–96. Stump, Gregory T. (2001), Inflectional Morphology: A Theory of Paradigm Structure [Cambridge Studies in Linguistics 93], Cambridge: Cambridge University Press. Stump, Gregory T. (2005), ‘Word- formation and inflectional morphology’, in Pavol Štekauer and Rochelle Lieber (eds), Handbook of Word-Formation, Dordrecht: Springer, pp. 49–72. Stump, Gregory, and Raphael A. Finkel (2013), Morphological Typology: From Word to Paradigm [Cambridge Studies in Linguistics 138], Cambridge: Cambridge University Press. Toman, Jindřich (1983), Wortsyntax, Tübingen: Max Niemeyer. Toman, Jindřich (1998), ‘Word Syntax’, in Andrew Spencer and Arnold M. Zwicky (eds), Handbook of Morphology, Oxford/Malden, MA: Blackwell, pp. 306–21. Wasow, Thomas (1993), ‘Grammatical theory’, in Michael I. Posner (ed.), Foundations of Cognitive Science, Cambridge, MA: MIT Press, pp. 161–205. ‘Welcome to the Surrey Morphology Group’, (last accessed 24 December 2014). Wheeler, Max W. (1993), ‘On the hierarchy of naturalness principles in inflectional morphology’, Journal of Linguistics, 29.1: 95–111. Whitney, William Dwight ([1885] 1945), The Roots, Verb-forms, and Primary Derivatives of the Sanskrit Language [American Oriental Series 30], New Haven, CT: American Oriental Society. Whitney, William Dwight ([1889] 1960), Sanskrit Grammar: Including both the Classical Language, and the Older Dialects, of Veda and Brahmana, London: Oxford University Press. Whorf, Benjamin Lee (1945), ‘Grammatical categories’, Language, 21.1: 1–11. Williams, Edwin (1981), ‘On the notions “lexically related” and “head of a word”’, Linguistic Inquiry, 12.2: 245–74. Wunderlich, Dieter (1996), ‘Minimalist Morphology: The role of paradigms’, in Geert Booij and Jaap van Marle (eds), Yearbook of Morphology 1995, Dordrecht: Kluwer, pp. 93–114.
174 Contemporary Morphological Theories Wunderlich, Dieter, and Ray Fabri (1995), ‘Minimalist Morphology: An approach to inflection’, Zeitschrift für Sprachwissenschaft, 14.2: 236–94. Wurzel, Wolfgang Ullrich (1987), ‘System-dependent morphological naturalness in inflection’, in Wolfgang U. Dressler, Willi Mayerthaler, Oswald Panagl, and Wolfgang U. Wurzel (eds), Leitmotifs in Natural Morphology [Studies in Language Companion Series 10], Amsterdam/Philadelphia: John Benjamins, pp. 59–96. Wurzel, Wolfgang Ullrich (1989), Inflectional Morphology and Naturalness, Dordrecht: Kluwer. Zwicky, Arnold M. (1985), ‘How to describe inflection’, Berkeley Linguistics Society, 11: 372–86. Zwicky, Arnold M. (1986), ‘The general case: Basic form versus default form’, Berkeley Linguistics Society, 12: 305–14. Zwicky, Arnold M. (1987a), ‘Morphological rules, operations, and operation types’, ESCOL ’87: 318–34. Zwicky, Arnold M. (1987b), ‘Phonological and morphological rule interactions in highly modular grammars’, ESCOL ’87: 523–32. Zwicky, Arnold M. (1992), ‘Some choices in the theory of morphology’, in Robert D. Levine (ed.), Formal Grammar: Theory and Implementation, New York: Oxford University Press, pp. 327–71.
Index
abstractive approach, 77, 110; see also constructive approach adjacency, 3, 7, 50, 77, 111, 133, 135–6, 146 Affix Ordering Generalisation, 39, 50, 141 affixation, 8, 11, 23, 24, 29, 44, 48, 50–1, 54, 73–4, 80, 83, 94, 98, 109, 111, 118, 131, 132, 136, 148n agreement, 12–14, 48, 60, 82, 112–26, 135, 147n allomorphy, 9n, 36, 37, 53, 59, 61, 79, 87n, 118, 126, 131, 135–7, 144, 149n, 162n analogy, 27, 30, 31, 62, 76–80, 109, 110, 126, 138–9, 144–5, 155, 156, 162n, 163n animacy, 12, 54, 65–6 aspect, 50, 60, 87n association, 8, 11, 32, 43, 59, 73–5, 89n, 96, 97, 107, 111, 116, 126, 133, 135, 137, 143–4, 145, 155, 159–60 attribute-value matrix (AVM), 28–9, 86n, 142 biuniqueness, 43, 55–8, 77–9, 101, 137–8, 152, 154 blocking, 36, 45, 51, 80, 99–100, 125, 127, 154, 155, 160 Canonical Typology, 71, 153–5, 162n; see also typology case, 12, 28, 40–1, 61, 83, 93, 95–101, 105–7, 111, 134, 146, 147n, 149n, 152, 162n cell, 51, 67–8, 71, 79, 91, 105, 109–10, 112, 123, 144, 153 clitic, 11, 14, 15–19, 23, 25, 45–6, 89n, 118, 135
co-derivative, 2, 30 compounding, 11–12, 14, 17, 24, 82–4, 95, 112, 128–30, 141–2, 149n, 159 concatenative morphology, 7, 8, 23–4, 26, 29, 48, 73–4, 76, 78, 81, 83, 109, 125, 143 congruence see congruity congruity, 56–8, 137 conjugation, 77, 128, 131, 133, 136–41, 144–5 construct, 27, 29, 81–2, 115–18, 129, 138 construction, 17, 26–7, 30–1, 40, 51, 81–4, 95, 112, 115, 123–4, 132, 146, 157 construction schema, 27, 31, 96–7, 116, 118, 132, 148n constructional iconicity, 55, 57–8, 101, 137, 157 constructive approach, 77–80, 125; see also abstractive approach context-free operation, 24, 37–8, 87n context-sensitive operation, 24, 37, 55, 94, 127 DATR, 63, 65, 88n, 123, 139 declension, 20, 64–6, 77–9, 93, 99, 104–5 deponency, 18, 138 derivation, 10, 16–17, 21, 24–5, 42–5, 54–6, 61, 65, 85, 94, 95–6, 101–2, 114, 116, 119, 122n, 123n, 135, 157, 176, 179–80, 182–6, 187–9, 191, 193, 195–200, 203–5, 218, 220, 225, 230, 231n; see also compounding derived environment, 43, 45, 52 Elsewhere Condition (EC), 13–14, 45, 93, 97, 100, 113, 127; see also Pān·inian Principle
176 Contemporary Morphological Theories environment, derived see derived environment exponence, 8, 11, 14, 19, 29, 33, 36–7, 48, 50, 60, 68, 70, 77–8, 88n, 98–9, 109–10, 115–16, 118, 119, 122–3, 124, 127, 135, 140, 142–3, 146, 148n, 159 Extended Word and Paradigm (EWP), 10, 86n, 88n extrafixation, 24–6 feature, morphosyntactic, 8, 11–14, 17, 21, 28, 34–40, 48, 50–1, 54, 63, 67, 71–2, 73–5, 80, 81–2, 87n, 88n, 92, 97–8, 101, 105, 107, 108, 111, 113, 115, 118, 119–20, 121–3, 127, 130–1, 133, 135, 140–2, 145–6, 147n, 148n, 149n, 152, 153 feature layering see layering, feature feature percolation, 7, 81–2, 111, 145–6 fission, 36, 118 formative, 2, 12, 136, 138 frequency, 6, 31, 45, 61–2, 74, 80, 103, 109, 122, 125, 128–9, 138, 144, 149n, 152, 156–9, 163n functor, 23–4 fusion, 35–6, 50, 55, 59, 61, 118, 119, 122–3, 126, 133, 152, 162n gender, 12, 17, 50, 65–6, 95, 98, 101, 146 head, 1, 12, 17, 22, 28, 33, 35, 37, 39, 81–4, 86n, 89n, 93, 130, 133, 141–3, 145–6, 149n Identity Function Default (IFD), 69, 88n, 107, 142 impoverishment, 36, 50, 98, 118–19 incremental (theory of morphology), 7–8, 45, 68, 77, 96, 119, 127 indexation of lexemes or stems, 32, 70, 88n, 137, 142, 154 of operations or rules, 22–3, 25, 93, 123–4 of segments, 29 indexicality, 55–6 inferential (theory of morphology), 37, 65–6, 116–17, 123 inflection, 4, 6, 7, 11–14, 16–17, 19–20, 28, 30, 33, 37–8, 39–40, 47–8, 50–1,
56–7, 60, 62, 65–6, 67–8, 70–2, 76–80, 82–3, 86n, 88n, 90–9, 101, 103, 104, 106, 109, 111–12, 114–15, 118, 119, 122, 124, 126, 128–9, 130, 133, 136–8, 141–2, 145, 146, 148n, 152–4, 159, 162n inheritance hierarchy, 27–30, 48, 50, 63–6, 69, 75, 88n, 104–5, 115, 137, 139–41, 147, 150n insertion, 14, 32, 34–8, 44, 48, 51, 83, 86n, 87n, 89n, 93, 97–8, 113–14, 118–20, 127, 130, 133, 137, 145–6, 147n interface, 14, 15, 18, 20–1, 34–5, 38, 41, 48, 52, 67, 83, 86n, 130–1 item and arrangement (IA), 20, 23, 77, 81, 98, 125 item and process (IP), 20, 23, 77, 98 layering, feature, 12–13, 39, 92, 96, 114, 123 level-ordering, 44–6, 99, 162n lexical entry, 6, 15, 19–23, 29, 42, 45, 47–51, 63–6, 70–1, 75, 78, 80, 83–4, 88n, 93–5, 105, 107, 111, 130–1, 137, 139–40, 143, 147n, 148n, 150n lexical strength, 59, 62, 103, 138, 145 Lexicalist Hypothesis, 5, 12, 83 lexicon, 4, 6–7, 11, 14–15, 20, 23, 27, 30, 32, 34, 39–41, 43–5, 48, 59, 60–1, 63, 64, 66, 67, 72, 84, 86n, 89n, 95, 122, 142 markedness, 53–8, 61, 101, 118, 120, 121, 127, 143 meta-redundancy rule, 45, 88n meta-template see meta-redundancy rule metatheory, 3–4, 23, 26, 67, 88n, 90 morph empty, 5, 16, 50 portmanteau, 5, 23 zero, 5, 7, 8, 13, 37–8, 45, 47, 50, 53, 58, 61, 69, 87n, 98, 107, 111, 117, 121–2, 126, 127 morpheme boundary, 25, 43–5, 50, 103, 130, 139 morphological component, 1, 3–4, 6, 7, 10, 15–21, 33, 38–9, 41–2, 43–4, 45, 47–8, 52, 86n, 87n, 114, 119, 130–1, 142, 146, 147, 149n
Index 177 morphological merger, 35–6, 133 morphological meta-generalisation, 70, 150 morphological module see morphological component Morphological Spelling (MS), 39–41, 87n, 98–9, 119–20, 133, 135, 149n morphologisation, 45, 56 morphome, 70, 80, 88n, 95–6, 109, 149n morphonology see morphophonology morphophonology, 1, 7, 14, 18–20, 42, 45, 52–3, 59–62, 70, 104, 118, 120, 125, 129, 130–1, 137, 150n morphosyntactic property, 58, 68, 80, 88n, 115, 124–5, 141 morphosyntactic representation (MSR), 11–14, 69, 91–2, 113–14, 123 neologism, 62, 122, 158–9 non-concatenative morphology, 20, 23–4, 29, 48, 54, 70, 73–4, 94–5, 97, 111, 125 number, 12, 50, 93, 95, 98–9, 101, 109, 115–16, 118, 120–1, 126, 127, 147n, 152 operation, 10, 14, 19–20, 22–5, 33–7, 55, 70, 73–4, 87n, 88n, 94–5, 97–8, 108, 111, 118–19, 128, 129–31, 133, 135, 146, 147n, 153 optimal symbolising, 54, 101 Optimality Theory (OT), 46, 52, 75–6, 108–9, 143 Pān·inian determinism hypothesis, 106, 123, 125 Pān·inian principle, 13, 69, 70, 106, 127; see also Elsewhere Condition; Pān·inian determinism hypothesis paradigm, 1, 4, 6, 9n, 14, 20, 47–8, 50–1, 56–7, 61–2, 64–8, 71–2, 76–80, 86n, 88n, 89n, 91, 96–7, 100, 101, 103–7, 109–10, 113–15, 121–3, 125–7, 128, 136–7, 140–1, 144–5, 148n, 153–5 paradigm construction, 51, 88n paradigm function (PF), 68–71, 106–7, 142 percolation, feature see feature percolation person, 12–13, 48, 50, 54, 113–16, 119, 120–1, 124, 126–7, 148n, 152
position class, 68–9, 72, 114, 115, 117–18, 123–4, 127, 148n pre-association, 73–4 predictiveness, 14, 46, 78–80, 82, 87n, 88n, 110, 114, 132, 148n principal parts, 76–8, 80 productivity, 6, 19–21, 28–9, 31, 45, 48, 53, 55–6, 62, 70, 83–4, 86n, 91, 95–9, 101, 103, 105, 109, 111, 122, 132, 138, 144–5, 148n, 150n, 155–60, 162n readjustment rule, 33, 35, 36, 87n, 97–8, 148n realisational (approach to morphology), 7, 8, 12, 14, 32, 36–9, 41, 66, 67–71, 87n, 88n, 96–8, 116, 118, 123, 125, 141–2, 147n, 148n, 162n reduction, 45, 51, 149n reduplication, 73–4, 87n, 108, 128 referral, 63, 66, 105, 107, 122 regularity, 6, 21, 25, 28, 31, 48–9, 62, 70, 91, 97–9, 102–3, 105, 109, 111, 122, 144–5, 148n, 149n, 156, 157–8, 159, 163n right-hand head rule (RHR), 81–3, 145–6 rule, lexical, 19–20, 24 cyclic, 43–6, 135 post-cyclic, 45 rule, post-lexical, 43 rule, readjustment see readjustment rule rule, realisation, 68–9, 106–7, 123–4 rule block, 14, 37–8, 45, 68–70, 72, 88n, 106–7, 113–14, 123–5, 127 rule ordering, 12–14, 38–9, 42, 44, 69, 113–14, 124, 125, 127, 145, 162n rule strata, 42–4, 46, 120, 127, 135, 145, 155 sandhi, 19, 128, 129, 135, 137, 150n schema, 24, 27, 29–31, 48, 60, 69, 95–7, 109–10, 115–18, 122, 124–5, 132, 139, 142, 147n, 148n, 158; see also construction schema schema unification, 29, 31, 132 Separation Hypothesis, 20, 39, 41, 70, 79, 87n, 98–9, 120, 149n specificity, 26, 51, 69–70, 92, 120–1, 127 Split Morphology Hypothesis, 7, 11, 39, 67, 141, 146
178 Contemporary Morphological Theories stem, 7, 10–11, 13–14, 15–17, 19–21, 29, 37, 39, 41, 46–51, 53, 56, 60–1, 64–6, 68, 70, 73–5, 77–9, 81, 86n, 88n, 89n, 91–6, 98, 99, 100–1, 103–9, 111, 114–15, 118–22, 128–9, 131–2, 135–46, 148n, 149n, 150n strata, rule see rule strata suppletion, 6, 9n, 48, 55, 70, 89n, 130, 136, 143, 149n, 159 syncretism, 57–8, 79, 95, 101, 107, 109, 116–18, 121, 126 system-defining structural property (SDSP), 56–7 tense, 35, 50, 60, 87n, 112–13, 133, 148n, 149n
Trisyllabic Shortening, 44–5, 87n typology, 152–5 under-specification, 21, 32, 50–1, 88n Vocabulary (item), 32–8, 97–8, 133, 147n word and paradigm (WP), 14, 76–8, 80–1, 88n, 89n, 125–6, 148n, 153; see also Extended Word and Paradigm (EWP) word-form, 4, 47, 51, 61, 76–80, 95, 97, 101, 102, 109–10, 114, 159 word-formation, 1, 3, 6, 8, 11, 92; see also compounding; derivation word-formation rule (WFR), 11–14, 92, 130, 139