Constraints on Suffixation (Linguistische Arbeiten) 3484303409, 9783484303409

The book is a detailed generative study of a number of derivational and inflectional processes of suffixation in contemp

266 117 8MB

English Pages 100 [108]

Report DMCA / Copyright

DOWNLOAD PDF FILE

Table of contents :
Preface
Chapter I. Theoretical Background
1.1. The structure of the Lexicon
1.2. Lexical Morphology
1.2.1. Morphological units
1.2.2. Morphological rules
1.2.3. The scope of morphological generalizations
1.3. Lexical phonology
1.4. Methodology
1.5. Constraints on rules
Chapter II. Constraints on Stem-based Suffocation in English - the Predictions of Lexical Phonology
2.1. Inflection
2.2. Derivation
Chapter III. Constraints on the Base of Morphological Rules - Verbalizing Suffixes in Polish
3.1. Verbalizing suffixes in Polish
3.2. The verbalizing suffix //i//
3.2.1. The phonological properties
3.2.2. The BASE of the morphological rule of //i// suffixation
3.3. The verbalizing suffix //ēj//
3.3.1. The phonological properties
3.3.2. The BASE of the morphological rule of //ēj// suffixation
3.4. Two more suffixes
3.4.1. Stem- based //i// suffixation
3.4.2. The root-based //ēj// suffixation
3.5. Root-based suffixes
3.5.1. The verbalizing suffix //aj//
3.5.2. The verbalizing suffix //a//
3.5.3. The verbalizing suffix //ē//
3.5.4. The notion of the root-based rule
Chapter IV. Constraints on the Operation of Morphological Rules - Deverbal Suffocation in Polish
4.1. The constraints
4.2. (ł) Participle
4.3. Derived Imperfective
Concluding remarks
References
Recommend Papers

Constraints on Suffixation (Linguistische Arbeiten)
 3484303409, 9783484303409

  • 0 0 0
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up
File loading please wait...
Citation preview

Linguistische Arbeiten

340

Herausgegeben von Hans Altmann, Peter Blumenthal, Herbert E. Brekle, Gerhard Heibig, Hans Jürgen Heringer, Heinz Vater und Richard Wiese

Adam

Wojcicki

Constraints on Suffixation A Study in Generative Morphology of English and Polish

Max Niemeyer Verlag Tübingen 1995

Die Deutsche Bibliothek-CIP-Einheitsaufnahme Wöjcicki, Adam: Constraints on Suffixation.: a study in generative morphology of English and Polish / Adam Wöjcicki. - Tubingen : Niemeyer, 1995 (Linguistische Arbeiten ; 340) NE: GT ISBN 3-484-30340-9

ISSN 0344-6727

© Max Niemeyer Verlag GmbH & Co.KG, Tübingen 1995 Das Werk einschließlich aller seiner Teile ist urheberrechtlich geschützt. Jede Verwertung außerhalb der engen Grenzen des Urheberrechtsgesetzes ist ohne Zustimmung des Verlages unzulässig und strafbar. Das gilt insbesondere für Vervielfältigungen, Übersetzungen, Mikroverfilmungen und die Einspeicherung und Verarbeitung in elektronischen Systemen. Printed in Germany. Druck: Weihert-Druck GmbH, Darmstadt Einband: Hugo Nadele, Nehren

Table of contents

Preface Chapter I. Theoretical Background

1 3

1.1. The structure of the Lexicon 1.2. Lexical Morphology 1.2.1. Morphological units 1.2.2. Morphological rules 1.2.3. The scope of morphological generalizations 1.3. Lexical phonology 1.4. Methodology 1.5. Constraints on rules

3 5 5 8 9 10 11 13

Chapter II. Constraints on Stem-based Suffixation in English - the Predictions of Lexical Phonology

14

2.1. Inflection 2.2. Derivation

Chapter III. Constraints on the Base of Morphological Rules - Verbalizing Suffixes in Polish 3.1. Verbalizing suffixes in Polish 3.2. The verbalizing suffix //i// 3.2.1. The phonological properties 3.2.2.'The BASE of the morphological rule of / i / suffixation 3.3. The verbalizing suffix //ej// 3.3.1. The phonological properties 3.3.2. The BASE of the morphological rule of //ej// suffixation 3.4. Two more suffixes 3.4.1. Stem- based //i// suffixation 3.4.2. The root-based //ej// suffixation 3.5. Root-based suffixes 3.5.1. The verbalizing suffix //aj// 3.5.2. The verbalizing suffix //a// 3.5.3. The verbalizing suffix //e// 3.5.4. The notion of the root-based rule

Chapter IV. Constraints on the Operation of Morphological Rules - Deverbal Suffixation in Polish 4.1. The constraints 4.2. (!) Participle 4.3. Derived Imperfective

Concluding remarks References

14 33

53 53 54 54 55 61 61 62 66 66 68 70 71 73 74 75

81 81 82 87

95 97

Preface You are in the first page of a book on generative morphology. The pages to follow are devoted to the discussion of a number of English and Polish word-building processes of suffixation and a variety of constraints on such processes. The theoretical underpinnings axe laid out in Chapter 1. As it is not intended to be an overview of the current state of the morphological theory only issues relevant to the present work are raised. Constraints on rules governing the distribution of English inflectional and derivational suffixes are discussed in Chapter 2. The allomorphy of English inflections is shown to follow from language-specific constraints on syllable structure but the main thrust is on constraints on suffixation inherent in the major tenets of the theory of lexical Phonology: lexical STRATA and the BRACKET ERASURE CONVENTION. Chapter 3 is devoted to the discussion of constraints on the BASE of morphological rules of verbalization in Polish. In Chapter 4 several constraints on the OPERATION of morphological rules proposed in the literature are evaluated against the background of Polish rules of deverbal suffixation. Each Chapter is divided into numbered sections and subsections some of which axe titled. The study is couched in the generative linguistic paradigm and it makes sense only within a broader descriptive framework provided for English by Rubach (1985) and Halle and Mohanan (1985) and for Polish by works of Laskowski (1975), Gussman (1978) (1980) and Rubach (1984). In my academic work I have benefited from assistance and scholary advice of several people. First and foremost, I owe an enormous intellectual and personal debt to Professor Jerzy Rubach who made me interested in generative linguistics through his inspiring teaching and many long and lively conversations. Professor Andrzej Boguslawski has influenced me both by what I have managed to read out of his immense scholary output and by what he said on various occasions. I have also learned a lot from Professor Wolfgang U. Dressier and Dr Heinz Giegerich. I wrote most of this book in the academic year 1989/90 while I was a student in the English Language Department at the University of Edinburgh. I am very grateful to the British Council for financing my stay there. I am grateful to Przemyslaw Pawelec, Professor Jerzy Rubach, Professor Piotr Ruszkiewicz, Professor Richard Wiese and Professor Jerzy Welna for reading earlier drafts of this work and making enlightening comments. My wife Beata's editorial assistance was invaluable. Finally I wish to thank my parents. All errors in the present work Eire mine.

Chapter I Theoretical Background 1.1. The structure of the Lexicon In more recent models of the generative linguistic paradigm the study of structure of complex words constitutes an essential part of the theory of one of the components of Grammar - the Lexicon. This state of affairs is in sharp contrast to what things used to be like in the early days of generative enterprise. Then, as part of the legacy of the post-bloomfieldian structuralism Lexicon was conceived of as a storage of idiosyncratic information about language, a mere "appendix of the grammar" (Bloomfield 1933:274); grammar being the sole locus of linguistic regularity (Lang (1990:9) aptly refers to this notion of the Lexicon as "the ragbag of oddities"). At the same time "for proponents of early generative grammar, grammar consisted of syntax and phonology (...) with the result that morphology was lost somewhere in between" (Aronoff 1976:4). If, following Anderson (1987), we assume that the principles of word structure can be divided between those that govern the distribution of sub-constituents of a word (i.e. morphotactics) and those that govern the variation in shape shown by these elements (allomorphy)1, then in the early generative framework morphotactics belonged to syntax and allomorphy to phonology. N. Chomsky in his early work defined syntax as the study of "all of the grammatical sequences of morphemes of a language" (Chomsky 1957:32) while all allomorphy regardless of the conditioning factors was encompassed by the domain of phonology. In Anderson's picturesque formulation "with neither morpheme distribution nor allomorphy to account for morphologists could safely go to the beach" (Anderson 1982:571). Subsequently, two distinct though simultaneous developments occurred to change this picture. First, the principles governing the structure of complex words were gradually recognised as distinct from either syntactic or phonological rules. Second, such principles were relegated from both the phonological and the syntactic components into the Lexicon. In the words of Scalise (19984:1) "the development of the transformational generative grammar from its beginning up to the present can be seen as a progressive refinement of the structure of the lexical component" This major breakthrough began in the late sixties with the advent of the Lexicalist Hypothesis (Chomsky 1970), on the one hand and the recognition of what was termed "readjustment rules" in Chomsky and Halle (1968), on the other. Chomsky (1970) contrasted two types of nominale in English: what he called derived nominalizations (since they axe traditionally regarded as the result of derivational morphology) and the gerundive nominalizations in (ing). The essence of his argument is 1

The distinction is valid for all students of morphology within the post-bloomfieldian structuralist framework (Joos 1957).

4 that derived nominals share many of the properties of words, including monomorphemic words, while the gerundive nominals behave more like syntactic collocations. Derived nominals are morphologically, syntactically and semantically idiosyncratic while gerundive nominalizations are regular and semantically transparent. Accordingly, Chomsky (1970) argued, it is only the gerundive nominals that should be derived by syntactic rules. The derived nominals, on the other hand, should be accounted for in terms of a separate set of grammatical rules, which, Chomsky (1970) suggested, apply in the Lexicon. The Lexicalist Hypothesis of Chomsky (1970) was further extended by Jackendoff (1972) to apply to all derivationally complex words regardless of their regularity. Jackendoff (1972) achieved this aim indirectly by stipulating that transformations should only be permitted to operate on syntactic constituents. Some linguists distinguish two versions of the Extended Lexicalist Hypothesis (the strong and the weak) depending on whether or not inflectional morphology is processed in the lexical or in the syntactic component. Halle (1973) and Lieber (1980) are among those who argue for the strong version while Anderson explicitly defends the weak version (counter arguments in Jensen and Stong Jensen (1984) and Rice (1985). Notice that Anderson's position crucially rests on the assumption that the distinction between the derivational and inflectional morphological processes is discrete, whereas this very assumption is seriously undermined in a number of recent studies (Bybee 1985, Carstairs 1987, Dressier 1988, Di Scullio and Williams 1988) where the distinction at hand is argued to be "more of a continuum than a matter of discreteness" (Taylor 1989 :179). Among the rules termed "readjustment rules"2 in Chomsky and Halle (1968) we find a significant number of plainly morphological ones. These are the readjustment rules which apply to specific derivable formatives spelling out the form of particular morphemes in specific morphological environments. (A good example of such a rule (110) in Chomsky and Halle (1968). Both Chomsky (1970) and Chomsky and Halle (1968) merely suggested (the latter implicitly) that morphology should exist without saying what it should be like. The task of elaborating a theory of words structure distinct from both phonology and syntax was undertaken in a number of works written in the seventies and early eighties the most influential of which have been Halle (1973), Jackendoff (1977), Allen (1978), Lieber (1980), Williams (1981) (the survey of major issues in Scalise (1984). The structure of the Lexicon was further enriched by the proponents of the most recent version of the theory of generative phonology - Lexical Phonology. (Kiparsky 1982, Mohanan 1986, Booij and Rubach 1987). In Lexical Phonology it is assumed that some of the phonological rules apply in the Lexicon rather than in the post-syntactic phono2

As pointed out in Aronoff (1976) the term "readjustment rule" employed in Chomsky and Halle (1968) is not particularly well-defined as this disparate body of grammatical structures encompasses: a. certain rules converting the surface structure generated by the syntactic component into a form appropriate for use by the phonological component b. rules which eliminate grammatical formatives in favour of phonological matrices c. other rules.

5 logical component. Some versions of Lexical Phonology impose the extrinsic ordering on lexical rules in terms of levels or STRATA (further discussion in section 3). The recognition of various grammatical regularities to be accounted for in terms of a rich body of lexical rules has rendered the very term "Lexicon" ambiguous. In the broader sense it denotes one of the components of the overall grammatical model along with the syntactic and the phonological components. In this sense it is the site of application of grammatical rules of various sorts. Accordingly, denoting a grammatical rule as "lexical" refers to its assignment to the lexical component. Lexical rules of various sorts jointly give account of the regular structural aspects of complex words. Still, in the narrower sense the Lexicon remains the repository of what is idiosyncratic and unpredictable about linguistic units. In this sense the Lexicon comprises both morphologically simple and irregular words as well as idioms.

1.2. Lexical Morphology 1.2.1. Morphological units Apart from the Lexicon the other crucial notion in modern morphological theory is the word. Spencer (1991:37) states that "each of these concepts hides a vast uncompleted research project in itself'. According to Stephen Anderson "the proper definition of the word is one of the classic chestnuts of traditional grammar" (Anderson 1985:153).3 All linguists writing on this subject stress, on the one hand, the intuitive appeal of the concept and, on the other hand, the non-existence of its satisfactory definition. 4 The chief difficulty of defining the concept "word" lies in the ambiguity inherent in its everyday usage. According to Lyons (1977) the term "word" can either refer to a lexical item (in which case we would say that run and running are two forms of the same word) or to a segmentable portion of an utterance (in which case each occurrence of run and running in an utterance constitutes a separate word). The former sense is usually referred to as LEXEME the latter as WORD-FROM. As pointed out in Bauer (1988) further disambiguation of the term word-form is necessary in as much as one word-form may correspond to more than one entities to which Bauer (1988) refers as grammatical words (morphosyntactic words in Spencer (1981). Consider the following sentences. 3 4

The most comprehensive traditional account is Kramsky (1966). See also Zirmunskij (1966).

To see this point clearly consider three quotations from the relevant literature "what constitutes a word is fairly clear intuitively but this concept has proved extremely elusive when it comes to giving it an exact definition" (Jensen 1990:1) "the word ( . . . ) elusive and barely definable at best ( . . . ) it nonetheless maintains great intuitive appeal" (Hyman 1978:445) "the word is one of the most basic and intuitively most salient of all linguistic categories of which it is notoriously difficult to give an adequate definition" (Taylor 1989:175).

6 He walked home. He has walked home. The word-form walked in both cases realises the lexeme WALK. Yet walked in these two sentences is not precisely the same element as in the first sentence walked realises WALK + past tense while in the second sentence it realises WALK + past participle (example and comment after Bauer 1988:9).5 Walked in the two sentences are different grammatical words even though they are the same word-form and realise the same lexeme. 6 It is obviously the grammatical words that constitute the chief concern of morphological theory. According to Bauer (1988) there are two major sets of criteria for delimiting the word-forms in the spoken language. 7 The first set is phonological and defines what is usually termed a phonological word while the second set comprises morphological and syntactic criteria. As is well known, at least since Pike (1967), the units as defined by the two sets of criteria not necessarily overlap (though in many languages they do in the overwhelming majority of cases). Hyman (1978) discusses the following four criteria for delimiting phonological words. a. A word can function as a unit in the statement of distributional constraints e.g. in English the sound [3] may not occur word initially. b. A word can function as a unit in the statement of phonological processes which occur at word boundaries e.g. in Sanskrit obstruents are devoiced in word final position (Gawrohski 1930). c. A word can function as a unit in the statement of accentual phenomena e.g. the dynamic stress in modern Czech falls on the word initial syllable. d. A word can function as a unit in the statement of phonological prosodies e.g. the word is the domain of vowel harmony in languages such as Turkish or Finnish. 8 There are three major morphosyntactic criteria of wordhood that seem to hold across a wide range of languages. The first is that a word is typically the smallest element of a sentence which has positional mobility i.e. it can be moved around without destroying the grammaticality of the sentence. Secondly, words are typically the largest units which resist interruption by the insertion of new material between their constituent parts. The 5

Bauer's analysis seems to be oversimplified in as much as in the second sentence WALK + past participle is realised not by walked alone but rather by has walked. 6

Lyons (1981) uses a twofold distinction: lexeme/word-form. His notion of the word-form i.e. "word

forms of a given lexeme are what would traditionally be described as its inflectional forms" (Lyons 1981:101) corresponds to Bauer's grammatical word. 7

The word in the written text is usually defined as "a unit which is bounded by spaces on both sides" (Bauer 1988:7). However the notion is made fuzzy by the use of a hyphen, as in word-form. 8 The conflicts between the criteria discussed in Hyman (1978) are not unattested. A widely quoted example is the case of compounds in modern Finnish (Karlsson 1983). Although the vowel harmony is bound by the word in compound words each component defines its own harmony span (i.e. on the strength of criterion d. compounds axe bilexical entities). The dynamic stress falls on the first syllable of the word and a single main stress falls on the first syllable of the compound (i.e. on the strength of criterion c. compounds are unilexical units).

7 third major characteristics of words is that they do not permit internal rearrangement of their constituent parts i.e. the morphemes constituting a single word have a rigidly fixed sequential order. Bauer (1988) refers to these three characteristics as positional mobility, uninterruptibility and internal stability. By virtue of what Bauer (1988:10) refers to as "the ability to recognise patterns and carry out substitutions" recurrent forms could be isolated within word-forms. Bauer (1983) terms such recurrent formal units "formatives", other names are also frequently used (e.g. morph). Furthermore, as pointed out by Lyons (1981:112) "minimal forms can be grouped into form classes on the basis of the criterion of intersubstitutability" For the purpose of this study we shall distinguish three classes of formatives: ROOTS, P R E F I X E S and SUFFIXES (affix being the traditional cover term for prefix and suffix) and we shall assume that they are morphologically simple i.e. indivisible units. Furthermore, we shall assume that another structural unit can be distinguished in every word-form: STEM. It is a traditional notion in Slavic descriptive linguistics and it denotes what Aronoff (1976:9) refers to as "word sans inflection". Says Milewski (1973:65) "in languages similar in types to the Slavic languages each word is composed of two principal parts i.e. a stem and an ending". The STEM is the common denominator of all word-forms realising the same LEXEME both in terms of their formal make-up and as "the principle carrier of lexical meaning" (Jensen 1990:1). From the structural point of view the STEM consists obligatorily of a ROOT which may be accompanied by derivational affixes. An important issue in current morphological theory is the problem of the lexical representation of morphological units. An influential line of research assumes that affixes have an autonomous lexical representation on a par with roots and stems (Lieber 1980, Williams 1981, Selkirk 1982) 9 i.e. they are endowed with semantic and grammatical representation (including) syntactic class) as well as idiosyncratic features, etc. In this work we will assume, following Aronoff (1976), that only stems and roots may have an autonomous lexical representation while affixes are introduced by morphological rules. The lexical representation of ROOT and STEM exhibit the following crucial differences. Firstly, STEM may be either morphologically simple (i.e. consist only of a ROOT e.g. the S T E M in the Polish noun walk(a) "fight") or morphologically complex (i.e. consist of a ROOT followed by suffix (es) and/or preceded by prefix (es) i.e. the STEMs in the Polish nouns: czyteln(y) "readable", rozbieralni(a) "cloakroom"). The ROOTs are, on the other hand ex definitione morphologically simple entities. 10 9

The distinction between root and stem made by Lieber (1980) and Selkirk (1982) is different from

ours and derives from the bloomfieldian distinction between free and bound linguistic forms (Bloomfield 1933). 1 0 Schultink (1988) quotes the following definition of a root from Doke (1935) "root is the irreducible element of a word, the primitive radical form without prefix, suffix or other inflection, and not admitting of analysis". Lyons (1981:112) states that "the difference between roots and stems is that roots are morphologically unanalysable, whereas stems may include, in addition to their root, one or more derivational affixes".

8 Secondly, ROOTs are purely formal, urn-lateral entities. A ROOT, strictly speaking, is merely a sequence of phonological segments listed in the lexical component and marked as ROOT. ROOTs are unmarked for syntactic and devoid of semantic representation. On the other hand, STEMs are bilateral entities, formal units endowed with semantic and grammatical representation. The properties of a STEM listed as part of its lexical entry include: a. its syntactic class b. its phonological representation c. its semantic representation d. all the relevant grammatical information: the specification for the values of relevant grammatical categories, the assignment to inflectional paradigms, et. e. all the idiosyncratic features (e.g. [+ LATINATE]). It will be clear now, that the lexical status of a formative such as [valk] (recurrent in various grammatical words realizing the lexeme walk(a) "fight" as well as in grammatical words realising the lexeme walczyc "to fight") is ambiguous. On the one hand it is a ROOT [valk]R on the other hand it is a nominal stem [valk]Ν and a component part of the verbal stem [[valk] -f i] V. Notice, that it is not apparently clear whether the two STEMs are related directly (e.g. it could be claimed that the verbal stem is derived from the nominal stem) or whether they are both derivationally related to the ROOT. This issue is discussed at some length in chapter 3.

1.2.2. Morphological rules Apart from the list of lexical units (roots and some stems) the Lexicon contains the set of grammatical rules which purport to characterise all the well-formed morphological structures of a language. In our study we will follow Aronoff (1976) in making the assumption that morphological rules (alternatively referred to as morphological processes) which form a proper subset of such rules along with lexical phonological rules fall into two categories. a. morphological rules proper, which correspond to Aronovian Word Formation Rules but on the adoption of the strong version of the Lexicalist Hypothesis also include rules of inflection. By virtue of such rules STEMs are built out of ROOTs or other STEMs while grammatical words in the sense of Bauer (1988) are built out of STEMs. b. readjustment rules, with two basic subtypes distinguished in Aronoff (1976), namely the rules of ALLOMORPHY and the rules of TRUNCATION. Morphological rules proper consist of the BASE 11 (i.e. all the morphological units that are subject to a given rule) the FORMAL OPERATION as well as (whenever possible) the SEMANTIC PARAPHRASE. 11

One important terminological and notional difference must be borne in mind: the distinction between the BASE as a group (class or set) and a base as an individual item member of the BASE referred to as a HOST.

9 A morphological rule proper should also fully determine all the grammatical properties of the output (syntactic class and other syntactic characteristics, the values of relevant grammatical categories, etc.). Aronoff (1976) claimed that the BASE must contain morphological units of one type only, namely STEMs (word sans inflection), members of major syntactic categories. We shall argue in chapter 3 that morphological rules whose BASES consist of ROOTs should also be recognised. Since ROOTs are purely formal units devoid of either semantic or grammatical representation it is only stem-based morphological rules that determine both semantics (by virtue of the SEMANTIC PARAPHRASE) and grammar of their outputs, Root-based morphological processes, on the other hand, merely account for the recurrent formal relationships between word forms. The advantages of this distinction become apparent in chapter 3. The FORMAL OPERATIONS employed by morphological rules vary considerably across world languages (for well exemplified surveys see Bauer (1988) and Jensen (1990)). In the present work we shall be concerned with one type of formal operation only: suffixation, e.g. an attachment of a morphologically simple unit to the right of the base. We adopt the notion of the SEMANTIC PARAPHRASE, an indispensable component part of any genuine stem-based morphological rule proper from Aronoff (1976). Aronoff (1976) argued that "the meaning of the output will always be a function of the meaning of the base" and pointed out that "the meaning of a W F R ["morphological rule proper" in our terms] is represented by a paraphrase containing a variable" (Aronoff 1976:50). Aronoff (1976) exemplified the notion of the SEMANTIC PARAPHRASE as follows "the agentive occupational suffix # er can be roughly paraphrased as: V # er Ν one who Vs habitually, professionally". We repeat the definitions of the two types of readjustment rules after Aronoff (1976). A rule of allomorphy adjusts the shape of a designated morpheme (or class of morphemes) in the immediate environment of another designated morpheme or morpheme class. A truncation rule deletes a designated stem-final morpheme before a designated suffix.

1.2.3. T h e scope of morphological generalizations One characteristic feature that distinguishes morphology from syntax is the fact that the overwhelming majority of the entries defined as well-formed by the rules of syntax are entirely regular and predictable while the individual characteristics of complex words are very often idiosyncratic. Morphologically complex words that exhibit idiosyncratic features must be listed (it seems to be the simplest way of recording their lexical idiosyncrasy) but it is not to say that they axe irrelevant for the operation of morphological rules. Usually, complex word-forms are aggregates of both the entirely unpredictable idiosyncratic properties as well as the predictable regular properties, which are best accounted for by means of grammatical rules. The entirely predictable complex words must not be listed.12 12

T h e significance of fully predictable complex words is highlighted in Aronoff (1988).

10 Listing and morphological rules are two theoretical constructs which respectively account for idiosyncratic (i.e. unpredictable) and regular (i.e. predictable) properties of complex words. Their mutual relationship can be summarised as below. morphologically simple words morphologically complex words with idiosyncrasies entirely regular complex words

MECHANISM listing listing and rules rules

1.3. The Lexical Phonology The following major tenets of Lexical Phonology are relevant for the present work. All morphological rules apply in the lexical component. This follows directly from Mohanan's statement "All grammatical rules which have access to morphological information apply in the Lexicon" (Mohanan 1986:18). Lexical Phonology incorporates the AFFIX ORDERING GENERALIZATION (discovered by Siegel 1974/79, christened by Selkirk 1982) and extends it to non-derivational morphology. AFFIX ORDERING GENERALIZATION claims that English derivational suffixes fall into two groups. CLASS I suffixes - which in Chomsky and Halle (1968) are preceded by a morpheme boundary in their lexical representation (e.g. + ity). CLASS II suffixes - which in Chomsky and Halle (1968) are preceded by a word boundary in their lexical representation (e.g. # ness). Siegel (1974/79) has substantiated the parallel distinction for English prefixes. The crucial insight of her influential work is that affixes in the two groups consistently differ with respect to two logically independent characteristics: the stress assignment and the ordering of attachment. Morpheme boundary affixes are stress-shifting i.e. they are able to alter the stress-pattern in their hosts. Word boundary affixes are stress-neutral i.e. they leave the stress pattern of their hosts intact. If both morpheme boundary and word boundary affixes attach to a given stem, then the former always precede the latter. For example, in a sequence of derivational suffixes morpheme boundary suffixes are never to be found following the word boundary suffix(es).13 The empirical observations are accounted for by Siegel (1974/79) in terms of rule ordering. The AFFIX ORDERING HYPOTHESIS - all morphological processes of affixation that consist in attaching a 13 The third difference between morpheme boundary affixes and word boundary affixes concern their distribution i.e. only the former can attach to bound morphemes while the attachment of the latter is confined to free morphemes. This distinction is irrelevant for our purposes since the present study is explicitly confined to the examination of stem-based suffixation and all English stems are free morphemes.

11 morpheme boundary affix are ordered before all morphological processes that consist in attaching a word boundary affix (there is no further extrinsic ordering within the two groups). The phonological rules of stress assignment axe ordered in between the two blocks of morphological rules. Notice that the latter proposal automatically leads to the simplification of the lexical representation of English affixes as it makes the boundary symbols superfluous (Strauss 1979). The two pivotal tenets of Siegel's model are taken up and extended by proponents of Lexical Phonology which results in the further enrichment of the lexical component. First, the number of blocks of morphological rules is multiplied to three or even four. Kiparsky (1982a) Mohanan (1986), Halle and Mohanan (1985) BLOCK 1 morpheme boundary derivation and irregular inflection BLOCK 2 word boundary word boundary derivation and derivation compounding BLOCK 3 regular inflection compounding BLOCK 4 regular inflection In Lexical Phonology all morphological rules are subject to the BRACKET ERASURE CONVENTION: morphological brackets are erased at the end of each block of rules (now called STRATUM). The BRACKET ERASURE CONVENTION stipulates that the morphological structure of complex words created by morphological rules at STRATA earlier than STRATUM η is non-transparent to rules applying at STRATUM n. Furthermore, the BRACKET ERASURE CONVENTION entails that no information concerning the morphological make-up of complex words is available to non-lexical grammatical rules. Second, the idea that phonological rules may be interspersed with morphological rules is extended to segmental processes. All phonological rules fall into two blocks: the lexical rules, which interact directly with the morphological processes in the Lexicon and the post-lexical rules, which apply outside the Lexicon to the outputs of the syntactic component. The lexical phonological rules are subject to the following restrictions: at least some of them comply with the Strict Cyclicity Principle (Mascaro 1976, Halle 1978, tested against a large body of data in Rubach 1984) i.e. they apply in "derived environments" (Kiparsky 1973), to the outputs of earlier morphological or phonological rules; they are assigned to lexical STRATA, i.e. apply in well-defined domains only, co-extensive with the domains of morphological rules.

1.4. M e t h o d o l o g y We shall assume that linguistics is an empirical science i.e. an inquiry which strives towards advancing testable hypotheses covering some domain of observable phenomena. Furthermore, we shall adopt the well-established view that in modern linguistics the

12 domain of observable data is constituted by language structures judged as grammatical by native speakers of a language. The test of a scientific hypothesis involves in the words of a leading philosopher of science "the confrontation between its consequences and the observable experiences" (Popper quoted in Magee (1990), the confrontation of the predictions entailed by a given hypothesis concerning the occurrence of some phenomena and the actual occurrence of these phenomena. The hypothesis is judged to be empirically adequate if the class of the predicted phenomena matches exactly the class of the observed phenomena. Two specific cases of empirical inadequacy must be distinguished: - the excessive empirical weakness: the hypothesis predicts the occurrence of certain non-existent phenomena; - the excessive empirical strength: the hypothesis predicts the non-occurrence of certain existent phenomena. Testing a hypothesis, to put it in simple words, we compare what it tells us with what is the case. Crucially, it may be too weak if it tells us something which is not the case, or two strong if it does not tell us something which is the case. At its simplest linguistic morphology is the theory of the grammaticality of complex words. It strives to characterise the notion "grammatical complex word" in the same sense in which rules of syntax were originally purported by N. Chomsky to define the notion "grammatical sentence". If the structure of complex words is conceived in terms of sequences of formatives then hypotheses that make up the morphological theory are general statements about the grammatical sequences of formatives. Such hypotheses predict that only certain sequences of formatives are grammatical and the class of the predicted grammatical sequences should coincide with what actually are grammatical sequences. Only in such case is the hypothesis at hand held to make correct empirical predictions i.e. to be empirically adequate. However, a morphological hypothesis could be empirically inadequate in either of two ways. The class of the predicted grammatical sequences could be broader than the class of the factually occurring grammatical sequences. The hypothesis is empirically too weak as it predicts the existence of certain unattested phenomena. The class of the predicted grammatical sequences could be narrower than the class of the actually occurring grammatical sequences. The hypothesis is judged to be empirically too strong as it rules out the existence of some attested phenomena. Morphological rules are generalizations about morphological facts i.e. well-formed complex morphological structures. We may call them first order generalizations. Modern linguistic research brings out the significance of the generalizations of the second order i.e. generalizations about grammatical rules (usually referred to as conditions or constraints on rules). Generalizations of the second order presuppose the prior formulation of rules, which with regard to them play the role of factual phenomena. Constraints on rules should be evaluated as empirical hypotheses against the relevant body of empirical data i.e. grammatical rules.

13 The relationship between first and second order generalizations in grammatical theory and the respective range of data is summarised in the table below. THE HYPOTHESIS a grammatical rule a constraint on grammatical rules

THE RELEVANT EMPIRICAL DATA grammatical language structures grammatical rules

1.5. Constraints on rules From what we say above it follows that we shall adopt the well-established distinction between grammatical rules and constraints on such rules. For PRACTICAL purposes (order of presentation, ease of exposition, etc.) we could distinguish rule-internal and rule-external constraints. No theoretical significance is attached to this distinction here. Rule-internal constraints concern the structure of the rule of a given sort and in the case of morphological rules they can be further divided into constraints on the BASE and constraints on the OPERATION (as pointed out above the BASE and the OPERATION are the principal components of all morphological rules). Rule-external constraints encompass the whole gamut of factors which have a bearing on the functioning of rules within the broader framework of an overall grammatical model. By way of example we could mention three types of rule-external constraints relevant for morphological rules: - extrinsic ordering (e.g. the A F F I X ORDERING GENERALIZATION i.e. the bifurcation of all morphological rules of derivation into blocks on the basis of non-morphological criteria), - constraints on syllable structure (which, as shall be argued in chapter 2 may have either the static implementation as conditions on the Underlying Representation or the dynamic implementation as rules), - various interactions with the permanent lexicon (e.g. Blocking of Aronoff 1976 or Synonymy Constraint of Kiparsky 1982).

Chapter II Constraints on stem-based suffixation in English — the predictions of lexical phonology 2.1. Inflection 2.1.1. In this section attention is focused on the following inflectional suffixes in modern English: two instantiations of what Zwicky (1975) refers to as Dental Suffix (i.e. Preterite Tense and Past Participle) and three instantiations of what he calls Sibilant Suffix (i.e. Third Person Singular, Plural, Saxon Genitive). The Dental Suffix exhibits three allomorphs: [t] after stem final voiceless segments with the exception of [t] [id] after stem final [t] or [d] [d] elsewhere. The Sibilant Suffix also exhibits three allomorphs: [s] after stem final voiceless segments with the exception of [s], [f], [tf] [iz] after stem final sibilants and affricates i.e. [s], [z], [J], [3], [tj], [d3] [z] elsewhere. The crucial analytical problems to be resolved are the following: 1) Is there a common phonological representation of each morpheme underlying its surface variants and if so what is it? 2) What rules govern the distribution of the allomorphs? In particular, are they lexical, morphological or phonological generalizations? 2.1.2. Important affinities between regular verbs and some irregular verbs should not be overlooked. There is a subgroup of irregular verbs (henceforth we shall refer to them as semi-regular verbs) with the following two characteristics. 1) The Preterite Tense and the Past Participle forms of these verbs invariably end with dental stops [t] or [d]. 2) Alternations which some of the stems in this subgroup exhibit occur in the Preterite Tense and the Past Participle forms and can be accounted for by means of independently motivated phonological rules (such as Cluster Shortening, Regressive Unvoicing, Vowel Shift). Furthermore, there are two parallels in morphological structure between semi-regular and regular verbs which set them apart from the remaining irregular (umlaut) verbs. 1) The Preterite Tense and the Past Participle forms are invariably homophonous, whereas umlaut verbs often have three distinct forms. 2) The Preterite Tense and the Past Participle forms in both regular and semi-regular verbs could be analysed as having the following structure: STEM + SUFFIX (the SUFFIX is the exponent of the Preterite Tense or the Past Participle). With the umlaut

15 verbs, on the other hand, it is the stem internal modification of (predominantly vocalic) segments that serves as the exponent of the categories at hand. Morphological analysis should reflect the difference in morphological structure between cases such as rubbed (regular) and kept (semi-regular), on the one hand, and sang (umlaut), one the other. Namely, the two surface forms of the stem keep: [kep] and [ki:p) should be reduced to the common underlying representation as allomorphs of the same lexical formative. In fact both allomorphs at hand could be derived from such common underlyer //kep//via independently motivated phonological rules: [ki:p] by means of Vowel Shift and [kep] by means of Cluster Shortening. [saeq] on the other hand, is not a morphologically simple form: the vowel [ae] is the exponent of the Preterite Tense while the consonants which flank it jointly form the exponent of the stem. An analysis (e.g. that of Halle and Mohanan 1985) attempting to reduce the two word-forms [sir)] and [saeq] to common underlying representation proceeds on the unwarranted assumption that they are phonologically conditioned variants of the same formative. 2.1.3. For ease of reference all semi-regular verbs can be divided into the following subgroups (1) a. put : put; hit, set, bet, burst, cut, etc. b. bleed : bled; hide, lead, chide, etc. c. lend : lent; build, bend, etc. d. leave : left; e. keep : kept; creep, sweep, sleep, f. feel : felt; kneel, mean, deal, dream, etc. If we assume that in all subgroups the exponent of the Preterite Tense and the Past Participle is one of the allomorphs of the Dental Suffix then the above mentioned affinities between regular and semi-regular verbs could be accounted in a straightforward manner. The two important and highly contentious issues are the following: 1) Is the underlying representation of the Dental suffix uniform throughout (la) to (If)? 2) Is it homophonous with the underlying representation of the Dental Suffix for the regular verbs? The members of subgroup (If) have a direct bearing on the underlying representation of the Dental Suffix in semi-regular verbs. Notice, that here the Dental Suffix exhibits the unvoiced allomorph [t] in a clearly nondevoicing context ( after sonorants)). This may be treated either as a case of allomorphy at the underlying level or as a result of the application of a lexically-governed rule of unvoicing. The primary disadvantage of stipulating allomorphy at the underlying level is that a generalization is missed the regular shape of the Dental Suffix (i.e. / / d / / ) and the putative irregular //t//are strikingly similar in their feature composition. In fact they differ in the value for one feature only. The solution invoking the notion of the lexically governed allomorphy rule is more restrictive: allomorphs that are phonologically unrelated cannot be accounted for in this manner. It is not an unreasonable assumption that various instances of

16 allomorphy such as / d / : / t / rubbed : felt versus //z//: //Vn//dogs : oxen differ in the degree of regularity, which is best captured by positing lexically governed allomorphy rule in the former case and allomorphy at the underlying level in the latter (Rubach 1982). Thus we can tentatively assume the existence of a lexically governed allomorphy rule of LEXICAL UNVOICING whereby the Dental Suffix changes the specification for the feature VOICE from [+] to [—]. The derivation of all members of subgroup (If) is exemplified below: (2) felt //fel+d// spelt //spel+d// fel+t spel+t LEXICAL UNVOICING felt CLUSTER SHORTENING felt spelt Given the rule of Lexical Unvoicing the derivation of the items in subgroups (Id) and (le) is straightforward since it involves the application of independently motivated phonological rules of Cluster Shortening and Regressive Unvoicing. (3)

left //lev+d// lev+t left left

kept //kep+d// kep+t kept

LEXICAL UNVOICING REGRESSIVE UNVOICING CLUSTER SHORTENING

The derivation of inflectional forms of verbs in subgroups (la) (lb) (lc) is more controversial since it involves the application of the rule of Degemination effecting the deletion of the Dental Suffix present in the underlying representation of the Preterite Tense and the Past Participle forms. The Dental Suffix (//d//) is first introduced by a morphological rule governing its distribution to be subsequently deleted via Degemination. Deletions of this kind are subjected to serious criticism in Stemberger (1981) who claims that rules of deletion of this kind do no more than recover input forms. He argues that "deletion rule creates a surface form identical to what would have been produced with no rules at all thus no evidence exists for the application of either rule" (Stemberger 1981:803) and concludes that deletion rules are undesirable and that the option of not adding an affix at all should be preferred. Notice, however, that on the "no affix attachment" solution no account is provided for the fact that all stems to which no affix is added share important phonological characteristics: they end with an alveolar stop. Likewise, contrary to Stemberger's claim it is not the case that the output forms are identical to what they would have looked like had no rules applied to them. Given the rule of Degemination we can provide a straightforward account of the stem internal alternations for which the consonant to be subsequently deleted constitutes the conditioning context. The presence of the Dental Suffix in the underlying representation of the Preterite Tense and the Past Participle forms of the verbs in (lb) and (lc) is crucial for the application of Progressive Unvoicing in (lc) (here it is preceded by Lexical Unvoicing) and Cluster Shortening in (lb).

17 (4)

lent //lend+d// lend+t lent+t

bled //bled+d// LEXICAL UNVOICING PROGRESSIVE UNVOICING CLUSTER SHORTENING DEGEMINATION

bled+d lent bled Incidentally, note that all the vocalic alternations in verbal stems involving the rule of Cluster Shortening can only be accounted for on the assumption that //d//is a separate formative for Cluster Shortening as a cyclic rule (Myers 1985)can only apply in a derived environment. Finally, Degemination is the only rule necessary for the correct derivation of the Preterite Tense and the Past Participle of the verbs in (la). Stemberger's criticism would be valid if all semiregular verbs behaved like those in (la). To sum up briefly the foregoing discussion of semi- regular verbs : 1. The Preterite Tense and the Past Participle forms of semi-regular verbs listed in ( l a ) - ( l f ) have the following morphological structure: STEM + Dental Suffix. 2. The STEM is underlyingly homophonous with whatever happens to be the underlying representation of a given verbal stem (the unmarked form: Infinitive or Present Tense). Its surface form is derived via independently motivated phonological rules. The presence of the Dental Suffix at the underlying level is crucial for the application of those rules. 3. The underlying shape of the Dental Suffix is not different from its variant that combines with "regular" stems (which we tentatively assume to be //d//). The above assumptions when accommodated within the framework of Lexical Phonology make our approach different from the treatment of verbal inflection in modern English in Halle and Mohanan (1985). In Halle and Mohanan (1985) "regular" and "semi-regular" verbs' forms of the Preterite and the Past Participle are derived at different STRATA in the Lexicon. All inflectional forms of "regular" verbs are processed at STRATUM 4 (along with all "regular" inflections) while semi-regular verbs are derived at STRATUM 1 (sharing the lot of all "irregular" inflections). This distinction is made explicit on both the phonological and the morphological side. 1) The "regular" underlying shape of the Dental Suffix (probably //id// though Halle and Mohanan (1985) are not explicit on this point) is distinct from its "irregular" shapes: //t// and //d// (//t// attaches to verbs in (la), (lc), (Id), (le), ( I f ) while //d// to those in (lb)). What is a single morpheme allegedly exhibits three way allomorphy at the underlying level. It makes the morphological account of the Dental Suffix somewhat unsatisfactory since no relationship between what axe clearly phonologically related allomorphs of a single morpheme is expressed anywhere in the Grammar while the necessity arises of multiplying the same lexical entry of the Dental Suffix. 2) On the phonological side the distinction is well motivated since a number of rules relating stem variants of verbs apply to "semi-regular" verbs only, but never to "regular" verbs. This important difference follows automatically from the assumption that phonological rules apply only at certain STRATA coupled with the well-grounded stipulation that all rules involved in the derivation of the Preterite Tense and the Past

18 Participle forms of semiregular verbs apply at STRATUM 1, whereas the Dental Suffix is attached to "regular" stems at a later stage (STRATUM 4). Notice that Halle and Mohanan's claims summarised in 1) and 2) above are logically independent. In fact given rule ordering stipulated in 2) we can safely abandon the hypothesis about the underlying representation of the Dental Suffix advanced in 1). This hypothesis seems to be a relic from the treatment of verbal inflection in a pre-lexical framework. Then it was indeed necessary to make leaped and leapt distinct at the underlying level to ensure the correct derivation, in particular to trigger off Cluster Shortening in the latter but not in the former. Within the lexical framework we can assume that exactly the same underlying shape of the Dental Suffix is attached to verbal stems at both STRATUM 1 and STRATUM 4. There are two crucial differences between derivations at the two STRATA under consideration. The hosts of the Dental Suffix attachment constitute a closed set at STRATUM 1 and an open class at STRATUM 4 while only STRATUM 1 derivations are subject to Lexical Unvoicing, Cluster Shortening, Degemination, Regressive and Progressive Unvoicing. 2.1.4. Now let us focus attention on the crucial difference between some "semi-regular" verbs (those in subgroups (la) (lb) (lc) and an open class of regular verbs, which share an important phonological characteristics: the feature composition of the stem final segment [t] or [d]. In both instances the attachment of the Dental Suffix to a verbal stem gives rise to a word final geminate cluster eradicated at a later stage in the derivation. The two groups of verbal stems mentioned above differ crucially in that different strategies of getting rid of the awkward cluster are employed. In the case of the verbs listed in (la) (lb) (lc) the word final geminate is deleted via the rule of Degemination, while in regular verbs the geminate cluster is split up by an inserted vowel, e.g. [put] vs. [paetid]. We have here a rather clear instance of "rule conspiracy" in the sense of Kisseberth (1970) where formally unrelated rules exhibit a functional unity in that they act to eliminate the same ill-formed sequence. In the words of Lass (1974)both rules Degemination and Vowel Insertion in the case at hand "could be taken as implementations of a higher-order well formedness condition" (Lass 1974:314). The important questions to be addressed now are the following: - in what sense exactly is the geminate cluster an ill formed sequence ? - why are the two different strategies employed to resolve what is apparently the same difficulty? are there any general principles of grammatical organisation which account for the variety and peculiar distribution of the remedial rules at hand? Degemination applies at STRATUM 1. The existence of some phonological process eradicating consonantal geminates at STRATUM 1 is entailed by an important structural constraint holding for well-formed English morphemes: "There are no morphemes with a sequence of identical consonants in English (Mohanan 1986:19) coupled with a general convention calledi Structure Preservation which is formulated as follows by Borowsky (1989:148) "Language particular structural constraints holding for underlying representation hold also for derived representation and vice versa".

19 Borowsky goes on to argue that" in English the domain of Structure Preservation appears to be the phonology of the 1st Level [STRATUM 1 in our terms] only" (Borowsky 1989:149 ftn 4). Since no geminates are allowed morpheme-internally such structures must also be banned across morpheme boundary followed by a STRATUM 1 suffix (or preceded by a STRATUM 1 prefix). Given the correctness of Mohanan's empirical claim quoted above STRUCTURE PRESERVATION motivates the existence of some grammatical mechanism eradicating all consonant geminates arising in the course of derivation at STRATUM 1. Contrary to what is claimed by Borowsky (1989) the existence of the rule of Degemination in modern English does not follow directly from STRUCTURE PRESERVATION as a general convention constraining the range of available grammars. STRUCTURE PRESERVATION offers no principled reason why Degemination should be selected rather than, say, the competing rule of Vowel Insertion. It seems that given the current theoretical apparatus of Lexical Phonology it is merely the achieved result that is motivated in the case at hand but not the particular manner of achieving it. The strong constraint against consonantal geminates holding throughout all structures at STRATUM 1 (whether simple or derived) is apparently relaxed at later STRATA but not to the extent to allow geminates that arise due to the attachment of the Dental Suffix to verbal stems with stem final alveolar stops. In other words Mohanan (1986) goes too far when he claims that "Cluster Degemination does not hold across non-class 1 affixes" (Mohanan 1986:18) although his claim is correct with regard to the examples he gives: prefix (un) and two suffixes (-ness) and (-less) (all the relevant examples in Mohanan 1986:18,19,40,45) e.g. un.natural, keen.ness, wrinkl.less. Since all the geminate clusters given in Mohanan (1986) consist of sonorants, while the attachment of the Dental Suffix to verbal stems with stem final [t] [d] creates sequences of obstruents, one is tempted to restrict the relaxation of the constraint against geminates to sonorant segments only. This generalization, on further scrutiny, turns out to be incorrect, however (cf. the attachment of the prefix (dis) to stems with initial [s] or the attachment of the suffix (dom) to stems with final [d] dissatisfaction, lord.dom). What seems to be of crucial relevance is the assignment of syllable structure to complex word-forms under consideration. Consider the syllabification in all instances where the constraint against geminates is apparently relaxed. d (5)

d

d

d

d

d

Λ Λ Λ Λ .Λ Λ un natural dis similar wrinkle less In all cases such as those given above members of geminate clusters belong to adjacent syllables, whereas in putative cases such as those given below they would have to be treated as tautosyllabic. 3 θ (6) pot d rod d

20 An additional syllable to which one of the geminate consonants can be assigned is created via the rule of Vowel Insertion, which inserts a vocalic segment ([i] or [a] subject to dialectal variation) between the Dental Suffix and the verbal stem whenever the latter ends with an alveolar stop. Three points are of interest concerning this rule. Firstly, a well-known alternative solution is discussed in the relevant linguistic literature. As mentioned above the rule of Vowel Insertion presupposes //d//as the underlying representation for the Dental Suffix. Alternatively, one may assume that the Dental Suffix is represented as //Vd//at the underlying level, which necessitates positing the rule of Vowel Deletion. The latter solution goes back to Bloomfield (1933) while the former originated with Hockett (1958) and is also defended in Shibatani (1972). It is by no means easy to adduce conclusive arguments in favour of either option. As pointed out by Shibatani (1972:119). "The phonological processes involved are very simple and the question raised is straightforward yet the current theory provides no single solution. This fact poses a serious problem regarding general claims made by generative p e n o logists" . What Shibatani refers to as "the current theory" is the early classic framework of generative phonology best exemplified by Chomsky and Halle (1968). Lexical Phonology deservedly acclaimed as a much more restrictive theory should offer a principled choice between the conceivable solutions for the issue at hand. Secondly, given the rule of Vowel Insertion (or, for that matter, Vowel Deletion), its exact status as a grammatical generalization should be established, whether it is an independently motivated phonological rule or a statement of a more restricted scope and/or conditioned by non-phonological factors. Incidentally, all proponents of either of conceivable solutions seem to assume that the putative rules are fully-fledged phonological generalisations. Thirdly, ever since Bloomfield's insightful analysis of English inflection (Bloomfield 1933) it is customary to assume that rules governing the distribution of the allomorphs of the Dental Suffix also account for the variation in shape of the Sibilant Suffix. Note the following parallels in the formal variants of the two suffixes. Both suffixes exhibit syllabic and non-syllabic allomorphs: (7)

[Vd] versus [d] or [t] [Vz] versus [z] or [s]

The syllabic variants appear whenever the stem final segment is identical (or similar) with regard to the place and manner of articulation to the consonant of the suffix attached: [id] after stem final [t] [d] [iz] after stem final [s] [z] [J] [3] [tj] [d3] The syllabic/non-syllabic variation is accounted for in terms of either Vowel Insertion or Vowel Deletion. Both suffixes exhibit voiced and voiceless allomorphs in exactly parallel contexts, [t] and [s] appear after stem final voiceless segments with the exception of those that

21 require the syllabic allomorph. Since both suffixes are commonly believed to contain a voiced consonant at the underlying level the voiceless allomorphs are produced via the phonological rule of Progressive Unvoicing (to be discussed later). An alternative solution would assume that the suffixes at hand have voiceless consonants at the underlying level (//s// and / / t / / respectively) which are voiced via the rule of Progressive Voicing in the context of - voiced segments and [s] [j] [tj] for the Sibilant Suffix - voiced segments and [t] for the Dental Suffix. The context for the putative rule of Voicing can hardly be claimed to constitute a natural class, hence this particular option is not given serious consideration in the relevant literature (but cf. Hoard and Sloat 1970 criticised in Kastovsky 1980) on the grounds of the phonetic implausibility of the rules involved. On the other hand, the competing rule of Progressive Unvoicing is widely believed to have a strong phonetic motivation (Hockett 1958, recently Hockett 1987, also Shibatani 1972) and it is treated accordingly as a phonological statement. Harms (1972) goes even as far as to deny its grammatical status, since as he claims "the unvoicing of obstruents preceded by voiceless obstruents is a purely physiological process, not a rule of grammar" (Harms 1972:7). The grammatical status of the rule of Progressive Unvoicing will be considered below. First, however, let us address another issue brought up above, namely the status of the putative rules of Vowel Deletion and Vowel Insertion in the Grammar within the framework of Lexical Phonology. Recall that by virtue of Vowel Deletion the underlying representations of the Dental Suffix and the Sibilant Suffix are simplified from //Vd//and //Vz//to / d / and / z / in all contexts except /1/ / d / and / s / / z / / J / / $ / / t f / / d ^ / respectively. Clearly, Vowel Deletion is not a post-lexical phonological rule since given such status it would incorrectly predict that all sequences conforming with the following pattern are ungrammatical: (9)

X d

X: all segments except / t / / d / where

Υ ζ

Y: all segments except / s / / z / / J / / $ / /tj"/ /d^f

Such sequences are quite common both morpheme internally and across word boundaries. As a lexical rule Vowel Deletion would be assigned to at least one of the STRATA. The data given below have a bearing on this issue. (10)

a. witty hurlish

b. singing oldest

Derivations such as those exemplified in (10) are effected by stress-neutral derivational suffixes, within the model of Lexical Phonology they are located at STRATUM 2 (Mohanan 1986). Regular inflection (including both the Dental Suffix and the Sibilant Suffix as well as suffixes exemplified in (10b) is posited at a separate STRATUM (STRATUM 3 in Kiparsky (1982a), STRATUM 4 in Mohanan (1986). If Vowel Deletion is assigned to exclusively inflectional STRATUM then it is automatically made inapplicable to forms such as those given in (10a) (which is a correct empirical prediction) but free to apply to all word-forms created via the attachment of the Dental Suffix or the Sibilant Suffix

22 (correct again). The unfortunate consequence of this solution is that it makes plainly incorrect predictions with regard to the remaining English inflectional suffixes, namely (ing) and (est) which sometimes satisfy the input conditions of the rule at hand but never show traces of its application. Vowel Deletion does not operate on either (ing) or (est) regardless of the phonological shape of the final segment of the stems to which they are attached. What the data exemplified in (10b) seem to show is that the putative rule of Vowel Deletion may not be accorded the status of a phonological statement even with the provision that it operates only at one (inflectional) STRATUM in the Lexicon. What the rule at hand seems to reflect is an idiosyncratic property of several morphemes that make up a closed set: all of them are instantiations of the Dental Suffix and the Sibilant Suffix. We can tentatively conclude that Vowel Deletion is an allomorphy rule in the sense of Aronoff (1976) in that, unlike phonological rules proper, which operate on open classes of morphemes sharing certain sets of phonological characteristics, it refers to a closed set of morphemes. What makes it different from prototypical Aronovian allomorphy statement, however, is its sensitivity to phonological rather than morphological context. (Allomorphy rules as discussed in Aronoff (1976) alter the phonological make-up of certain morpheme(s) in a well-defined morphological context usually before (after) certain other morphemes). Vowel Deletion, contrary to general consensus on this issue (Anderson 1974) seems to be a phonologically conditioned allomorphy rather than a fully-fledged phonological rule. 2.1.5. In the syllabic allomorphs of the Dental Suffix and the Sibilant Suffix the consonants are invariably voiced while the non-syllabic allomorphs exhibit further variation with regard to the feature [voiced]. The latter alternation in commonly accounted for in terms of the rule of Progressive Unvoicing of Obstruents (Anderson (1974) Miner (1975)). This solution crucially rests on the assumption that both the Dental Suffix and the Sibilant Suffix have a voiced consonant (obstruent) at the underlying level, which becomes unvoiced whenever preceded by a voiceless segment (obstruent, as it must be). The rule at work here is commonly considered to be a phonological generalization per se in as much as it is automatic (Hockett 1958), (hence it allows of no exceptions or variation) as well as phonetically well-motivated (Harms 1972).It seems to be an instance of assimilation with regard to the value of feature [voiced] among adjacent obstruents, a universally well-attested phonological process. Let us now inquire into the status of the rule at hand in the model of Lexical Phonology. Is it indeed a purely phonological generalization? Is it indeed the case that English obstruents get unvoiced when preceded by a voiceless obstruent? Consider the following data. (11)

a. disdain disgust

b. that day this boy

The data exemplified in ( l i b ) show that Progressive Unvoicing of Obstruents is not a post-lexical process, since it does not obtain across word boundaries. The data exem-

23 plified in (11a), moreover, show that the process at hand does not apply morpheme internally hence it is presumably conditioned by the presence of a morphological boundary. Given the validity of the above conclusions let us pose the following question: Is it the case that obstruents get unvoiced when preceded by a voiceless obstruent across a morphological boundary? If it were indeed true then the following statements would hold: (12) a. Voiced suffix-initial obstruents become unvoiced when preceded by a stem-final voiceless obstruent. b. Voiceless prefix-final obstruents induce unvoicing of the stem-initial voiced obstruents. One can follow Anderson (1974) and Miner (1975) in imposing a further restriction on suffixes and prefixes involved in (12a) and (12b) and assume that they must be followed (preceded) by a word boundary (In Lexical Phonology such affixes are attached at STRATUM 2 and the inflectional STRATUM). Now let us look at all data bearing on generalizations given in (12a) and (12b) scarce as it admittedly is. (-dom) is the only suffix in modern English which begins with a voiced obstruent. Marchand (1960) claims, however, that "the suffix is currently very productive, though most of the words are slightly humorous and not exactly recognised as standard vocabulary" (Marchand (1960:205ff). Many of the formatives listed by Marchand (1960:205ff) exhibit stem final voiceless obstruents which, however, do not seem to induce the change of suffix initial [d] to [t] in (-dom), (13)

priest.dom prince.dom soviet, dom wife, dom horse.dom birth.dom

brute.dom pope.dom sheriff, dom duke, dom mice.dom church.dom

In modern English there are five prefixes which end with voiceless obstruents, none of which triggers off the unvoicing of the stem-initial voiced obstruent. (14)

(ex) ex-bishop, ex-boarded-out (Marchand 1960:116) (dis) dis-belief, dis-join, dis-burden (Marchand 1960:109) (mis) mis-govern, mis-behave, mis-deem (Marchand 1960:126) (trans) trans-dialect, trans-border (Marchand 1960:147) (step) step-brother, step-daughter (Marchand 1960:142)

One could still attempt to maintain the status of the Progressive Unvoicing of Obstruents as a general phonological process by restricting its application to a subset of obstruents only, minimally to [d] and [z]. Yet the data in (13) and (14) do not warrant the assumption that alveolar obstruents vary in their behaviour with respect to unvoicing from other members of the class of obstruents e.g. dis. dain vs. dis. gust, mis.govern vs. mis.deliver. Incidentally, the only suffix initial obstruent happens to be [d], a sheer but fortunate coincidence.

24 Note, that within the framework of Lexical Phonology a further restriction is possible. Apart from making the rule of Progressive Unvoicing of Obstruents sensitive to the presence of a morphological boundary we could assign it to one STRATUM only within multilayered Lexicon. At first glance it seems to accord well with the systematic difference between the morphological status of all affixes in (13) and (14), on the one hand and the Dental Suffix and the Sibilant Suffix that undergo unvoicing, on the other. The former are invariably derivational stress-neutral affixes (hence attached at STRATUM 2) while the latter are all inflectional suffixes and as such belong to a separate "inflectional" STRATUM. The assumption that the inflectional STRATUM constitutes the domain of application for the Progressive Unvoicing of Obstruents seems to be borne out by all the data considered so far. As a lexical rule it is inapplicable across word boundaries. It is conditioned by the presence of a morphological boundary, hence inapplicable to morpheme-initial clusters. By the same token it is inapplicable across morpheme boundaries preceding/following derivational affixes (this holds given the validity of the B R A C K E T ERASURE CONVENTION). It applies only to complex word forms generated by the attachment of various instantiations of the Dental Suffix and the Sibilant Suffix at a separate inflectional STRATUM. Thus, we could treat the Progressive Unvoicing of Obstruents as a phonological rule assigned to the inflectional STRATUM. Were it the only solution available it would in fact provide strong evidence in favour of two important constraints on grammatical rules inherent in the basic tenets of Lexical Phonology (Mohanan 1986): the existence of a separate STRATUM for regular inflectional processes in modern English (as it is a highly contentious issue, all evidence is welcome) as well as the validity of the B R A C K E T ERASURE CONVENTION. One can, however, equally well consider the Progressive Unvoicing of Obstruents to be a phonologically conditioned allomorphy rule valid for the Dental Suffix and the Sibilant Suffix only. The issue of the grammatical status of the generalization under consideration remains unresolved for the time being. Recall that in this respect the Progressive Unvoicing of Obstruents parallels the putative rule of Vowel Insertion discussed earlier. Both generalizations admit of two interpretations either as phonological rules that belong to the inflectional STRATUM or as phonologically conditioned allomorphy statements valid for a couple of suffixes only, which sets them apart from the putative rule of Vowel Deletion. The latter, as shown earlier, can only be interpreted as an allomorphy statement. In view of this descriptive indeterminacy it would be of no inconsiderable interest to adduce evidence disqualifying one of what now seem to be two equally admissible interpretations of the status of the rules under consideration. Such evidence comes from the consideration of one instantiation of the Sibilant Suffix - the so-called Saxon Genitive. 2.1.6. Linguists of various theoretical persuasions agree that the Saxon Genitive exhibits markedly different grammatical behaviour in comparison to the remaining two instantiations of the Sibilant Suffix: Plural and Third Person Singular (Vachek 1976, Zwicky

25

1975). Although the details need not concern us here one indisputable fact is, however, of paramount importance for our analysis - the morphological process of the attachment of the Saxon Genitive morpheme must apply outside the Lexicon. It is on this assumption only, that we are able to account for what Zwicky (1986) aptly refers to as "promiscuous attachment" of the Saxon Genitive suffix, namely its ability to join the last element of virtually any nominal phrase embedded in a non-head position of a nominal phrase. The Saxon Genitive morpheme, in other words, frequently appears in syntactic configurations with the following structure: NP (15) \ P [Saxon Genitive] Ν e.g. the man I saw yesterday's hat (the celebrated example from Bloomfield 1933). For the historical development of this interesting phenomenon known as "group genitive" see Jespersen (1909-1949)). Since syntactic configurations conforming with the above pattern arise in the syntactic component the Saxon Genitive suffix must be free to join them outside the Lexicon. This, in turn, forces us to assume that rules governing the distribution of the various allomorphs of the Saxon Genitive morpheme must also be available outside the Lexicon. Hence, we no longer need to consider seriously the possibility that they are phonological rules assigned to the inflectional STRATUM, which, automatically, gives priority to the other conceivable interpretation, namely that both the rule of Progressive Unvoicing of Obstruents and the rule of Vowel Insertion are phonologically conditioned allomorphy statements valid only for the Dental Suffix and the Sibilant Suffix. Notice, however, that the well-grounded claim that both rules under consideration apply outside the Lexicon (more precisely, after the SYNTACTIC component since it is only there that the Saxon Genitive morpheme on which they operate becomes available) entails, in turn, the stipulation that, contrary to one of the basic tenets of Lexical Phonology (Mohanan 1986), not all grammatical rules sensitive to morphological information apply in the Lexicon. In the case at hand we seem to have to do with two extra (post)lexical allomorphy statements. Furthermore, viewed as post-lexical allomorphy rules both Progressive Unvocing of Obstruents and Vowel Insertion provide disconfirming evidence against the BRACKET ERASURE CONVENTION. Notice that both rules apply to Plural Suffix and Third Person Singular Present Tense Suffix, which, unlike the third instantiation of the Sibilant SufBx, are, for lack of evidence to the contrary, supposed to attach in the Lexicon. The BRACKET ERASURE CONVENTION must now be prevented from wiping out the morphological identification of the [z] segment in all cases where it represents Plural or Third Person Singular Present Tense; otherwise the allomorphy statement may not be expected to apply. This is not the end of troublesome data provided by the Saxon Genitive suffix. Note that the attachment of this morpheme is subject to a very peculiar restriction. Consider the following data: (16) a. the two kids' ideas ['kidz] *['kidziz] the two children's ideas ['tfildranz] the two women's ideas ['wimanz]

26 b. anyone who hurries' ideas [ΊΙΛΓΙΖ] *['hAriziz] anyone who is hurrying's ideas ['ΙίΛτπηζ] anyone who hurried's ideas ['hAridz] c. a friend of my children's ideas ['tfildranz]*['tj'ildr3nziz] Apparently the otherwise "promiscuous" Saxon Genitive suffix systematically fails to attach in certain contexts. First of all, note, that the generalization at hand is not of a phonological nature, it is N O T an instance of HAPLOLOGY - the allegedly universal constraint, whereby morphemes fail to attach to hosts which end with sequences of segments (morphemes) homophonous with the attaching morpheme (Dressier 1977, Stemberger 1981). Admittedly, variation presumably governed by Haplology constraint is well attested in many varieties of contemporary English and many, if not all, nouns with stem final [s] or [z] may be subject to this constraint. Consider the data in (17) below: (17) Fox' tail ['foks teil] or ['foksiz 'teil] Lass' contention ['laes kan'tenjan] or ['laesiz kan'tenjan] Now, with regard to (16) as compared to (17) note the following two points. First, linguists differ in their interpretation of the data in (17). Stemberger (1981) treats it as an instance of dialectal variation, while Zwicky (1975) explains the first variant in each case in (17) as a spelling pronunciation. Second, an all-important difference between cases exemplified in (16) and (17) is that with nominal stems ending with an alveolar fricative the haplology constraint is merely optional, whereas the Saxon Genitive's failure to attach in cases such as those in (16) is systematically obligatory. Two other possibilities of restricting the scope of the Saxon Genitives attachment should be dismissed. Examples in (16a) show that the rule distributing Saxon Genitive's failure to apply (note, incidentally, a fine instance of "group genitive" in the preceding clause) cannot be attributed to the fact that the host contains the feature [PLURAL]. It is rather the presence of merely one of the bearers of this feature (the regular exponent of plurality - the Sibilant Suffix) that inhibits the attachment of the Saxon Genitive morpheme. The latter is allowed when the host contains other exponents of plurality. The data in (16) imply that the rule at hand's non-application is not due to the presence of a suffix boundary. The latter has no inhibitive effect on the Saxon Genitive attachment whenever preceded by morphemes other than the Sibilant Suffix. All in all, we can conclude that the attachment of the Saxon Genitive morpheme is obligatorily inhibited whenever the host ends with an instance of the Sibilant Suffix: Plural, Third Person Singular Present Tense or Saxon Genitive as exemplified in (16a), (16b), (16c). This conclusion has again vital bearing on one of the basic tenets of Lexical Phonology: the B R A C K E T ERASURE CONVENTION, since it forces us to assume that it is suspended in the case of Plural and Third Person Singular Present Tense. 2.1.7. Let us briefly summarise the foregoing discussion of the Sibilant Suffix. 1) We argued on the basis of highly compelling evidence that the Saxon Genitive morpheme is attached outside the Lexicon.

27 2) In view of the fact that no such evidence exists with regard to the other instantiations of the Sibilant Suffix we assume that both Plural and Third Person Singular Present Tense are attached in the Lexicon. 3) The two generalizations that account for the distribution of the allomorphs of English inflectional suffixes i.e. Progressive Unvoicing of Obstruents and Vowel Insertion (or possibly Vowel Deletion) are phonologically conditioned allomorphy rules and they apply only when ALL the three instantiations of the Sibilant Suffix are available, that is outside the Lexicon. 4) The Saxon Genitive morpheme fails to attach whenever the potential host ends with any instantiation of the Sibilant Suffix (Plural, Third Person Singular Present Tense or Saxon Genitive). Given the validity of the stipulation in 2) above the claims inherent in 3) and 4) run contrary to one of the basic tenets of Lexical Phonology - the BRACKET ERASURE CONVENTION. Note that both 3) and 4) imply that grammatical rules posited outside the Lexicon must have access to morphological structure created in the Lexicon. We are faced with the dilemma of either admitting exceptions to the BRACKET ERASURE CONVENTION or abandoning the stipulation given in 2) above i.e. A. positing the processes of Plural and Third Person Singular Present Tense attachment outside the Lexicon. Note that in the case of 3) above we can also attempt to show that B. the rules in question are not allomorphy statements after all. We focus on the options spelled out in A. and B. above in sections 1.8. and 1.9. respectively. 2.1.8. There are several reasons for assuming that Plural and Third Person Singular Present Tense suffixes are attached in the Lexicon. Firstly, the grammar of contemporary English contains a well-known lexically-governed allomorphy rule, by virtue of which stem final voiceless fricatives become voiced when followed by at least three different suffixes: - the denominal adjectivizing suffix [i] e.g. worth : worthy - a denominal verbalizing suffix e.g. house: to house, proof: prove - the plural Sibilant Suffix e.g. knife: knives The rule in question runs as follows: (18) Fricative Voicing

As a statement conditioned by both lexical and morphological factors Fricative Voicing can be safely assumed to belong to the Lexicon. Notice, however, that its application in some cases (see the examples above) crucially hinges upon the prior attachment of the plural suffix. As all rules preceding a lexical rule are, of necessity, part of the Lexicon as well, we can conclude that given the rather uncontroversial assumption that Fricative Voicing is a lexical rule it follows that the Plural suffix attachment takes place

28 in the Lexicon as well. Secondly, an interesting historical development could be nicely accounted for, if we assume that the attachment of the Plural suffix, unlike that of the Saxon Genitive takes place in the Lexicon. Jespersen (1909-1949) VI provides ample evidence that the Saxon Genitive morpheme used to constitute the conditioning context for the rule of Fricative Voicing. "The same alternation between [f, Θ, s] and [ν, Ö, z) that is found in the Plural also occurred in the genitive singular" (Jespersen 1909-1949 VI:264). According to Jespersen [f] was regularly spelled as (u) whenever it preceded the Saxon Genitive morpheme till the early seventeen century. This peculiar spelling often corrected in modern editions of writers such as Chaucer, Malory or Shakespeare reflected the voiced pronunciation of [f] in this particular context. Consider the following examples (after Plank (1985)): (19) for my wiues seke my liues counsell a kniues point According to Jespersen (1909-1949) VI two diachronic developments coincided in time (early seventeen century): the Saxon Genitive Suffix ceased to trigger off the rule of Fricative Unvoicing while, simultaneously, the syntactic phenomenon known as "group genitive" arose i.e. from then on the Saxon Genitive suffix has attached freely to nominal phrases rather than to single nominal stems. "From the early seventeen century the genitive ending is often added to a whole group of words instead of the word that might, perhaps, logically expected to be genitive" (Jespersen (1909-1949) VI:281). Some early instances of the group genitive are given in Plank (1985:218) e.g. my wife and children's ghosts. This peculiar coincidence can be nicely accounted for, if we assume that sometime in the early seventeen century the lexical process of the Saxon Genitive suffix attachment moved out of the Lexicon. The displacement reflecting its newly-acquired ability to join nominal phrases entailed its failure to trigger off the lexical rule of Fricative Unvoicing. Admittedly, arguments discussed above concern solely the Plural suffix, while the grammatical status of the process governing the attachment of the Third Person Singular Present Tense remains indeterminate. Notice, however, that the assumption that even one instantiation of the Sibilant Suffix is attached in the Lexicon coupled with the, by now familiar, constraint restricting the attachment of Saxon Genitive constitute evidence against the universal validity of the BRACKET ERASURE CONVENTION. 2.1.9. Various allomorphs of the Sibilant Suffix and the Dental Suffix are accounted for in terms of two grammatical rules: the Progressive Unvoicing of Obstruents and either the Vowel Deletion or the Vowel Insertion. We argued above that these rules exhibit two logically independent characteristics: 1) They are allomorphy statements; of all the morphemes in the English language they are valid merely for those subsumed under the Dental Suffix and the Sibilant Suffix. 2) They apply outside the Lexicon.

29 Recall that this particular configuration of characteristics calls into question one of the basic constraints Lexical Phonology puts on grammatical rules: the BRACKET ERASURE CONVENTION, as the latter, in principle, does not allow extra-lexical rules to have access to morphological structure created in the Lexicon. The Bracket Erasure Convention, in other words, seems to make incorrect predictions with regard to the extra (post) - lexical allomorphy processes under consideration. However, instead of concluding that we have discovered another piece of evidence against the BRACKET ERASURE CONVENTION we should perhaps reconsider the arguments advanced in favour of the claims inherent in 1) and 2) above. As there is little point in questioning the compelling evidence which led us to the conclusion that the rules in question apply outside the Lexicon we shall, once again, focus attention on the other assumption-that concerning the grammatical status of the rules under consideration. First of all, notice, that two striking features set both rules apart from prototypical allomorphy statements: they apply outside the Lexicon (while it is the lexical component where all of the well-described allomorphy rules belong, Lieber (1982) Rubach (1984)) and they are conditioned by phonological factors (while allomorphy rules as described by Aronoff (1976) and exemplified by others apply to morphemes in a well-defined context of other morphemes). While arguing that the Progressive Unvoicing of Obstruents is an allomorphy statement we pointed out that the rule fails to apply to voiced obstruents (including [d] and [z]) preceded by voiceless obstruents, unless the former are instances of the Sibilant Suffix or the Dental Suffix. Although I can see no reason to doubt the accuracy of this observation, still nothing, in fact, rules out the possibility that purely phonological factors are at play here. It may very well be the case that the rules apply, to instances of the Dental Suffix and the Sibilant Suffix while at the same time failing to apply to obstruents in the initial position of stems and derivational suffixes merely because the former share certain phonological property which the latter consistently lack. Consider, again, some of the relevant data: , . mis.behave step.daughter ^ dis.believe priest.dom In all the cases such as those given above the rule of Progressive Unvoicing of Obstruents would have to apply across the syllable boundary, since the conditioning obstruent and the obstruent undergoing unvoicing belong to adjacent syllables. The situation is different as far as all instantiations of both the Sibilant and the Dental Suffix are concerned, as they invariably belong to the same syllable with the preceding obstruents which condition unvoicing. This may well be the relevant phonological property of the Dental and the Sibilant Suffix which sets them apart from those obstruents that fail to undergo unvoicing in an otherwise parallel phonological context. The phonological generalization at work here may simply concern tautosyllabic obstruents and say: (21) Adjacent tautosyllabic obstruents must agree in the specification for the feature [voiced].

30 The above restriction seems to be implemented by grammatical generalizations of two kinds: - STATIC (of a kind analogous to Morpheme Structure Conditions): Well-formed lexical representations conform to the following condition: d (22)

-(-obstruent α voiced

^

+obstruent a voiced

- DYNAMIC: rules. The Progressive Unvoicing of Obstruents may be one of the conceivable options here. d [+obstruent] —* [-voiced] / [—voiced]— By virtue of the above rule voiced obstruents become unvoiced when they join a syllable that end with a voiceless obstruent. The rule of Progressive Unvoicing of Obstruents can now be viewed as a phonological statement operating at a post-lexical level. Notice that it is not necessary to stipulate that the rule involves morphological information of any kind. The confirming evidence in favour of the above interpretation of Progressive Unvoicing of Obstruents comes from the consideration of the weak forms of two auxiliary verbs: "has" and "is". As is well known they exhibit the following variation in rapid/casual speech: [hsz] : Μ

/ ζ / /β/

[ιζ] : / « / / ζ / / s /

Whenever the vowel is dropped the obstruent joins the preceding syllable undergoing obligatory unvoicing if only the syllable final obstruent is unvoiced. John has done it.

Bob has done it.

/bob 3

d / \ /dgaek a z/ /d3aek s/ The process of unvoicing evidently at work here could be accounted for by means of the rule of Progressive Unvoicing of obstruents provided the latter is a post-lexical phonological statement relevant for tautosyllabic sequences only. Note that the rule at hand, although a post-lexical generalization, is precluded to apply across word boundaries as long as they coincide with syllable boundaries. The latter stipulation is borne out by all available data e.g. that day, this zone, etc. It remains to be seen whether the other process governing the distribution of the allomorphs of English inflectional suffixes (Vowel Insertion or Vowel Deletion) could also be reinterpreted as a phonological rule, perhaps along the same lines. This indeed seems to be the case in as much as the following generalization is true: Jack has done it.

31

In modern English the following configurations of segments are ill-formed: d (24)

+obstruent +coronal a continuant

+obstruent +coronal a continuant

In plain words: (25) The adjacent tautosyllabic coronal obstruents must not agree in the manner of articulation. The above formulation presupposes the bisegmental interpretation of affricates [t$] and [d3], which given the apparatus of the current phonological theory is formalised as two "melodies" attached to a single "timing slot" (Clements and Keyser 1983) X

X

Μ [fl [d] [3] The constraint given in (25) above refers to the melody tier and rules out the following sequences of adjacent tautosyllabic segments d d d d (26)

[t] [d] [t] [t] [d] [d] [d] [t] d d d

d

Η Μ Μ Μ LH Μ [3] W (The sequences with [s] [J] [3] as the second element Eire non-attested). In a manner parallel to (21) generalization (25)could be implemented both as a constraint on the underlying representation and as phonological rules. As a matter of fact, it may provide justification for several conceivable phonological rules in modern English at least some of which must be assumed to exist. For example, we may hold it responsible for the rule deleting the Dental Suffix in semi-regular verbs, which, as argued earlier, is also motivated by, possibly related, conditions on permissible geminate clusters. The constraint (25) is apparently at work in the case of optional Vowel Deletion, the rapid/casual speech phonological rule, which governs the distribution of syllabic/non-syllabic variants of the weak forms of auxiliary verbs "has" / "is". As is well-known the vowel in those verbs is never dropped, regardless of the tempo or degree of casualness of speech involved, if it is preceded by any of the following segments: [J] [3] [s] [z] (it will be remembered that [j], [3] may be second segments in affricates). The sentence such as "John has done it" may have two different pronunciations in rapid casual speech: /d3onaz/ /d3Dnz/ but only one when the same statement refers to George: /d30:d38z/. In the latter case the rule of Vowel Deletion is blocked. In general, Vowel Deletion is blocked in rapid speech in those contexts only where it would generate tautosyllabic sequences of obstruents banned by condition (25). Hence, we can assume it is constrained precisely by this condition. Some "dynamic" implementation of constraint (25) is also apparently

32 at work in the case of English inflectional endings. Note, however, that the two conceivable options (Vowel Insertion and Vowel Deletion) are equally well-justified on the basis of constraint (25). Vowel Insertion inserts a vowel, whenever the Dental Suffix (//d//) or the Sibilant Suffix (//z//) is preceded by [t] [d] or [s] [z] [J] [3] respectively. Thus, it eliminates ill-formed, tautosyllabic sequences of obstruents by splitting them up and creating an additional syllable. Vowel Deletion deletes the underlying vowel, unless the Dental Suffix (//id//) or the Sibilant Suffix (//iz//) is preceded by [t] [d] or [s] [z] [J] [3] respectively, thus, it prevents the ill-formed tautosyllabic sequences of obstruents from arising. Granted the validity of all arguments adduced above the rules governing the distribution of the allomorphs of the Dental and the Sibilant Suffix could be treated as post-lexical phonological statements. As such they no longer endanger the B R A C K E T ERASURE CONVENTION since no morphological information stemming from the Lexicon is involved in their operation. Finally, we can address the issue of the underlying representation of the Dental and the Sibilant Suffix. It is still unresolved, as, until now, no conclusive arguments have been found in favour of either of the two rules that conceivably could account for the variation between syllabic and non-syllabic allomorphs of the morphological units in question. The two rules at hand imply various underlying representations for the two suffixes: Insertion implies the non-syllabic allomorph at the underlying level (//z//and //d//). Vowel Deletion implies the syllabic allomorph at the underlying level (//iz//and //id//). The vowelless solution is perhaps preferable since it enables us to assume that the Dental Suffix has a uniform representation for both regular and semi-regular verbs. The multiplication of the lexical representations of the Dental Suffix can thus be avoided. The cost of this solution is the necessity to set up the lexically-governed allomorphy rule unvoicing the Dental Suffix in certain semi-regular verbs whose stems end with sonorants. On the other hand, the rule of Vowel Deletion seems to have an independent motivation of some sort, given the following premises: - Rubach (1977) has established that optional phonological rules are often reflexes of the obligatory rules. - There is in modern English an optional rule of Vowel Deletion. Thus, we could justifiably conclude that the optional rule of Vowel Deletion is a reflex of an analogous obligatory rule. All in all, there seems to be very little independent motivation for either solution and whatever there is points in opposite directions. 2 . 1 . 1 0 . The foregoing discussion of inflectional suffixation in modern English leads to the following conclusions with regard to constraints on rules inherent in the model of Lexical Phonology. First, various instantiations of the Dental Suffix and the Sibilant Suffix are attached in different places in the Grammar. The Dental Suffix is attached at STRATUM 1 in the Lexicon to a closed set of verbal stems (semi-regular verbs) and to an open class of verbal stems (regular verbs) at STRATUM 2. Arguably, the

33 lexical representation of the Dental Suffix (more exactly its underlying phonological representation) is the same at both STRATA, at STRATUM 1 the Suffix exhibits idiosyncratic behaviour accounted for in terms of a lexically-governed allomorphy rule of unvoicing. Alternatively, the underlying representation of the Dental Suffix varies at the two STRATA (Halle and Mohanan 1985). The Plural instantiation of the Sibilant Suffix is attached in the Lexicon, the Saxon Genitive outside the Lexicon. The attachment of the Saxon Genitive violates the B R A C K E T ERASURE CONVENTION, since it crucially involves morphological information created in the Lexicon (it is blocked by the presence of the Sibilant Suffix in its host's final position). Second, the phonological properties of the Dental Suffix and the Sibilant Suffix motivate the distinction between STRATUM 1 and STRATUM 2. Crucially, the semi-regular verbs at STRATUM 1 exhibit stem alternations in the context of the Dental Suffix which never occur with regular verbs at STRATUM 2. This is best accounted for by assuming that certain phonological rules apply only at STRATUM 1. On the other hand, the phonological properties of the Dental and the Sibilant Suffix provide no evidence for a separate STRATUM at which regular inflection would be processed (STRATUM 3 in Kiparsky 1982a, STRATUM 4 in Mohanan 1986). Syllabic/non-syllabic and voiced/voiceless allomorphs of both Suffixes are accounted for in terms of post-lexical phonological rules: the Progressive Unvoicing of Obstruents, which is also at work in rapid and casual speech in the case of weak forms of auxiliary verbs ("has" "is") and either of the equally well-motivated Vowel Insertion or Vowel Deletion. All these rules can be shown to be specific implementations of wider ranging constraints on syllable well-formedness, which stipulate that tautosyllabic clusters of adjacent obstruents must not be too similar in certain respects (in the case of coronal obstruents they must not agree in the manner of articulation) while at the same time they must be similar (identical) in other respects (agreement in voicing). The putative rule of Vowel Insertion is in equal measure motivated by a similar constraint on well-formedness of geminate clusters in modern English and as such it competes with the phonological rule of Degemination.

2.2. Derivation 2.2.1. In the previous section we argued that the descriptively adequate account of the attachment of the Saxon Genitive suffix involves violation of the BRACKET ERASURE CONVENTION at the post-lexical STRATUM. The B R A C K E T ERASURE CONVENTION is a vital constraint on the overall organisation of the Grammar, by virtue of which in a multilayered Lexicon morphological and phonological rules may not have access to morphological information introduced at an earlier STRATUM. (By the same token, morphological structure created in the Lexicon is not transparent to post-lexical rules). The unquestionably post-lexical rule of the Saxon Genitive suffix attachment is governed by a restriction inhibiting its application to hosts containing an instantiation of the Sibilant Suffix (Plural, Third Person Singular Present Tense, Saxon

34 Genitive) in the stem final position. The formalisation of this restriction must refer to the following structural configuration: [//z//]; the morphological boundary preceding the Sibilant Suffix is of paramount importance, as the rule at hand is not precluded from application by a mere presence of the segment /z/ in stem final position. Yet, if the B R A C K E T ERASURE CONVENTION were permitted to erase the boundary preceding //z//this crucial distinction would be obliterated and the restriction on the Saxon Genitive attachment could not be encoded in the Grammar. We have also argued that an adequate description of inflectional suffixation in modern English does not necessitate positing a separate STRATUM in the Lexicon at which regular inflectional phenomena would be processed (STRATUM 3 in Kiparsky (1982) STRATUM 4 in Mohanan (1986)). Prom our discussion of the English inflectional suffixes the following picture of Lexical Phonology as a morphological theory seems to be emerging: the framework as presented in e.g. Mohanan (1986) is perhaps too richly structured as far as the internal organisation of the Lexicon is concerned. It may well be the case that an adequate morphological theory could do with fewer (or even possibly no) STRATA. On the other hand, at least one major tenet of Lexical Phonology - the B R A C K E T ERASURE CONVENTION makes it too restrictive as a theory of morphological processes i.e. an account of what possible morphological rules are allowed to do. The possibility that some of them behave like the Saxon Genitive Suffix attachment in that they crucially involve morphological information created at earlier STRATA is definitely worth exploring. In what follows I wish to develop both of these lines of argument. After sketching briefly some well-known arguments against unnecessary stratal distinctions (separate STRATA for compounding and inflection - STRATUM 3 and STRATUM 4 respectively in Mohanan (1986)) we focus on empirical predictions entailed by two major morphological hypotheses of Lexical Phonology: the bifurcation of all derivational processes in modern English into two STRATA (the A F F I X ORDERING GENERALIZATION as Selkirk (1982) aptly dubbed it) and the B R A C K E T ERASURE CONVENTION. An attempt will be made both to state these predictions clearly and to test them with a well-defined class of derivational rules: all processes of derivational stem-based suffixation in modern English. It is my firm belief that viewed against such a sufficiently large body of relevant data (and not merely illustrated with a handful of carefully selected examples) the constraints on morphological rules inherent in the both major tenets of Lexical Phonology are deficient as explanatory principles with regard to suffixation in contemporary English. As shown in Aronoff and Sridhar (1987) the AFFIX ORDERING GENERALIZATION seems to be too restrictive a claim about grammatical configurations of derivational suffixes in modern English. It rules out certain classes of well-formed complex word-forms. At the same time and perhaps more significantly as we will argue in some detail, it is consistently far too permissive, in that it does not rule out the occurrence of certain classes of complex word-forms that systematically fail to appear. The overwhelming majority of derivational suffixes in modern English are attached subject to a very peculiar restriction first pointed out by Kiparsky (1982a) with regard to the denomi-

35 nal verbalizing zero suffix, namely, they never attach to an already suffixed stem. The above restriction on suffixation in modern English disqualifies the AFFIX ORDERING GENERALIZATION as too permissive a claim. It also, however, seriously undermines the BRACKET ERASURE CONVENTION proving it to be too restrictive in as much as rules of derivation posited at STRATUM 2 are inhibited, in equal measure, by suffixes from both STRATUM 1 and STRATUM 2. Hence the morphological boundaries preceding STRATUM 1 suffixes must be available also at STRATUM 2. Additionally, a few derivational STRATUM 2 suffixes that attach only after a well-defined set of other suffixes (e.g. the de-adjectival verbalizing suffix (ize) which attaches only to adjectives created by means of 4 adjectivizing suffixes (ic) (iv) (al) (an)) also never distinguish between STRATUM 1 and STRATUM 2 suffixes (In the example given here (ize) is a STRATUM 2 suffix while all 4 adjectivizing suffixes are attached at STRATUM 1). 2.2.2. The hypothesis that the Lexicon consists of several ordered STRATA (Kiparsky (1982 a,b,c) Mohanan (1986)) is an elaboration of an influential claim first advanced by Siegel (1974/79) with regard to English derivational affixes. Siegel argued that combinatorial properties of affixes correlate with their phonological properties, in particular their ability to affect the stress pattern of their host. In a widely held view "the explanatory adequacy of this hypothesis derives from the consistency with which combinatorial properties of affixes can be determined by their STRATUM membership and from the interesting correlations (...) between STRATUM membership and various phonological properties of affixes" (Pesetsky 1979:17). The stratal structure of the Lexicon is a universal property of human language - part and parcel of the UNIVERSAL GRAMMAR in the sense of Chomsky (1965). The number and characteristics of STRATA in a given human language is a language-specific fact, a contingent trait of a given GRAMMAR. As noted by Mohanan (1986) both phonological and morphological justification should be provided for each STRATUM, since, ideally, it is a locus of both morphological and phonological processes. It will be remembered that the leading proponents of Lexical Phonology differ with regard to the number of STRATA they propose for modern English (4: Mohanan (1986) 3: Kiparsky (1982a) 2: Booij and Rubach (1987)). The bifurcation of derivational rules into two consecutive STRATA (affix boundary affixation at STRATUM 1, word boundary affixation at STRATUM 2) seems to be a well-established distinction. Most current work on descriptive phonology and morphology of modern English presupposes its validity and both phonological and morphological evidence is adduced in favour of it. The existence of separate lexical STRATA for compounding and inflection (STRATUM 3 and STRATUM 4 respectively in Mohanan (1986)) is a much more contentious issue. Our discussion will reflect this unequal measure of recognition that various stratal distinctions receive in the influential linguistic literature. First, we shall briefly review the evidence adduced in support of separate STRATA for compounding and regular inflection. Then, at somewhat greater length, we shall attempt to challenge the distinction between STRATUM 1 and STRATUM 2. 2.2.3. Kiparsky (1982a) argues for the conflation of compounding with word boundary affixation at one STRATUM on the basis of well-known distributional properties of

36

word boundary affixes, namely their ability to appear both inside and, more significantly, outside compounds (see also Strauss (1982)). Mohanan (1986) does not question the relevance of such data for the validity of the distinction between word boundary affixation and compounding in terms of stratal organisation. In his own words "Given (...) the distinction between class 2 derivation and compounding in terms of stratal organisation one would expect that compounding cannot be an input to class 2 derivations" (Mohanan 1986:51). Note that the above statement is tantamount to formulating empirical predictions inherent in the claim that word boundary affixation operates at a STRATUM preceding the STRATUM at which compounding takes place. Mohanan (1986) goes on to admit "That this is not so is amply demonstrated by words like [hard-hearted] ness" (Mohanan 1986:51). This statement, in turn, is nothing else, but an open admission that the empirical predictions are falsified by available evidence which calls the hypothesis at hand into serious doubt. It should be emphasised that the relevant counterexamples are not isolated words but open classes of well-formed complex words e.g. Adj-V-ed-ness. The hypothesis is too strong as it rules them out, incorrectly predicting their non-occurrence (ungrammaticality). There is no reason to stop short of the following conclusion: positing word boundary derivation and compounding on separate STRATA entails wrong empirical predictions about the properties of English word boundary derivational suffixes. Mohanan (1986) however, attempts to salvage the hypothesis at hand by resorting to the following reasoning: 1. He regrets that not all STRATA are supported by both morphological and phonological evidence. 2. Having admitted that there is no morphological evidence in favour of placing word boundary affixation and compounding at separate STRATA he goes on to present what he considers to be phonological evidence for the distinction between STRATUM 2 and STRATUM 3. 3. He attempts to circumvent incorrect morphological predictions following from the assumption that word boundary affixation and compounding are effected at separate STRATA by introducing the LOOP, device which allows the output of STRATUM to go back to a previous STRATUM. Each of the stages of Mohanan's reasoning deserves a brief comment. Lack of evidence corroborating an empirical hypothesis should not be confused with counter-evidence effectively disconfirming (falsifying) it. Corroborating evidence in favour of an empirical hypothesis is constituted by all such phenomena, whose existence is predicted by the hypothesis. No amount of corroborating evidence can ultimately prove a hypothesis (i.e. constitute a LOGICAL proof that it is true - Popper (1936), yet linguists and indeed all empirical scientists are at pains to produce such evidence, since there is no denial that it renders the hypothesis it supports more plausible (Mortimer 1982). We agree with Mohanan (1986) that no morphological phenomena in modern English could serve as corroborating evidence in favour of his hypothesis that word boundary affixation and compounding take place at separate lexical STRATA. (This is how one should restate Mohanan's statement about the lack of supportive morphological data in favour of the distinction between STRATUM 2 and STRATUM 3). More significantly, however, as

37 mentioned above, there exists counter-evidence discontinuing the hypothesis Mohanan (1986) is at pains to defend. It is received wisdom of current methodology of empirical science that an empirical hypothesis is effectively disconfirmed (falsified) by showing that it predicts the non-occurrence of some well attested phenomena. Such claims are usually labelled as empirically too strong (too restrictive). As mentioned above the available morphological evidence seems to suggest that Mohanan's hypothesis that word boundary affixation and compounding should be processed at separate STRATA is a case in point. All in all, as far as relevant morphological data are concerned, not only is nothing gained by setting up STRATUM 2 as distinct from STRATUM 3 (compounding) but much is lost, much more than a linguist with genuine concern for the empirical content of his claims should be prepared to lose. The allegedly remedial expedient introduced by Mohanan (1986) i.e. the LOOP-device, which legitimises violations of the STRATA ordering hypothesis by allowing the output of a STRATUM to go back to the previous STRATUM instead of salvaging the hypothesis at hand renders it vacuous (this point is made clearly in Aronoff and Sridhar (1987), Booij and Rubach (1987), Gussmann (1988). Mohanan's phonological evidence in favour of the distinction between STRATUM 2 and STRATUM 3 involves a single phonological rule, whose "status is dubious" (Gussmann 1988:234) - the Stem Final Tensing. Even if we disregard the fact that Mohanan's data find no support in the most comprehensive survey of modern English dialects published to date (Wells 1982) there seem to be alternative ways of accounting for them (Booij and Rubach 1987) without invoking the stratal distinction. Mohanan (1986) argues for positing word boundary derivation and compounding at separate lexical STRATA. In view of the fact that there is neither morphological nor phonological evidence corroborating the distinction between STRATUM 2 and STRATUM 3 and, on top ofthat, enough morphological evidence to disconfirm it, we conclude, following Kiparsky (1982a) and Booij and Rubach (1987) that the distinction at hand could safely be abandoned. The existence of a separate lexical STRATUM for English regular inflection is upheld by Kiparsky (1982a) and Mohanan (1986). It appears to be strongly motivated since both phonological and morphological evidence are adduced in its favour. Mohanan (1986) argues that the hypothesis that regular inflection and compounding should be posited at separate STRATA entails one correct empirical prediction: the left hand member of a compound is not inflected in English. If this assertion were indeed true, then it would constitute a valid piece of corroborating evidence in favour of the hypothesis under consideration. The non-occurrence of compound internal inflection is the state of affairs predicted by the hypothesis at hand, so if found to be correct it could validly be argued to be accounted for by this very hypothesis. The following regular inflectional endings, however, seem to appear systematically in a compound internal position: PLURAL //z//. Sproat (1985), for one points out numerous instances of compound-internal plural endings, e.g. systems analyst, arrivals lounge, parts department. A left

38 hand side member of a compound may be inflected for plural and Sproat (1985) goes on to argue that the correct generalization can only be captured by a constraint on rule application and not by postulating separate STRATA for inflection and compounding. The following is the constraint proposed by Sproat (1985): The left hand member of a compound must be unmarked for number, unless the plural is interpreted collectively or idiosyncratically. / / V r / / - the exponent of the comparative degree of adjectives may appear compound internally e.g. harder-skinned, longer-lasting / / z / / t h e exponent of the Saxon Genitive may appear compound internally e.g. hair's breath. In fact, as argued convincingly by Gold (1975) it exhibits the tendency to expand in such context e.g. rabbit foot is replaced by rabbit's foot In view of the facts exemplified above one is not only ready to agree with Booij and Rubach (1987) that "there is no morphological motivation for STRATUM 4" (Booij and Rubach 1987:26), but, more significantly, it must be admitted that there is enough morphological evidence to effectively falsify the hypothesis that compounding and inflection are effected at separate STRATA. Phonological evidence for STRATUM 4 involves two phonological rules: Sonorant Syllabification and [1] Velarization. (27)

Consider the following pair of words: twinkling: a disyllabic noun twinkling: a trisyllabic participle with syllabic [1].

The observation that the noun is always pronounced as a disyllabic word while the verb may be pronounced as a trisyllabic one (with syllabic [1]) goes back to Chomsky and Halle (1968). The rule responsible· for this contrast is stated as follows (Chomsky and Halle 1968:85) (28)

[+son] -

[+syll]/_#

and is known as Sonorant Syllabification. The desired result is achieved by representing: - twinkling (noun) as /twinkl + ing/ - twinkling (participle) as /twinkl # ing/ thus making the rule of Sonorant Syllabification applicable to the participle but not to the noun. The simplest solution within the multi-stratal approach of Lexical Phonology is to assume that the rule of Sonorant Syllabification (now conditioned by the presence of]) applies at STRATUM 4 1 . Only word forms effected by morphological rules placed at the same STRATUM still retain internal boundaries while the rule applies (this concerns participles such as twinkling V) whereas nominalizations that have arisen at STRATUM 2 (twinkling N) are now morphologically opaque due to the operation of the BRACKET ERASURE CONVENTION: twinkling Ν twinkl] ing V STRATUM 4. 1 Here we do not follow Mohanan's (1986) approach to English syllabification since it deserves a separate treatment for which a work on morphology is no proper place (but see severe and strong-worded criticism in Gussmann (1988).

39 Since the rule of Sonorant Syllabification is triggered off by the presence of morphological boundary it can apply to the participle but not to the nominal. Notice that this account rests on the assumption that regular inflection (including participle (ing)) is processed at a separate STRATUM than word boundary derivation (including nominalization (ing)) hence, if correct, would constitute a valid piece of corroborating evidence in favour of setting up STRATUM 4 (a separate inflectional STRATUM in the Lexicon). A more careful look at relevant data shows, however, that the above analysis rests on incorrect factual assumptions. In point of fact, English word final sonorants preceded by a consonant Eire regularly syllabic, they regularly lose syllabicity when followed by a morpheme boundary suffix and only occasionally when followed by a word boundary suffix. The unmarked case, so to say, is to have a syllabic sonorant in the following context: C - word boundary, whether word boundary is followed by a non-STRATUM 1 suffix or not. Heinz Giegerich has kindly pointed out to me that this claim is borne out by nonce-formations with the word boundary agentive suffix (er) such as (29)

wrinkl.er: [1] is syllabic bottl.er: [1] is syllabic fickl.er: [1] is syllabic

All the cases where a word boundary suffix triggers off desyllabification of a stem final sonorant preceded by a consonant (including nominale with (ing) such as twinkling) should be treated as exceptions to the above generalization, as instances of lexicalization. In all such cases we can assume that a given word boundary suffix may attach at STRATUM 1 to lexically specified items. (30)

wrangle]V

wrangl]V er]N

[1]

-

bottl]V er]Ν [1]

STRATUM 1 STRATUM 2 Sonorant Syllabification

The rule of Sonorant Syllabification is prevented from applying to wrangler due to the operation of the BRACKET ERASURE CONVENTION. At times even a word-form produced via the attachment of word boundary inflectional suffix may become lexicalized in this respect. Compare (31a) to (31b) in this respect. (31)

(a) simple: [1] syllabic [1] simpler: [1] non-syllabic [1] fickle : fickler "j (b) feeble : feebler

> both forms with syllabic [1]

brittle : brittler J There is nothing surprising in having the comparative suffix attach to a few adjectives already at STRATUM 1. In fact, well known exceptions to the rule of word final [g] deletion (longer, stronger and younger)call for the same stipulation (Halle and Mohanan 1985). The distribution of syllabic sonorants does not seem to provide valid justification for a separate inflectional STRATUM. Booij and Rubach (1987) reach the same conclusion with regard to the rule of [1] Velarization. All in all, there is neither mor-

40 phological nor phonological evidence corroborating the positing of STRATUM 4 (the allegedly supportive evidence can be accounted for otherwise) and enough morphological counter-evidence to disconfirm it. 2.2.4. In the foregoing discussion an attempt was made to show that the number of lexical STRATA set up in some current versions of Lexical Phonology could be reduced. In particular, there seems to be no evidence to warrant the assumption that the morphological processes of compounding and regular inflection operate at separate STRATA (STRATUM 3 and STRATUM 4, respectively, in Mohanan 1986) (similar conclusion in Booij and Rubach 1987). Before we examine English derivational suffixes for the light they may throw on the remaining stratal distinction (STRATUM 1 and STRATUM 2 the AFFIX ORDERING GENERALIZATION in Selkirk 1982) let us focus on a single morphological process - the denominal verbalization by means of a zero suffix (conversion). I shall discuss the analysis of this process carried out in Kiparsky (1982a) since, to my mind, it highlights in a very instructive fashion two crucial points I try to make about English derivational suffixes in general: 1. many derivational suffixes do not attach to an already suffixed stem, 2. the BRACKET ERASURE CONVENTION fails to make correct predictions with regard to stem-based suffixation in modern English. The syntactic reclassification of nouns as verbs without an overt affixal or segmental marker is a highly productive verbalizing process in modern English (Marchand 1960, Clark and Clark 1979). The simplest way to account for it is to set up a morphological rule by virtue of which a verbalizing suffix (phonetically null) is attached to [X] NOUN. The nominal base of this rule is subject to a very interesting restriction, first pointed out in Kiparsky (1982a), which, as he goes on to argue, provides firm evidence in favour of the AFFIX ORDERING GENERALIZATION and the BRACKET ERASURE CONVENTION. Kiparsky (1982) claims that zero-derived verbs may be formed from nouns derived at STRATUM 1 (i.e. arising via morpheme boundary suffixation) but not from nouns derived at STRATUM 2 (verbs such as* to singer and *to freedom are ill-formed). Kiparsky (1982a) notes, furthermore, that there is a general principle prohibiting the affixation of the verbalizing null suffix to already affixed words. This prohibition coupled with the BRACKET ERASURE CONVENTION and the assumption that the denominal verbalizing conversion takes place at STRATUM 2 correctly (according to Kiparsky) predicts: 1. the occurrence of verbs derived via zero-derivation from nouns derived at STRATUM 1. 2. the non-occurrence of verbs derived via zero-derivation from nouns derived at STRATUM 2. The prohibition on the application of verbalizing zero-derivation, discovered by Kiparsky (1982a), prevents the word boundary nominal derivatives from undergoing verbalizing zero-derivation, which, at the same time, due to the operation of the BRACKET ERASURE CONVENTION treats morpheme boundary nominal derivatives as morphologically opaque, i.e. on a par with underived nominale. Kiparsky (1982a) explicitly

41

claims that the AFFIX ORDERING GENERALIZATION and the BRACKET ERASURE CONVENTION are empirical hypotheses which make correct empirical predictions with regard to the data at hand. "These facts fall out from Bracket Erasure and Level Ordering we assumed" (Kiparsky 1982a:141). The logic of Kiparsky's argument is, as one would expect, flawless. It is the facts at hand that, let us repeat after Sproat (1985) "are unclear" (Sproat 1985:191). Kiparsky's second prediction is borne out by available data. The non-occurrence of verbs zero-derived from word boundary nominal derivatives is a well-attested phenomenon in modern English (Marchand 1960) Sporadic exceptions are noted by Sproat (1985). Kiparsky's first prediction is, however, false. (Sproat 1985). The zero verbalizing suffix systematically fails to attach to nouns formed at STRATUM 1 by means of stem-based morphological rules. There are at least four nominalizing suffixes in modern English that attach to stems at STRATUM 1: deverbal (ation): conversation (ion): diversion denominal (y) (//j//): democracy de-adjectival (ity) : profundity All the four classes of words exemplified above are systematically excluded as bases for the verbalizing zero-derivation. The account of the data at hand requires that the prohibition on verbalizing zero suffixation be extended also to STRATUM 1 derivatives. Notice, however, that if we adhere to the assumption that the morphological rule in question applies at STRATUM 2, then the BRACKET ERASURE CONVENTION effectively prevents it from access to morphological structure arising at STRATUM 1. Hence, given the BRACKET ERASURE CONVENTION Kiparsky's prohibition (revised as required by the relevant data) cannot be encoded in the Grammar. To sum up, Kiparsky (1982a) discovered an important and well-motivated restriction constraining the operation of one morphological rule in modern English (the denominal verbalizing zero-suffixation) but he incorrectly assumed that STRATUM 1 nominal stem-based suffixal derivatives should be exempt from it. This assumption led him to the conclusion that the analysis of the verbalizing zero-suffixation in modern English lends support to basic tenets of Lexical Phonology with regard to morphology, in particular the BRACKET ERASURE CONVENTION. On closer scrutiny, however, the data at hand seem to show that the empirical claim inherent in the BRACKET ERASURE CONVENTION concerning morphological rules merits a more detailed investigation. 2.2.5. The standard model of Lexical Phonology espoused by Kiparsky (1982) and Mohanan (1986) advances two important hypotheses with regard to derivational processes in modern English: 1) Derivational processes fall into two classes according to phonological properties of affixes effecting them. All "stress affecting" derivations are placed at STRATUM 1 hence they precede all "stress neutral" derivations placed at STRATUM 2. Following Selkirk (1982) we have been referring to this hypothesis as the AFFIX ORDERING GENERALIZATION.

42 2) Word internal morphological brackets are erased at the end of STRATUM 1. (This follows directly from the BRACKET ERASURE CONVENTION). Hypotheses 1. and 2. are logically independent. Now, let us note several important empirical predictions that they entail with regard to the morphological properties of English derivational affixes. AFFIX ORDERING GENERALIZATION. Let us start with a familiar observation that the AFFIX ORDERING GENERALIZATION restricts the permissible sequence of morphological operations introducing affixes from different STRATA. In particular, no morphological rule introducing a "stress-affecting" affix is allowed to apply to a complex word-form which already contains a stress neutral affix introduced by an earlier morphological rule. As has been repeatedly pointed out by proponents of the AFFIX ORDERING GENERALIZATION the above prediction is corroborated by the non-existence of certain pairs of suffixes which it effectively rules out. A well-known example concerns the restricted distribution of the stress-affecting de-adjectival nominalizing suffix (-ity) as compared with the co-functional stress-neutral suffix (-ness). Although both suffixes attach to adjectival bases, it is only (-ness) and not (-ity), that yields well-formed words by attaching to adjectives created due to the previous application of some adjectivizing "stress neutral" suffixation e.g. forget.ful.ness vs. * forget.ful.ity. On the other hand both (-ity) and (-ness) are free to attach to adjectives created via suffixation at STRATUM 1. e.g. verbose: verbosity: verbosness. A number of similar examples, all of which are rightly held to constitute evidence corroborating the AFFIX ORDERING GENERALIZATION as an empirical hypothesis are well known from the relevant literature (Siegel 1974/79, Selkirk 1982). At the same time, the AFFIX ORDERING GENERALIZATION has been repeatedly shown to be too strong as an empirical hypothesis, in as much as it predicts the non-occurrence of complex word forms which Eire in fact well-formed. First, Strauss (1982) argued convincingly that morphological rules introducing stressaffecting suffixes must often be preceded by morphological rules attaching stress neutral prefixes. The most celebrated and widely quoted example involves two rules: the de-adjectival negative prefixation with (un) and the de-adjectival nominalization with (ity). The former must crucially precede the latter, as it applies to adjectives but not to nouns. The morphological structure of words such as [[un.[grammatic.al]]A ity]N reflects this ordering. At the same time, (un) belongs to STRATUM 2 while (ity) is attached at STRATUM 1. Secondly, Aronoff and Sridhar (1987) point out that in at least three well-attested cases stress-affecting suffix follows a stress-neutral suffix: abil.ity e. g. analyzeability iz.ation e.g. computerization ist.ic e.g. Qaddafistic Another sequence that comes to mind: ment.al e.g. developmental It is of paramount importance that all the above examples refer to open classes of words rather than isolated instances. ("The three classes (...) are rather productive"

43 Aronoff and Sridhar 1987:12). The AFFIX ORDERING GENERALIZATION makes wrong predictions not merely with isolated formations (this seems to be the view of Selkirk 1982) but rather with regard to certain possible words; it is deficient as a recursive definition of well-formed complex words in modern English. Aronoff and Sridhar (1987) conclude that "all potential words of the forms X#ability, X#istic, X#ization falsify the A F F I X ORDERING GENERALIZATION" The demonstrably excessive empirical strength of the A F F I X ORDERING GENERALIZATION is the only aspect of this interesting hypothesis that has received critical attention. What, in my view, merits an equally detailed consideration is the fact that the A F F I X ORDERING GENERALIZATION is an extremely weak empirical claim as a hypothesis about grammatical sequences of derivational morphemes it does not rule out many pairs which apparently never occur. Note the following empirical predictions the AFFIX ORDERING GENERALIZATION entails with regard to distributional properties of English derivational suffixes: 1) Stress-affecting suffixes are not allowed to follow stress-neutral suffixes, but no restrictions are placed on them with regard to the number of other stress-affecting suffixes that may precede them. 2) stress-neutral suffixes are totally unrestricted in this respect. In other words the A F F I X ORDERING GENERALIZATION coupled with familiar subcategorization properties of suffixes predicts what follows: Given a STRATUM 1 suffix which attaches to (a subclass of) members of major syntactic category X the suffix may attach to hosts conforming with the following structural condition: [ ]X, with the familiar exception of configurations such as [[ ]Y SUFF2]X (SUFF2:STRATUM 2 suffix; X , Y major syntactic categories). Otherwise the content of the brackets labelled with X is immaterial, a STRATUM 1 suffix may attach both to morphologically underived forms and forms created via suffixation at STRATUM 1. Given a STRATUM 2 suffix which attaches to (a subclass of) members of major syntactic category X the suffix may attach to hosts conforming with the following structural condition: [ ]X, regardless of the content of the brackets labelled with X. In particular a STRATUM 2 suffix may attach to morphologically underived forms, to forms created via suffixation at STRATUM 1 and to forms created via suffixation at STRATUM 2. Very, clearly, the AFFIX ORDERING GENERALIZATION is tantamount to the following stipulation: - morphological rules attaching derivational suffixes in modern English are subject to categorial conditions only and are free to apply regardless of the morphological make up of the hosts (the only principled exception to this claim being the familiar restriction on the distribution of STRATUM 1 suffixes mentioned above). Now, a close scrutiny of the relevant data with a view of checking if it warrants such empirically weak claim is, in our view, the next logical step to take. Morphological hypotheses are, more often than not, merely illustrated with a handful of carefully selected examples (as has been so far the case with the AFFIX ORDERING GENERALIZATION). Contrary to this prevalent practice we wish to analyse all the relevant data. In other words in section 2.6 we intend to answer the following question: how many of

44 the derivational suffixes in modern English are as promiscuous in their morphological behaviour as it is predicted by the AFFIX ORDERING GENERALIZATION? B R A C K E T ERASURE CONVENTION. It follows from the BRACKET ERASURE CONVENTION that none of the morphological rules attaching stress-neutral suffixes (STRATUM 2) may have access to morphological information introduced by the morphological rules attaching stress-affecting suffixes (STRATUM 1). Thus, the BRACKET E R A S U R E CONVENTION entails the following claim (s) with regard to derivational suffixes in modern English: The attachment of a stress-neutral suffix may not be either triggered off or blocked by the presence of a stress-affecting suffix in the base. We hope to show in section 2.6 that the above claim (s) are incorrect with regard to more than one instance of a stress-neutral suffix in modern English. 2.2.6. In the preceding section we laid out, as clearly as possible, empirical predictions entailed by the AFFIX ORDERING GENERALIZATION and the BRACKET ERASURE CONVENTION with regard to English derivational suffixes. The present section is devoted to testing these predictions against the relevant data: all morphological rules attaching derivational suffixes to members of major syntactic classes (stems) in modern English. For the ease of exposition we adopt the commonsensical division of derivational suffixes according to the syntactic class membership of the words they create into VERBALIZING, NOMINALIZING, ADJECTIVIZING and discuss them in that order. Each suffix is characterized, first of all, in terms of the syntactic class of its hosts as deverbal, denominal, deadjectival or attaching to roots (elements unspecified for syntactic category). Thus, it is assumed that homophonous suffixes that attach to members of more t h a n one syntactic class are different suffixes, even if they share many other characteristics. This is consistent with analytical principles underlying traditional descriptive works on English word formation (e.g. Marchand 1960) where the verbalizing denominal (ize) (e.g. victim.ize) and the verbalizing deadjectival (ize) (e.g. modern.ize) are treated as different derivational suffixes. Note that the third separate item is the suffix (ize) which attaches to roots (e.g. bapt.ize, fratern.ize). VERBALIZING SUFFIXES (ify) denominal suffix: attaches to [ ]N class.ify, test.ify. It is attached at STRATUM 1 as it is stress-affecting (person. personify) and the word-forms it creates are subject to the phonological rule of Trisyllabic Shortening undoubtedly a STRATUM 1 rule (Kiparsky 1982, Mohanan 1986): type: typify, code: codify, mode: modify. Given all the above characteristics the following empirical predictions with regard to the distribution of (ify) follow from the AFFIX ORDERING GENERALIZATION. 1. (ify) should not attach to hosts of the following morphological structure: [[X]Y SUFF2]N SUFF: any nominalizing suffix attached at STRATUM 2 Y: variable ranging over symbols for major syntactic classes: Ν, A, V 2. (ify) is free to attach to

45

a. morphologically underived nouns b. nouns created via the attachment of STRATUM 1 nominalizing suffixes to roots c. nouns created via the attachment of STRATUM 1 nominalizing suffixes to stems. The claim 1. above is borne out by all relevant data. In modern English there are 10 nominalizing suffixes, which due to their phonological characteristics are attached at STRATUM 2. betray.al deverbal

(-er) kill.er (-ment) contain, ment (-er) prison, er denominal (-hood) brother, hood (-ism) mao.ism (-ist) method.ist deadjectival (-ness) clever, ness (-ist) modern.ist (-ism) popular.ism (ify) is never found to attach to nouns created due to the attachment of any of the STRATUM 2 nominalizing suffixes. Thus, the AFFIX ORDERING GENERALIZATION correctly constrains the attachment of (ify)· Still, it fails to constrain it sufficiently. Notice that (ify) does NOT attach to nouns created due to the attachment of STRATUM 1 nominalizing suffixes to stems (prediction 2c). There are 10 such suffixes in modern English: deverbal (-age) break, age (-ent) defend, ent (-ance) admitt.ance (-ion) divers, ion denominal (-age) bagg.age (-an) republic, an (-ary) revolution.ary democrac.y

deadjectival ) material

1

de-V(-er)

de-v(-ence)

de-V(-ent)

inventoried

referential

confidential

directorial

existential

residential

editorial de-(ion)

(-less) breathless

2

providential de-V(ation)

expressionless informationless foundationless

de-v(-er) driverless prayerless leaderless

de-Α (-ist): dualistic, formalistic, nationalistic

(-ic> slavic

1

de-N(-ist): cubistic, sadistic

The foregoing analysis of derivational suffixation in modern English seems to lead to the following conclusions with regard to the two major constraints Lexical Phonology imposes on morphological processes: the AFFIX ORDERING GENERALIZATION and the BRACKET ERASURE CONVENTION.

52

According to the AFFIX ORDERING GENERALIZATION derivational suffixes in modern English fall into two groups with regard to their sensitivity to the morphological structure of the bases they attach to (the division is co-extensional with the distinction between stress-shifting and stress-neutral suffixes) - stress-shifting suffixes do not attach to hosts that contain a stress-neutral suffix and they are free to attach to hosts containing a stress-shifting suffix - stress-neutral suffixes are free to attach to hosts regardless of their morphological make-up. Our survey of English derivational suffixes shows: Only three stress-neutral suffixes (the deverbal adjectivizing (-able), the deverbal nominalizing (-er), the de-adjectival nominalizing (-ness)) behave as predicted by the AFFIX ORDERING GENERALIZATION i.e. attach to stems regardless of their morphological structure. At least 12 suffixes attach to words created via the attachment of a specified set of derivational suffixes to stems (i.e. members of major syntactic classes) but not to words created via the attachment of the remaining derivational suffixes in a given syntactic class (e.g. the de-adjectival nominalizing suffix (-ism) attaches to words created via the attachment of merely 4 (out of 20) adjectivizing suffixes to members of major syntactic classes2. The majority of English derivational suffixes never attach to stems created via the attachment of derivational suffixes to members of major syntactic classes (i.e. stems). In other words the bulk of stem-based rules of derivational suffixation in modern English are systematically blocked by the following structural configuration in the potential bases [[ ]X SUFF]Y X,Y:A,N,V. As pointed out by Kiparsky (1982a) it follows from the BRACKET ERASURE CONVENTION that "secondary (i.e. STRATUM 2 - A.W.) suffixes may not have access to morphological information created at STRATUM 1" (Kiparsky 1982a). This prediction has been shown to be incorrect with regard to: - all STRATUM 2 suffixes which are blocked by the presence of the following configuration in the host: [[ ]X SUFF 1]Y SUFF1: STRATUM 1 suffix. - several STRATUM 2 suffixes (e.g. (ism) (ist) (ize) (less)) which are triggered off by the following configuration: [[ ]X SUFF 1]Y SUFF 1: STRATUM 1 suffix.

2 Jerzy Rubach has pointed out to me, having read the first draft of this chapter, that similar conclusions were reached by Fabb (1988) and Halle (ms).

Chapter III Constraints on the base of morphological rules — verbalizing suffixes in polish 3.1. Verbalizing suffixes in Polish In the most extensive monograph on morphophonological structure of contemporary Polish that exists up to date, Rubach (1984) states that "verbs fall into several classes depending on the phonological structure of the stem" (Rubach 1984: 35). He proceeds to draw a distinction between what he calls C-verbs "those which end in a consonant or a glide" and "the remaining classes distinguished on the basis of different verbalizing suffixes which attach to roots" (Rubach 1984: 35). These various classes of verbal stems Eire listed in (1). (1) i. C-verbs (-t, -d) plesc "blab" - plot+ (3) 3p.pl. (-s, -z) pasc "graze" - pas+ (3,) 3p.pl. (-k, -g) wlec "drag" - wlok+ (3) 3p.pl. (-m, -n) dqc "blow" - dm+ (3) 3p.pl. (-r) drzec "tear" - d r + (3) 3p.pl. (-j) myc "wash" - myj+ (3) 3p.pl. ii. - a j stems czytac "read (cf. czyt + a j + 3) iii. -ej stems siwiec "get grey" iv. -e stems slyszec "hear" v. -a stems pisac "write" vi. -i steins kosic "mow" vii. -nVn stems taknqc "crave" The above list amounts to making a descriptive analytic generalization about the morphological structure of verbal stems in modern Polish: a verbal stem is a sequence of two morphological units. The morphological status of the first unit is unclear, the second unit determines the syntactic class membership of the whole configuration which is subject to morphological process of verbal inflection and deverbal derivation. The problems we intend to address below concern the exact characteristics of the morphological generalizations at hand. If verbal stems are indeed sequences of two morphological units the vital question to be answered is the following: is the second unit i.e. the verbalizing stem really attached to the first unit via a morphological rule of some sort or are the two units just listed together in the Lexicon. On the former solution two further problems arise: what is the exact morphological status of the first unit i.e. does the base of verbalizing suffixation in modern Polish consist of roots in all classes ("verbalizing suffixes attach to roots" Rubach 1984: 35), or is it perhaps sometimes made up of stems i.e. members of a major syntactic class?

54

What contribution do the two units make to the overall grammatical and semantic characteristics of the verbal stem as a whole? (So far we merely know that the second unit determines the major syntactic class membership of the whole configuration). The relevant morphological rules will be discussed with a view to establishing constraints they are subject to. In particular it will be argued that constraints relevant for Polish verbalizing processes concern the BASE of morphological rules. Two properties of the B A S E are of paramount importance in this respect. As pointed out in Chapter 1 all non-affixal morphological units fall into two classes: ROOTs and STEMs (for their characteristics see (Chapter 1). Accordingly, as we wish to claim, morphological rules of verb forming suffixation in Polish fall into two classes, whose bases are characterized as ROOTs and STEMs, respectively. As we shall do our best to demonstrate it is an entirely different empirical hypothesis to claim that a given suffix attaches to ROOTs rather than to STEMs. The B A S E of a morphological rule being the group of morphological units that are subject to it, morphological rules may vary depending on whether their bases can be characterized as classes or merely as sets. The B A S E as a class - the BASE is determined by giving a general condition a morphological unit must fulfil to qualify as its member. General conditions in question may concern common phonological, morphological, semantic or syntactic properties of morphological units. The BASE as a set - morphological units that are subject to a given rule form a set. The base is determined of necessity by giving a list of its members.

3.2. The verbalizing suffix / / i / / 3.2.1. T h e phonological properties The suffix surfaces in two phonological shapes [ i ] (front high, lax) and [i] (central high, lax). The underlying representation must be the front vowel [ i ] as the suffix invariably causes palatalization of the immediately preceding root-final consonant, ([i] as non-palatalizing is classified phonologically as [-bk]). (2) Velar Palatalization caused by the Verbalizing Suffix //i//: walczyc "to fight" m§czyc "to torture" r$czyc "to guarantee"

walka "a fight" πιφα "a torture" rqka "a hand"

(3) Anterior Palatalization caused by the Verbalizing Suffix //i//: koste "to mow" mrozic "to freeze" placic "to pay" nudzic "to bore" ranic "to wound"

kosa mroz optata nuda ran a

"a "a "a "a "a

scythe" frost" price" boredom' wound"

55 The surface variant [i] is derived via the phonological rule of retraction (Gussman 1980, Rubach 1984). The phonological derivation of verbal stems with the verbalizing suffix //i// is unproblematic. //valk + i// valt s + i 1 Velar Palatalization valt s + i Retraction There is no reason to assume that the segmental makeup of the lexical representation of the suffix under consideration is anything but identical to its underlying phonological representation.

3.2.2. T h e B A S E of t h e morphological rule of / / i / / suffixation. The morphological properties of the verbalizing suffix //i// are much more complicated. Consider the following exemplary data. (4) walka - walczyc "fight" Ν - V rana - ranic "wound" Ν - V zqdlo - zqdlic "sting" Ν - V zaszczyt - zaszczycic "honour" Ν - V brud - brudzic "grime" Ν - V The left hand side column in (4) contains nouns and the right hand side column verbs, whose stems consist of two elements: the root homophonous with the noun on the left and the verbalizing suffix //i// both subject to phonologically predictable modifications. The number of such pairs in modern Polish is substantial and it is rather obvious conclusion that the members of each pair are derivationally related. Notice, that a priori there are at least two ways of accounting for this relationship. A. One of the members in each pair is the derivational source of the other. B. Both members in each pair are derived from some common source. Conceivably, solution A has two variants. Αχ: verbal stems in the right hand side column in (2) are derived from corresponding nouns. On the plane of form the morphological operation consists in attaching the verbalizing suffix //i// to nominal bases. This solution is propounded in traditional grammars of Polish and in Grzegorczykowa and Puzynina (1979). A2: nouns in the left hand side column in (2) are back-derived from corresponding verbs. On the plane of form the morphological operation consists in removing the verbalizing suffix. The claim to this effect is advanced in Szpyra (1987). On solution Β the common derivational source for both the noun and the verb is the ROOT - a purely formal unilateral unit unspecified for syntactic category and devoid of semantic content. Verbs in the right hand side column in (2) are derived via the morphological operation which formally consists in attaching the verbalizing suffix //i// to ROOTs. Nouns in the left hand side column in (2) are derived from the ROOTs via the morphological operation of zero derivation

56 On the face of it solution A. has an obvious advantage over solution B. It is simpler, so the criterion of the economy of description should resolve the controversy in its favour. The relative descriptive simplicity of solution A. as compared to that of B. could hardly be denied. Solution B. seems to be more complicated both in terms of representations and morphological rules. Notice, that on solution A. (in either of its variants) we have to do with two basic representations related directly by a single morphological rule i.e. verbal STEM which arises due to the attachment of the verbalizing suffix to nominal STEM or, alternatively, nominal STEM back-derived from verbal STEM. On solution B, however, we have three representations two of which are related to the third one via two different morphological rules. Apart from the verbal STEM and the nominal STEM we have here their common structural core factored out and treated as a unit in its own right t h a t requires an autonomous representation (ROOT). Although considerations of descriptive economy seem to resolve the case in favour of solution A. we wish to suggest that they are irrelevant in the case at hand. Namely, it should be borne in mind that the criterion of the economy of description is applicable only to notational variants i.e. accounts which are equivalent in the sense that they make the same empirical predictions (Halle 1962). Thus, the two conceivable solutions at hand should first and foremost be evaluated as empirical hypotheses, which requires a very careful scrutiny of the empirical consequences they entail. They could be compared on the grounds of their relative simplicity only if it is first shown that they are genuine notational variants. As a matter of fact we intend to show that solutions A. and B. are not empirically equivalent notational variants since they differ substantially in their empirical consequences. Viewed in that perspective the seemingly less adequate (less economical) solution B. turns out to score much higher as an empirical hypothesis as it makes correct predictions concerning a range of important linguistic facts. Let us first consider the empirical consequences of solution A. (We are not very strict about keeping its two variants apart, since what we say about one could easily be rephrased to show its relevance for the other). The pivotal assumption of this solution t h a t the process at hand is a denominal operation, should be encoded in formal grammar as a very simple condition on the base of the relevant morphological rule: Condition C: the verbalizing suffix //i// attaches to hosts fulfilling the following structural condition: [ ]N Notice, that condition C entails no requirements with regard to the content of the brackets labelled as N. i.e. the verbalizing suffix //i// is allowed to attach to nominal stems regardless of their morphological structure a. [ ] N (5)

b. [[ ]SUFF] N SUFF: nominalizing suffix

c. [[ ]X SUFF]Ν X: a major syntactic class The morphological rule under consideration predicts that members of its base range from underived nominal stems (5a) to the outputs of the attachment of nominalizing suffixes to both unlabelled (5b) and labelled brackets (5c). At the same time, however, the condition C effectively bars the verbalizing suffix //i// from attaching to unlabelled

57

brackets (i.e. ROOTs) as all members of the base of the rule which attaches the suffix must be nominal stems. Both of the empirical predictions entailed by solution A. can easily be shown to be undesirable. First, there are several nominalizing suffixes in modern Polish but apparently the verbalizing suffix //i// does not attach to an already suffixed form. The range of permissible inputs to the morphological rule which attaches the verbalizing suffix //i// is much more constrained than predicted by solution A. (see the first empirical consequence of the Condition C given above). It is only configurations such as those listed in (5a) but never those in (5b) or (5c) that are well-formed-members of the BASE of the morphological rule in question. As it happens the underlined statement in the preceding paragraph is an important empirical finding about the morphological structure of verbal stems in modern Polish because the verbalizing suffix //i// is paralleled in this respect by some other verbalizing suffixes. Secondly, with regard to the other empirical consequence of condition C, there are numerous instances where the verbalizing suffix //i// apparently attaches to ROOTs rather than nominal STEMs. Consider the exemplary data in (6). UNIQUE ROOTS i.e. roots not attested outside verbal stems rose + i(e) "claim" CMC + i(c) "revive" bul + i(c) "pay" karc + i(c) "reproach" korc + i(c) "haunt" klec + i(c) "patch up" chyb + i(e) "miss" holub + i(c) "adore"

ROOTS also attested in members of other syntactic classes plae + i(e) "pay" (platny "not free") piesc + i(c) "caress" (pieszczota "a caress") pros + i(c) "ask" (prosba "request") lieζ + y(c) "count" (liezba "number") vrob + i(c) "do" (robota "work") krop + i(c) "drizzle" (kropla "a drop") dus + i(c) "suffocate" (duch+ota "stuffy air")

The evaluation of the empirical consequences of solution A. envisaged as an empirical claim leads to the following conclusions: Solution A. is a prototypical case of a bad empirical hypothesis as it is: too strong (as shown above it predicts the non-occurrence of certain well-attested phenomena i.e. attachment of the verbalizing suffix //i// to ROOTS) and at the same time too weak (it predicts the occurrence of certain non-attested phenomena i.e. attachment of the verbalizing suffix //i// to morphologically complex nominal stems). The distribution of the verbalizing suffix //i// in modern Polish is apparently restricted to ROOTs and morphologically simple (underived) nominal STEMS. Note, that it is exactly this state of affairs that follows from the competing hypothesis - the solution B. Here one important remark is in order. What ROOTs and underived (morphologically simple) STEMs have in common is their formal structure - both are morphologically simple, formally indivisible entities. In other words, given a morphological unit such as

58 walk (which shows up in both walk(a) "a fight" and walcz+y(c) "to fight") its morphological status judging on purely formal grounds is ambiguous i.e. it is either the R O O T [valk]R or a homophonous nominal stem [valk]N (i.e. walk(a) "fight" ). Thus it could be assumed that the verbalizing suffix / / i / / attaches to ROOTs only, some of which happen to be homophonous with nominal STEMs. This assumption when formalised as a morphological rule equivalent to solution Β gives the correct account of the distribution-of the verbalizing suffix //i// in modern Polish. Given the R O O T [valk]R two distinct morphological processes apply to it giving rise to verbal and nominal STEMs respectively (7)

a. [valk]R

[[valk]Ri]V

b. [valk]R

[[valk]R]N

The process of verbalization in (7a) is also responsible for forming configurations such as those listed in table (6) e.g. trzebi(c) "destroy" gubi(c) "lose", etc. Although we believe to have pointed out some rather obvious advantages of solution B. over the competing hypothesis A. the matter may still not be considered to be resolved. The reason why linguists of different theoretical persuasions adopted hypothesis A. (apart from its apparent simplicity) is the fact that it seems to capture a non-formal relatedness between the nouns and the corresponding verbs. Indeed it is obvious to a native speaker of modern Polish that words in such pairs as given below are related as far as their meaning is concerned. (8)

walk(a)N "fight" walczy(c)V "fight" hanb(a)Ν "disgrace" hanbi(c)\ "disgrace" szczerb(a) "notch" szczerbi(c)V "notch" kos (a)Ν "scythe" kosi(c)Y "mow" garb Ν "hump" garbi(c)V "stoop" gwalt Ν "rape" gwalci(c)~V "rape"

The morphological rule whereby verbs were derived directly from nouns (or vice versa) (i.e. solution A.) would allegedly capture both formal and semantic relatedness between the two STEMs. It would be all too easy to dismiss the above mentioned semantic intuitions on the grounds that since very little is known about meaning in general semantic considerations should have no place in formal grammar. All the more so that in fact it is not very difficult to show that whatever semantic data are available here they provide strong arguments for the novel solution (solution B) rather than the traditional account which relies on them heavily. Implicit in the Aronovian concept of a stem-based morphological rule adopted here is that not all cases of non-formal (semantic and grammatical) relatedness exhibited by stems should be accounted for by means of stem-based morphological rules. The latter account ONLY for recurrent structural patterns of both formal and non-formal (semantic and grammatical) nature. A closer scrutiny of the relevant data shows that very few genuine generalizations about semantic and grammatical relatedness between nouns and allegedly denominal verbs with the suffix //i// are to be disclosed. In fact the meaning of the nouns covers the whole range of conceivable nominal meanings

59 (9)

the name of the action

gwalcic V gwalt Ν "rape" walczyc V walka Ν "fight" chwycic V chwyt Ν "grasp" poscic V post Ν "fast" the result of the action pienic V piana Ν "foam" rzezbic V "carve" rzezba Ν "sculpture" lupic V "rob" tup Ν "booty" the instrument zqdlic V zqdlo Ν "sting" koste V "mow" kosa Ν "scythe grabic V grabie Ν "rake" the object kaleczyc V "harm" kaleka Ν "cripple" goscicV "host" go sc Ν "guest" the agent nianczyc V nianka Ν "nurse" Many relations are notoriously difficult to classify (e.g. ziqb "frost" ζίφίέ "freeze", hanba, hanbic "disgrace", zaszczyt, zaszczycic "honour"). Recall, that the recurrent semantic relation between a member of the BASE and the corresponding output is accounted for by means of a PARAPHRASE (Aronoff 1976) an indispensable component of a genuine stem based morphological rule. Note that for every group listed above we would need a different paraphrase e.g. Ν = an instrument used in V-ing Ν = the person who habitually V Ν = the action of V-ing not to mention the cases of wholly erratic (unique) relationship. It is not at all clear how the putative stem-based rule could in every case select the correct meaning relationship between the formally related stems 1 Furthermore, a stem-based morphological rule should unambiguously determine all the grammatical features of its output. Notice that neither verbal stems with the verbalizing suffix //i// nor the corresponding nouns exhibit uniform grammatical properties. VERBS - most of the verbs with the verbalizing suffix //i// are marked for the value "imperfective" of the category ASPECT e.g. walczyc "fight", hanbic "disgrace", gubic "lose", etc. There are many verbal stems with the verbalizing suffix //i// which exhibit the value "perfective" e.g. strzelic "shoot", rzucic "throw", skoczyc "jump", ruszyc "move". A putative rule of denominal verbalization would not determine the value of the category ASPECT of its output. The verbal stems at hand do not exhibit uniformity with regard to their syntactic characteristics, either. While some of them are intransitive (e.g. skoczc "jump", ruszyc "move", etc.) many are transitive (e.g. nianczyc "nurse"). 1 Should the reader still feel unconvinced as to the unpredictable nature of semantic relations between the nominal stems said the corresponding verbal stems with the verbalizing suffix //i// she (he) is kindly invited to consider the following examples: zldb "crib": ziobic "groove", czub "head": czubic "quarrel", rzezba "sculpture": rzezbic "carve", ziqb "frosty wind": ζίφίέ "freeze", sposob "way": sposobic "prepare", zrebiq "colt": zrebic "foal", grzmot "thunder": grzmocic "beat", bioto "mud": btocic "soil with mud", szcz$scie "luck": szcz^scic "have good luck".

60 NOUNS - the nominal stems under consideration vary with regard to the value of the category GENDER often even within a uniform semantic class: (10) names of action: gwatt "rape" (gwalcic) masculine donos "denunciation" (donosic) masculine walk(a) "fight" (walczyc) feminine names of instrument: zqdt(o) "sting" (zqdlic) neuter mydl(o) "soap" (mydlic) neuter kos (a) "scythe" (kosic) feminine names of result of action: pian(a) "foam" (pienic) feminine wyraz "expression" (wyrazic) masculine. Apparently grammatical properties of the verbal and the nominal stems under consideration as well as semantic relations between them are too erratic (idiosyncratic) to be accounted for by means of morphological rules which relate them directly (solution A.) Note, however, that the irregularity (unpredictability) at hand follows automatically from the assumption that both STEMs are derived from ROOT. ROOTs, it will be recalled, are purely formal entities devoid of either semantic or grammatical features; morphologically complex units derived from ROOTs are related to them only on the plane of form. Non-formal characteristics (grammatical features and semantic representation) are idiosyncratic properties of the STEMs unrelated to properties of any other units and not determined on the strength of grammatical generalizations (unless a STEM is morphologically derived from another STEM). All in all, for each pair such as those listed in (8) we have three morphological units: [valk]R : ROOT, a unilateral purely formal unit unspecified for syntactic class membership, [[valk]R i]V VERBAL STEM [[valk]R ]N NOMINAL STEM The STEMs are bilateral units endowed with semantic representation, syntactic class membership, grammatical characteristics (values of relevant grammatical categories, inflectional paradigm, etc.) The STEMs are formally related to the ROOT by means of morphological rules which perform a formal operation (e.g. attach a suffix). The non-formal relations between STEMs derived from the same ROOT lie outside the scope of morphology. To sum up our discussion of the verbalizing suffix //i//, both formal and non-formal properties of the relevant verbal and nominal stems point to the following conclusion: the morphological rule which governs the distribution of the verbalizing suffix //i// in modern Polish is subject to an important constraint on the BASE, namely the verbalizing suffix //i// attaches to morphological units labelled as ROOT [ ]R. We claim that it is an important finding about the morphological structure of the verbal system in Polish as the behaviour of the verbalizing suffix //!// is paralleled by some other ver-

61 balizing suffixes. The assumption that the verbalizing suffix //i// attaches to ROOTs only is instrumental in accounting for important morphological, semantic and grammatical properties of the verbal stems with the verbalizing suffix //i// which could be summarised as follows: Given a morphological configuration STEM of the following structure [[M] SUFF]V, where SUFF stands for a verbalizing suffix 1) Μ is a morphologically simple, unimorphemic underived structure. It may be a unique ROOT or a ROOT which appears as a component in other STEMs. It may also be homophonous with morphologically simple STEM. 2) The semantic relations between [[M] SUFF]V and all STEMs homophonous with Μ are highly irregular. 3) The grammatical properties of both [[M] SUFF]V and STEMs homophonous with Μ are not uniform. The three points given above are hallmarks of morphological rules which are subject to the constraint on their BASEs to the effect that all members of the BASE must belong to the category ROOT. As could be expected not all verbalizing processes in modern Polish are restrained in this manner. In the next section we shall have a somewhat closer look at a verbalizing process which is free to apply to STEMs, the process of the attachment of the verbalizing suffix (ej). An attempt will be made to show that root-based and stem-based morphological rules differ in their crucial characteristics.

3.3. The verbalizing suffix / / e j / / 3.3.1. T h e phonological properties Two points are worthy of note concerning the phonological properties of this suffix, which surfaces as [ej] (unless the glide is deleted via the phonological rule of j-deletion; Laskowski 1975, Gussmann 1978, Rubach 1984). First, the verbalizing suffix //ej//invariably causes the palatalization of the immediately preceding consonant: Velar Palatalization dzicze(c) "get wild" dzik(i) "wild": srog(i) "severe": sroze(c) "become severe' Anterior Palatalization rud(y) "red-haired": rudzie(c) "get red-haired" pstrokat(y) "motley": pstrokacie(c) "get motley" szkaradn(y) "disgusting": szkaradnie(c) "get disgusting' wesele(c) "cheer up" wesot(y) "cheerful": iysie(c) "get bald" lys(y) "bald": The rules of Palatalization are triggered off by the suffix initial segment.

62

Secondly, the suffix initial vowel is represented as tense front mid vowel at the underlying level. As pointed out in Rubach (1984:28) "In some cases phonetic [e] must be set up as underlying //e// as distinct from //e//. The point is that in surface terms some e's which alternate with [a] are set up as tense //e// while all the others are represented as //e//". Rubach points further on that the vowel in the verbalizing suffix //ej// is underlyingly tense [e] since it alternates with [a] in the surface forms e.g. lysiej(q) "get bald" 3p.pl. non-preterite: lysial "get bald" 3p.sing. preterite. The verbalizing suffix (ej) is entered in the Lexicon in its underlying phonological form as //ej//.

3.3.2. T h e B A S E of t h e morphological rule of / / e j / / suffixation In section 2.2. we discussed in some detail the morphological rule of verb forming //i// suffixation which, as we tried to demonstrate, is subject to an important constraint on the BASE, namely, the verbalizing suffix //i// attaches only to morphological configurations marked as ROOT. As would be expected not all processes of verbalization are restrained in this manner. Our purpose in this section is twofold. We want to discuss the morphological properties of the verbalizing suffix //ej// which, it will be shown, attaches to morphological units labelled as members of some (major) syntactic class i.e. STEMs. Furthermore, we would like to claim that morphological rules which apply to STEMs exhibit markedly different characteristics from rules restricted in their application to ROOTs. Let us start with the process of deadjectival verbalizing suffixation with the morpheme //ej//. We claim that the verbalizing suffix //ej// attaches to adjectival STEMs i.e. all morphological configurations which comply with the structural condition (11) [ ]A A: adjective The above hypothesis is tantamount to positing no restrictions with regard to the morphological complexity of the content of the labelled brackets in (11). In other words the BASE of the verbalizing process at hand could contain morphological configurations of two kinds. 1) The adjectival stems homophonous with ROOTs. In all such cases the brackets in (11) contain morphologically simple (indivisible) units. 2) The adjectival STEMs which arise due to the application of a rule of adjective formation. The brackets in (11) would then contain morphological configurations of the following structure [[ ]X SUFF] ADJ SUFF - an adjectivizing suffix [[ ] SUFF] ADJ X - a syntactic class symbol Now let us look at the relevant data (12) a. morphologically simple stems bogaty "rich": gruby "fat":

bogaciec "get rich" grubiec "get fat"

63 tlusty "thick": tlusciec "thicken" niemy "dumb": niemiec "get dumb" tani "cheap": taniec "cheapen" dr§twy "numb": dr^twy "get numb" czerstwy "stale": czerstwiec "get stale" trzezwy "sober": trzezwiec "sober up" nudny "boring": nudniec "get boring" t$py "blunt": t§piec "get blunt" wqtiy "slight": wqtlec "get slight" wesoly "cheerful": weselec "cheer up" dziki "wild": dziczec "get wild" b. stems formed due to the application of adjectivizing suffixes - ( - a t ( y ) ) the adjectivizing suffix (at(y)) attaches to nominal stems and to roots: popielaty "grey": popielaciec "get grey" (popiol Ν "ash") garbaty "hunchbacked": garbaciec "become hunchbacked" (garb N"hunch") pstrokaty "motley": pstrokaciec "become motley" kosmaty "hairy ": kosmaciec "become hairy" rosochaty "gnarled": rosochaciec "become gnarled" (ow-at(y)) the adjectivizing suffix (ow-at(y)) attaches to nominal stems and to roots: flakowaty "flabby": flakowaciec "get flabby" (flak Ν "flab") mechowaty "mossy": mechowaciec "get mossy" (mech Ν "moss") rogowaty "horny": rogowaciec "get horny" (rog Ν "horn") korkowaty "cork like": korkowaciec "get cork like" (korek Ν "cork") tykowaty "fibrous": tykowaciec "become fibrous" (lyko Ν "bast") lodowaty "icy": lodowaciec "get icy" (lod Ν "ice") rakowaty "cancerous": rakowaciec "become cancerous" (rak Ν "cancer") chropowaty "coarse": chropowaciec "coarsen" kartowaty "dwarfish": karlowaciec "become dwarfish" - (-om) the adjectivizing suffix (-om) attaches to nominal stems and to roots: ruchomy "moving": lakomy "greedy":

ruchomiec "become moving" (ruch Ν "movement") lakomiec "become greedy"

- (ist) the adjectivizing suffix (ist) attaches to nominal stems and to roots: wyrazisty "expressive": soczysty "juicy": srebrzysty "silvery": przejrzysty "transparent":

wyrazisciec "become expressive" (wyraz Ν "expression") soczysciec "become juicy" (soft Ν "juice") srebrzysciec "become silvery" (srebro Ν "silver") przejrzysciec "become transparent"

(n) the adjectivizing suffix (n) attaches to nominal stems and to roots: mocny "strong": mocniec "get strong" (moc Ν "strength") szkaradny "ugly": szkaradniec "get ugly" (szkarada Ν "ugliness") swobodny "free": swobodniec "get free" (swoboda Ν "freedom") giodny "hungry": gtodniec "get hungry" (giod Ν "hunger")

64 pouuszedni "common": lagodny "mild": widny "bright":

powszedniec "become common" lagodniec "become mild" widniec "get bright"

- (ow(y)) the adjectivizing suffix (ow(y)) attaches to nominal stems: woskowy "waxen": szafirowy "sapphire": purpurowy "purple":

woskowiec "get waxen" (wosk Ν "wax") szafirowiec "get sapphire" (szafir Ν "sapphire") purpurowiec "get purple" (purpura Ν "purple")

- (aw(y)) the adjectivizing suffix (aw(y)) attaches to nominal stems: kulawy "lame": rdzawy "rusty": dziurawy "leaky":

kulawiec "get lame" (kul(a) Ν "crutch") rdzawiec "get rusty" (rdz(a) Ν "rust") dziurawiec"get leaky" (dziura Ν "leak")

All the data exemplified above show that the verbalizing suffix //ej// attaches to adjectival stems regardless of their morphological complexity i.e. both to morphologically simple stems homophonous with roots and to stems derived via various processes of adjectivizing suffixation. Recall that the distribution of the verbalizing suffix //i// discussed earlier was much more restricted in that it would attach only to morphologically underived configurations. This difference between the morphological behaviour of the two verbalizing suffixes is accounted for by making different claims with regard to the BASE of the respective rules: the suffix //i// attaches to ROOTs while //ej// attaches to members of a major syntactic class (Adjectives) i.e. STEMs. Thus, the data in (12) confirm one of the empirical consequences of the hypothesis that the verbalizing suffix //ej// attaches to STEMs. The other empirical consequences of this claim concern the grammatical properties of deadjectival verbs with (ej) as well as the semantic relations between them and the corresponding adjectives. In both respects //ej// attachment as a stem-based morphological process turns out to be a mirror-image of the root-based //i// attachment discussed in section 2.2. First of all, let us compare two groups of pairs of derivationally related items. (13)

A. bogaty "rich": bogaciec "get rich" gruby "thick": grubiec "get thick" tlusty "fat": tlusciec "get fat"

B. walka "a fight": walczyc "to fight" gwalt "a rape": gwaicic "to rape"

For more pairs such as those in (13A) the reader is requested to go back to (12). The data exemplified in (13B) are discussed in section 2.2 (see also ftn 1). It will be recalled that the semantic relations between the nouns and the corresponding verbs in (13B) are not recurrent, i.e. they may vary considerably from pair to pair. Hence, one is hard put to it to find a semantic formula common to all cases of putative denominal verbalization exemplified in (13B). The putative rule of //i// attachment to nominal stems would,'of necessity, lack the SEMANTIC PARAPHRASE a crucial component part of any genuine stem-based morphological rule (Aronoff 1976). In the case at hand the formal operation whereby a verb arises due to the attachment of the

65 suffix //i// to a nominal host may not be accompanied by a corresponding operation on a semantic plane on the strength of which the meaning of new formation (the output) is defined on the basis of the meaning of the host. Notice, that in this respect the pairs given in (13A) are different in that the semantic relation between an adjective and a corresponding verb is recurrent throughout the column (more examples in (12). Here each verb denotes the process of acquiring the feature(s) denoted with the corresponding adjective. 2 Thus, the formal operation of creating verbal stems by attaching the suffix //ej// to adjectival stems is coupled with a corresponding operation on the semantic plane whereby the meaning of the verb is constructed on the basis of the meaning of the adjectival base. The meaning of all deadjectival verbs with the verbalizing suffix (ej) can be paraphrased (in the technical sense of Aronoff 1976) as follows: [[X] ADJ + ej] V the meaning of [ ]V: to be in the process of acquiring the feature(s) denoted by [X] ADJ (simplified as "to be in the process of becoming X") Although all the adjectival stems in (13A) are morphologically simple entities the paraphrase given above is also valid for all deadjectival verbs with the verbalizing suffix //ej // regardless of the morphological make-up of the stem. To see this let us look at one adjectival stem from each group given in (12): pstrokaciec (derived from the adjective pstrokaty "motley") means "to be in the process of becoming motley", lodowaciec (derived from the adjective lodowaty "icy") means "to be in the process of becoming icy", robaczywiec (derived, from the adjective robaczywy "wormy") means "to be in the process of becoming wormy", purpurowiec (derived from the adjective purpurowy "purple") means "to be in the process of becoming purple", wyrazisciec (derived from the adjective wyrazisty "expressive") means "to be in the process of becoming expressive", ladniec (derived from the adjective ladny "pretty") means "to be in the process of becoming pretty". The outputs of the stem-based morphological process of //ej// suffixation are not only uniform in their semantic representation, which is invariably the function of the semantics of the host, but also with regard to their grammatical features. All verbs formed by attaching the verbalizing suffix //ej// to adjectival bases exhibit the following properties: - they invariably display the value "imperfective" of the grammatical category ASPECT - they share the syntactic characteristics at least in as far as they are invariably intransitive verbs. Recall that verbal stems with the verbalizing suffix //i// axe different in both respects i.e. although most of them are "imperfective" quite a few are marked as "perfective". Also, there is no uniformity among them with regard to the syntactic characteristics 2

The PARAPHRASE of //ej// attachment is subject to a qualification to the effect that the object referred to by the subject in the sentence with the verb is the SOURCE of the changes which lead to the acquisition of the feature(s) denoted with the adjective.

66 (some are intransitive, other transitive). Unlike the process of //i// attachment the deadjectival verb forming suffixation by means of the suffix (ej) determines both the semantic and the grammatical properties of created verbs. We claim that all the differences between the two processes of suffixation at hand are the consequences of the fact that the BASE of //i// attachment is confined to ROOTs while the BASE of //ej// suffixation is made up of members of a major syntactic class i.e. STEMs. Constraining the BASE of a morphological process to members of the category ROOT or STEM entail markedly different empirical predictions whose correctness is in every case to be checked against the relevant data. The constraints seem to us to be of some explanatory value since they account for convergences of both formal (distribution) and non-formal (semantics, grammatical properties) characteristics of the constrained rules. The table belowe contains the hallmarks of two types of morphological processes under consideration. root-based

stem-based

morphological

morphological

processes

processes

the

a morphologically

the degree of the morphological

BASE

simple unit

complexity of the base is irrelevant

the

a set of items with

a uniform class with

OUTPUT

disparate grammatical properties

regard to grammatical properties

the semantic relations

the semantic relations

between the outputs

between the members of

and those STEMs which

the BASE and the

happen to be homophonous

corresponding outputs

with the members of the

are recurrent

BASE are highly erratic

(regular)3

3.4. Two more suffixes 3.4.1. Stem-based / / i / / suffixation The outputs of the root-based process of //i// attachment discussed in section 2 are not the only verbal stems in modern Polish in which the verbalizing suffix //i// appears. Let us consider verbs such as those in (14). All of them seem to be derivationally related to nominal stems. 3 This is not to deny the reality of a well-known process of lexicalization which may entail the acquisition of idiosyncratic semantic features by individual lexical items. It has been repeatedly emphasised that lexicalization lies outside the scope of morphology (Halle 1973, Booij 1977, Ba.uer 1983).

67 (14)

kozak "cossack": cygan "gypsy": chuligan "hooligan": furman "carter": tajdak "scoundrel": rybak "fisherman": gospodarz "host":

kozaczyc "behave like a cossack" cyganic "cheat" chuliganic "behave like a hooligan" furmanic "drive a cart" lajdaczyc "behave like a scoundrel" rybaczyc "to fish" gospodarzyc "keep house"

If indeed we assume that the verbal stems on the right hand side in (15) are directly derived from the corresponding nouns (or vice versa) 4 then this claim, it will be recalled, has at least three important empirical consequences. First, if the verbalizing suffix //i// is allowed to attach to nominal stems it joins brackets labelled as NOUN (i.e. [ ]N) regardless of the morphological complexity of the morphological unit enclosed in such brackets. This prediction seems to be borne out by the relevant data since the nominal stems in the right hand side column in (15) are both morphologically simple (homophonous with roots) as well as morphologically complex (derived by earlier rules of nominalization) e.g. cygan "gypsiF vs. rybak "fisherman" (ryb(a) "fish"). Second, the semantic relations between the members of the BASE and the corresponding outputs of a stem-based morphological process are regular (recurrent). The meaning of the verb could be paraphrased using the meaning of the noun as "behave in a manner characteristic of X" 5 . On the assumption that the direction of the derivation is reverse i.e. that we have to do with a deverbal back-derivation the meaning of the resultant nouns could be paraphrased using the meaning of the verbs as "a person who characteristically (habitually) Xs". It is not, however, the exact wording of the paraphrase that matters for our purposes but the fact that it seems to exist. Thus, the morphological rule at work here is a genuine stem-based process and the regular formal operation is accompanied by equally regular changes in the semantic representation. Third, whichever direction of the derivation we assume to be correct the outputs turn out to form a uniform class as fax as their grammatical features are concerned. VERBS - are invariably intransitive and are marked for the "imperfective" value of the category ASPECT. NOUNS - are all marked for the value "masculine" of the category GENDER. They are also invariably "animate" and "human" 6 . All in all, the process of //i// suffixation on nominal STEMs exhibits the characteristic properties of a stem-based morphological rule. 4

It is not the direction of the derivation that matters but the fact that the STEMS are directly related and NOT both derived from ROOT. 5

In several cases (e.g. cyganic "cheat", zydlaczyc "speak with a Jewish accent") idiosyncratic semantic features make this general meaning more specific by supplying the reading of the formula "characteristic behaviour"; the characteristic features of the Gypsies are different from those of the Jews. 6 The assignment to various inflectional paradigms is due to phonological factors alone (J. Rubach personal communication) e.g. walkonie "lazy bone" pi. vs. furmani "carter" pi. appears after coronal sounds including [j].

68 3.4.2. T h e root-based / / e j / / suffixation The verbalizing suffix //i// discussed in section 2 and 4.1 turns out to be introduced by both root-based and at least one stem-based morphological rule. Likewise, the outputs of the stem-based morphological process discussed in section 3 are not the only verbal stems in which the verbalizing suffix //ej// appears. Consider the following data: (15)

gor.e.(c) "burn" inf. (cf. gor.ej.(q) "burn" 3p.pl.) szal.e.(6) "go mad" inf. bol.e.(c) "feel pity" inf (cf. bol.ej.fa) "feel pity" lp.sing.) kul.e.(c) "hobble" inf. mdl.e.(c) "faint" inf. rozumi.e.(c) "understand" inf. smi.e.(c) "dare" inf. umi.e.(c) "know" inf.

Such formations are not very numerous in modern Polish. It seems that they are best accounted for by assuming a morphological rule by virtue of which the verbalizing suffix //i// attaches to ROOTs. Notice that the empirical predictions entailed by this hypothesis are well borne out by available data. First, the morphological entities to which //ej// attaches are morphologically simple (underived). It is true in equal measure with regard to "unique" roots (e.g. smi.e.(c) "dare" inf. or umi.e(c) "know" inf.) roots that turn up in other non-verbal stems (e.g. gor.e.(c) "burn" inf. cf. gor.qc(y) "hot") and roots homophonous with free stems (e.g. kul.e.(c) "hobble" cf. kul(a) "crutch"). Secondly, the semantic relations between the nominal stems homophonous with roots to which //ej// attaches and the corresponding verbs with the verbalizing suffix //ej// are far too erratic to assume that //ej// attaches to nominal stems e.g. : szal "madness" kul(a) "crutch" rozum "mind" bol "pain"

szalec "go mad" inf. kulec "hobble" inf. rozumiec "understand" inf. bolec "feel pity" inf.

The semantic relations between the items in the above pairs could hardly be conceived as recurrent. Thus, a putative rule of denominal //ej// attachment would lack a crucial component the SEMANTIC PARAPHRASE. The root-based morphological process of verb forming //ej// suffixation plays a marginal role in the morphological system of modern Polish, even in comparison with other root-based rules. Yet, it should not be confused with the stem-based process that introduces the same verbalizing suffix (discussed in section 3). The latter is a well-established, highly productive rule, which applies freely to adjectival stems to produce verbs with predictable meaning and grammatical properties. The scope of the former is not only limited to a handful of ROOTs but it steadily diminishes. Moreover, there axe interesting differences between the two groups of verbal stems with the suffix //ej// which concern

69 their inflection. In the non-preterite tense some of the Polish verbal stems are extended with a front vowel [i] or [e]. Rubach (1984) established that the [i] extension patterns with the verbalizing suffixes expressed by a front vowel i.e. //i// or //e// while the [e] extension with the remaining verbalizing suffixes. Neither of the extensions shows up in the 1 person singular or the 3 person plural. Notice that the verbal stems produced via the stem-based rule of //ej// suffixation invariably pattern with the vocalic extension in the non-preterite paradigm e.g. lysiec "get bald" inf.: lysi.ej.e.(sz) "get bald" 2p.sing., lysi.ej.(e) "get bald" 3p.sing., lysi.ej.e.(my) "get bald" lp.pl., tysi.ej.e.(cie) "get bald" 2p.pl. The root-based verbal stems with //ej//, on the other hand, are much more erratic in this respect. Some of them seem to be regular : e.g.

kul.e.(c) "hobble" inf. kul.ej.e(sz) 2p.sing. kul.ej.e 3p.sing.

kul.ej.e.(my) kul.ej.e.(cie)

lp.pl. 2p.pl.,

similarly bolec "feel pity" inf., mdlec "faint" inf., szalec "go mad" inf.. Others do not show any traces of the vocalic stem extensions in the non-preterite tense. e.g. rozumie(c) "understand" inf. rozum.e.(sz) 2p.sing. rozum.e.(my) lp.pl. rozum.e. 3p.sing. rozum.e.(cie) 2p.pl. (the vowel [e] in the above forms is what is left of the verbalizing suffix //ej// after the application of the phonological rule of j-deletion: Laskowski 1975, Gussmann 1978, Rubach 1984). Similarly: umiec "know" inf., smiec "dare" inf. The verbal stem gore(c) "burn" inf. seems to belong to both groups e.g. gor.e, gor.ej.e. are the alternative forms of the 3 person singular (both are considered to be appropriate by the normative dictionary "Slownik Poprawnej Polszczyzny"). In the diachronic perspective the lack of the vocalic extension in the non-preterite paradigm of some of the verbal stems with the suffix //ej// is a novelty. Klemensiewicz et al. (1981) give the following relevant form of rozumiec "know" in Old Polish: rozum.ej.e.(sz) 2p.sing. rozum.ej.e. 3p.sing.

rozum.ej.e.(my) rozum.ej.e.(cie)

lp.pl. 2p.pl.

Umiec "know" inf. and smiec "dare" inf. also used to behave regularly in this respect (Klemensiewicz et al. 1981). There exists another important difference between all verbal stems with the suffix //ej// derived from adjectival stems and some of the verbal stems with //ej// formed from roots. Compare the forms of the 1 person singular of the non-preterite tense of lysiec "get bald" and rozumiec "understand": [wip.ej.ew] : [rozum.e.m]. Rubach (1984) established the underlying representation of the ending [ew] as //Vm// (the exact phonological shape of the vowel is irrelevant for our purposes but see Rubach 1984). The crucial difference between the 1 person singular inflectional ending in lysiec

70 "get bald" and rozumiec "understand" is that in the latter case the morpheme at hand is underlyingly WITHOUT a vowel: //rozum.ej.m// (otherwise [j] would surface as there would be no context for j-deletion). Again, the modern form here is a historical novelty (Klemensiewicz et al. 1981 give the historical form as rozumi. ej. ς) Now, notice that both tendencies set apart the root-based verbal stems with the suffix //ej// from the stem-based verbal stems with the same suffix. At the same time they make the root-based verbal stems with //ej// analogous to another group of root-base verbal stems, those with the verbalizing suffix / / a j / / (discussed later in section 5.1). There are two points of analogy. First, as is well-known the non-preterite vocalic extension invariably fails to occur throughout the non-preterite paradigm of verbal stems with the suffix //ej//. Compare rozumiec "understand" and machac "wave": rozum.e.sz 2p.sing. rozum.e 3p.sing. rozum.e.my lp.pl. rozum. e. cie 2p.pl.

mach.a.sz 2p.sing. mach, a 3p.sing. mach.a.my lp.pl. mach. a. cie 2p.pl.

([j], it will be remembered, is present in the underlying representation to be deleted by virtue of the phonological rule of j-deletion). Second, all the verbal stems with the verbalizing suffix / / a j / / invariably pattern with the vowelless representation of the 1 person singular suffix e.g. rozum. e.m "understand" lp.sing. mach.a.m "wave" lp.sing. (in both cases the desinance-initial vowel would effectively block the rule of j-deletion). To conclude, the two groups of verbal stems with //ej// (one produced by a stem-based rule, the other by a root-based one) are gradually becoming more and more dissimilar as far as their inflection is concerned. At the same time, the outputs of the root-based morphological rule are slowly drawn towards another root-based morphological pattern. Both facts seem to show the relevance of the category ROOT in morphological analysis as well as the validity of the distinction between root-based and stem-based morphological processes. Our discussion of two verbalizing suffixes in modern Polish (sections 2,3,4): //i// (gwatci(c) "rape", rybaczy(c) "to fish") and //ej// (robaczywiec "to get wormy", umiec "know") leads to the following conclusions: - morphological rules fall into two classes: root-based and stem-based rules - the same suffix may be introduced by more than one morphological rule.

3.5. R o o t - b a s e d suffixes In the present section we shall analyse three Polish verbalizing suffixes : //e// (krzyczec "shout"), / / a / / (szczypac "pinch"), / / a j / / (machac "wave") and show they are invariably introduced by root-based morphological rules. In section 5.4 we shall consider various conceivable interpretations of the notion "a root-based morphological rule".

71 3 . 5 . 1 . T h e v e r b a l i z i n g suffix / / a j / /

The verbalizing suffix / / a j / / appears in full shape only in two word-forms of every relevant lexeme the present participle:

the 3 person plural non-preterite:

mach.aj.qc "wave" migaj.qc "twinkle" mrug.aj.qc "wink" mach.aj.q "wave" mig.aj.q "twinkle" mrug.aj.q "wink"

In all other word-forms the suffix final glide [j] is deleted as predicted by the phonological rule of j-deletion mach.a.m. "wave" lp.sing. 1 , , ., , , u „ _ > (non-preterite) mach, a wave 3p.sing. J mach. a. I "wave" 3p.sing. (preterite) strug.a.ny "whittle" (the past participle) The verbalizing suffix / / a j / / never triggers changes in the roots to which it attaches. The verbalizing suffix / / a j / / should not be confused with two other homophonous suffixes / / a j / / - the exponent of the derived imperfective (discussed in chapter 4) / / a j / / the exponent of repetitive action in stems such as bij.a.(c) "beat" inf. (cf. bij.aj.q "beat" 3p.pl.), pij.a.(c) "drink" inf. etc. 7 All the remaining verbal stems which contain the verbalizing suffix //aj// are derived by virtue of a morphological rule which attaches the verbalizing suffix / / a j / / to morphological units labelled as ROOTs. Consider the following representative sample of data: (16)

A. si§gac "reach" inf. scigac "pursue" inf. migac "twinkle" inf. szargac "bully" inf. drgac "shake" inf. machac "wave" inf. kichac "sneeze" inf. gtaskac "stroke" inf. trzymac "hold" inf. umierac "die" inf. narzekac "complain" inf. spotkac "meet" inf.

B.

rzygac "vomit" inf. slizgac "slide" inf. bluzgac "gush" inf. achac "cheer" inf. usmiechac "smile" inf. frygac "spin" inf. czotgac "creep" inf. troskac "care" inf. szlochac "weep" inf. stuchac "listen" inf. zyskac "profit" inf.

7 Such words are usually referred to as iterative verbs. They are produced by attaching the suffix //aj// to verbal stems e.g. //bij//: //bij+aj// (bijaj)q) "beat" 3p.pl. Another exponent of iterativity is the morpheme //iva// e.g. pisa(c): pisywa(c) "write". The distribution of the two exponents of iterativity mirrors that of the two homophonous exponents of the Derived Imperfective. The two suffixes distinct, by all appearances, in modern Polish are related historically (Klemensiewicz et al. 1981).

72

The hypothesis that the verbalizing suffix / / a j / / attaches to ROOTs accounts in a straightforward manner for the derivation of all items such as those listed in (16 A). An alternative solution for items in column (16 B) is to assume that they are directly related to corresponding nominal stems (e.g. usmiechac "smile" inf. is, on that claim, derived from usmiech "a smile") Note, however, the following facts. First of all, all morphological units that immediately precede the suffix / / a j / / are morphologically simple (underived). All the nominal stems from which some of the aj-verbs are allegedly derived are morpholagically simple rather than produced by an earlier nominalizing process. The hypothesis that / / a j / / attaches to nominal stems is thus empirically too weak as it predicts the attachment of / / a j / / is insensitive to the degree of morphological complexity of the stem, while / / a j / / seems to attach to roots only (some of which happen to be homophonous with nominal stems). 8 Secondly, the semantic relations between the verbal stems with / / a j / / and the corresponding nominal stems such as those listed in (16 B) are highly irregular and cannot be captured by a single simple formula which could act as a semantic paraphrase for the putative rule of denominal verbalization (or, assuming the reverse order of the derivation, the deverbal nominalization). The examples in (17) show that the meanings of the nouns cover the whole range of conceivable nominal meanings. (17) a. fryg(a) Ν "spinning toy " (cf. frygac "spin" inf.) the nominal stem denotes an agent b. strug Ν "plane" (cf. strugac "whittle" inf.) siodl(o)N "saddle" (cf. siodlac "saddle" inf.) the nominal stems denote instruments c. siad Ν "sitting posture" (cf. siadac "sit" inf.) usmiech Ν "smile" (cf. usmiecha6 "smile" inf.) kaszel Ν "cough" (cf. kaszlac "cough") the nominal stems denote actions Last but not least, neither the verbs nor the corresponding nouns form a uniform class with regard to their grammatical properties. It cannot be assumed, therefore, that the morphological rule of denominal verbalization with the suffix / / a j / / uniquely determines the grammatical features of the resultant verbs. VERBS Aspect: Although the vast majority of the verbal stems with the verbalizing suffix / / a j / / are marked for the value "imperfective" of the category ASPECT some of these stems are marked as "perfective" e.g. zyskac "gain", spotkac "meet" Syntax: many of the verbal stems with the suffix / / a j / / are intransitive (e.g. rzygac "vomit", wierzgac "kick", migac "twinkle", kaszlac "cough", szlochac "weep", klqkac "kneel", szczekac "bark") many are transitive i.e. require at least two arguments (e.g. glaskac "stroke", ciskac "hurl", targac "carry", dotykac "touch") 8 The data in (17) are fully representative in this respect, //aj// attaches to many roots homophonous with nominal stems but there is not a single case recorded in the Dictionary of Modern Polish of this suffix, following a morphologically complex nominal stem.

73 NOUNS The nominal stems under consideration belong to all three GENDER classes: masculine: sluch "hearing", czolg "tank", wynik "result" feminine: fryg(a) "spinning toy", trosk(a) "care" neuter: siodl(o) "saddle" Notice that some of the stems allegedly directly related to verbs with the suffix / / a j // are not nominal stems at all but rather members of some minor syntactic class (perhaps interjections) e.g.: achac "cheer" (ach! "ο gosh!") ciachac "cut" (ciach: the onomatopoeic sound imitating the sound emitted by working scissors) biadac "complain" (biada! the exclamation of pity, fear) pukac "knock" (puk! the onomatopoeic sound imitating the sound produced by knocking on the door). From all the above considerations it follows quite clearly that the verb forming suffixation with / / a j / / is a typical root-based morphological process.

3.5.2. T h e verbalizing suffix / / a / / The verbalizing suffix / / a / / plays a marginal role in the morphological system of modern Polish. It seems that it is invariably introduced by a single root-based morphological rule. Consider the following data: (18)

a. rqbac "chop" skubac "pluck" skrobac "scrape" pisac "write" wiqzac "bind" kazac "order" deptac "tread" taskotac "tickle" sypac "sprinkle" klepac "pat" szarpac "yank" kqpac "bathe"

b. szeptac "whisper" dziobac "peck" klopotac "trouble" karac "punish" tapac "catch"

All the empirical predictions entailed by the claim that / / a / / attaches to ROOTs are borne out by the data. The bulk of the morphological units with which the suffix //a// appears are "unique" roots, some turn up in other formations (e.g. taskotaö "tickle": iaskotk(a) "a tickle", pisac "write": pism(o) "writing") very few are homophonous with nominal stems (all such cases are listed in 19b). The nominal stems to which a-verbs could be claimed to be directly derivationally related (a conceivable solution alternative to ours) are all morphologically non-complex.

74 None of them is an output of an earlier rule of nominalization. Likewise, the semantic relations between the nominal and the verbal steins in (18 b) are non-recurrent and the chances of reducing them to a single semantic formula are slight. Last but not least, neither of the stem classes is grammatically uniform. All the available data, scarce as they are in the case at hand, seem to support our claim that the Polish verbalizing suffix / / a / / attaches to morphological units labelled as ROOT.

3.5.3. T h e verbalizing suffix / / e / / The verbalizing suffix //e// is similar to the verbalizing suffix //a// discussed in the preceding section in that it plays a relatively marginal role in the morphological system of modern Polish and also is invariably introduced by a single root-based process of suffixation. It differs from / / a / / , however, in that it systematically causes palatalization of the immediately preceding root final consonant. Velar Palatalization krzyczec "to shout": krzyk "a cry" lezec "to lie": legowisk(o) "a lair" Anterior Palatalization leciec "fly": lot "a flight" srebrzec "shine": srebro "silver" widziec "see": widok "sight" To see that the verbalizing suffix //e// attaches to ROOTs consider the following linguistic data: (19)

a.

rzec "neigh" brzmiec "sound" patrzec "look" kl§cze6 "kneel" milczec "keep silence"

b. szumiec "hum" zrec "devour" srebrzec "shine" lezec "lie" j^czec "moan" bzyczec "buzz" myslec "think" wolec "prefer" kaszlec "cough"

: szum "a rustle" : zer "fodder" : srebro "silver" : leze "lair" : "a moan" : bzyk "a buzz" : mysl "a thought" : wola "will" : kaszel "a cough"

The data in (19) are representative as far as the length of the two columns (a and b) is concerned. Namely, the suffix //e// attaches to just a handful of "unique" roots or roots that appear in other formations (e.g. [milk]R: milczec "keep silence", milkliw(y)

75 "reticent"). Yet, the fact that most roots joined by the suffix //e// to form verbal stems are homophonous with nominal stems does not in our view warrant the hypothesis that nouns and verbs in (19 b) are directly derivationally related i.e. one is the derivational source of the other. The empirical predictions of the claim that //e// forms verbs by attaching to nominal stems are not supported by the available data. First, note that the nouns in the right hand side column in (19 b) are invariably morphologically simple. If we claim that //e// attaches to nominal stems then the generalization that such stems are never derived by an earlier nominalizing process remains unaccounted for. Second, the semantic relations between nominal and the corresponding verbal stems in (19 b) are too irregular to warrant the claim that the meaning of a verb could be paraphrased (in the sense of Aronoff 1976) in terms of the meaning of the corresponding noun in the same manner for all pairs. The putative rule of denominal verbalization would thus lack the semantic paraphrase - a crucial component of any genuine stem-based morphological process.

3.5.4. T h e notion of t h e root-based rule An important property of root-based verbalizing processes is that they, unlike stem-based rules, invariably account for a limited number of morphologically complex units. All verbal stems in which the verbalizing suffix //a// appears can be listed while all verbal stems with the suffix //ej// in principle cannot. Conceivably, there are two interpretations of a morphological rule relevant for a limited number of items. A. Morphologically complex units are listed in the Lexicon in their full shape and whatever structural regularities they exhibit are extracted and expressed separately in the form of redundancy statements in the sense of Jackendoff (1975). Thus, redundancy rules do not participate in the derivation of morphologically complex items they account for but merely reduce the informational load of the Lexicon measured in terms of independent information (Jackendoff 1975). Given this interpretation of root-based processes all Polish verbal stems which contain verbalizing suffixes attaching to ROOTs (i.e. //e//, //a//, / / a j / / //i//) would be listed in the Lexicon. For each group of verbal stems there will be a redundancy statement to the effect that a well-formed verbid stem may consist of a ROOT followed by a verbalizing suffix e.g. V:[X]R+[e] B . Alternatively, a morphological rule relevant for a limited number of items may be interpreted as a morphological process whose BASE forms a closed SET i.e. it is specified by giving a full list of its members rather than defining properties which all its members must share. Therefore, we may say that the BASE of such a morphological rule is a SET rather than a CLASS. Morphological processes of this sort, unlike redundancy statements, take part in the derivation of complex lexical items they account for. What sets them apart from

76 morphological rules whose bases are defined as classes is that unlike the latter they apply to a limited number of morphological units. On this interpretation of the root-based verbalizing processes the verbal stems which contain //i//, //e//, / / a j / / or / / a / / are NOT listed in the Lexicon. Each verbalizing suffix is introduced by a morphological rule which consists of two elements: the BASE - a finite list (i.e. a SET) of ROOTs the MORPHOLOGICAL OPERATION - the attachment of a verbalizing suffix to ROOT. Since ROOTs are unilateral, purely formal units the rule lacks SEMANTIC PARAPHRASE which would determine the meaning of the output in terms of the meaning of the base. At first glance it is difficult to judge the superiority of one of the above interpretations. From the point of view of predictability they seem to be merely two different ways of expressing one simple truth, namely, which root patterns with which verbalizing suffix is totally unpredictable i.e. it is an entirely arbitrary (idiosyncratic) property of the root [koch]R that it is verbalized via the attachment of the suffix / / a j / / rather than, say, //a //, / / ! / / or //i//. Notice, however, one crucial difference between the two interpretations. Interpretation (A) implies that roots and verbalizing suffixes are listed together, hence it could be expected that they form relatively stable units. On the competing interpretation (B), however, verbalizing suffixes attach to ROOTs marked as members of the BASE of the relevant rule of suffixation. As far as roots pattern with the same verbalizing suffix on an entirely arbitrary basis this should result in: 1) some degree of synchronic variation i.e. there should be roots which pattern with more than one verbalizing suffix; 2) diachronic shifts i.e. in the diachronic perspective roots should, without difficulty, change their assignment from the BASE of one rule of root-based verbalizing suffixation to another. As a matter of fact, both phenomena are well attested. As far as synchronic variation is concerned the relevant data can be classified in two groups. First, there are several roots that pattern with two verbalizing suffixes //a// and / / a j / / . Consider the verbal stem of chlasta(c) "slap"; forms such as chlaszcz(q) "slap" lp.sing., chlaszcz(q) "slap" 3p.pl. point to the verbalizing suffix //a//, while forms such as chlasta(m) "slap" lp.sing., chlastaj(q) "slap" 3p.pl. indicate that the verbalizing suffix is / / a j / / . The form chlaszcz(q) is underlyingly //xlast+a+om// (the verbalizing suffix is followed by the inflectional ending). The sequence of two vowels is necessary to trigger the phonological rule of j-insertion, which, in turn, feeds the rule of I-otation accounting for the surface alternation [t]: [ts] (Gussmann 1978, Rubach 1984). Secondly, three of the very few roots which pattern with //ej // could also appear with the suffix //i //: umiec "know" rozumiec "understand", smiec "dare". In umiec and rozumiec //i// can be found in all inflectional forms of the non-preterite tense (with the exception of the 1 person singular where it never replaces //ej//) and also in the

77

infinitive form. All the forms with //i// are considered to be highly inappropriate by the normative grammarians (Buttler et al. 1986). In smiec //i// shows up only in the 3 person plural and it is thought of as acceptable (Buttler et al. 1986). Consider the following data: , ν ^ '

rozum+i+(sz) "understand" 2p.sing., rozum+i "understand" 3p.sing., rozum+i+(cie) "understand" 2p.pl., rozum+i+(q) "understand" 3p.pl.

Obviously the [i] that shows up on the surface in (20) is NOT a verbalizing suffix but a non-preterite stem extension suffix. The underlying representation of rozumisz is as follows: //rozum+i+i+sz//. The verbalizing suffix (the first vowel in the sequence) is deleted by the phonological rule of Vowel Deletion (discovered by Jakobson (1948). Although the verbalizing suffix //i// does not appear on the surface in the inflectional forms we have two pieces of evidence for its presence at the underlying level. First, recall that the non-preterite stem extension suffix //i// patterns with two verbalizing suffixes only (//i// and //e//). Second, the verbalizing suffix shows up on the surface in the infinitive form which for speakers who use forms such as rozumisz "understand" 2p.sing., umisz "know" 2p.sing. are umic "know" inf., rozumic "understand" inf.. The relevant diachronic data are notoriously difficult to obtain but some inferences may be drawn from descriptive studies of Klimek (1963) and Smiech (1967). With regard to roots changing their assignment from the BASE of one root-based process to another two tendencies can be observed. First, quite a few roots which used to pattern with the verbalizing suffix / / a j / / are nowadays suffixed with //a//. According to Klimek (1963) the verb czerpac "scoop" assumed the following inflectional forms in the non-preterite tense up to the 17th century: czerp+a+(m) lp.sing. czerp+a+(my) lp.pl. czerp+a+(sz) 2p.sing. czerp+a+(cie) 2p.pl. czerp+a 3p.sing. czerp+aj+(q) 3p.pl. Such forms undoubtedly indicate that the verbalizing suffix in the verbal stem czerpac was / / a j / / . Similarly to "czerpac" behaved verbs such as those in (21). (21)

kiamac "lie" kopac "dig" lapac "catch"

szarpac "yank" dziobac "peck" hfill kiapac "snap" (Klimek 1963)

All of the verbs in (21) contain the verbalizing suffix //a//in contemporary standard Polish. Second, many of the verbal stems that used to be made up of a root followed by a verbalizing suffix //a// now contain the verbalizing suffix / / a j / / . According to Klimek (1963) the verb kqsac "bite" used to have the following inflectional forms: kqsz.($) lp.sing. kqs.i.(my) lp.pl. kqs.i.(sz) 2p.sing. kqs.i.(cie) 2p.pl. kqs.i 3p.sing. kqsz.(q) 3p.pl. Given the infinitive form of kqsac all the above inflectional forms indicate the presence of the verbalizing suffix / / a / / in the underlying representation of the stem, //kons+a//.

78 Similarly to kqsac behaved other verbal stems such as those in (22) (Klimek 1963) (22)

zyskac "gain" gwizdac "whistle" miotac "hurl" ciosac "hew" wiklai "twine"

In contemporary standard Polish all these stems contain the verbalizing suffix / / a j / / . As a matter of fact, the synchronic variation of the type //smokt+a//: / / s m o k t + a j / / "suck" is a reflection of this ongoing linguistic change, it is well-known that language changes are reflected on the synchronic plane as language variation (Aitchison 1981). Although we cannot see an easy way of reconciling the view that root-based verbalizing processes in modern Polish are merely redundancy rules with both the synchronic and the diachronic data discussed above it is not tantamount to resolving the controversy at hand in favour of the competing interpretation. On the contrary the bulk of the relevant facts seem to be compatible with both interpretations since the overwhelming majority of roots form stable units with verbalizing suffixes (a rough estimate is that variation concerns one item in 100). To all appearances we must accept the validity of the distinction between the stem-based and the root-based morphological processes and acknowledge that the notion "a root-based morphological rule" admits (at least) two interpretations. Another intriguing problem needs to be touched upon. It will be remembered that in the foregoing discussion we distinguished two types of constraints on the BASE of a morphological rule. 1) The BASEs of some morphological rules consist exclusively of morphological units labelled as ROOTs while the BASES of other rules include members of (major) syntactic classes i.e. STEMs. 2) The BASEs of some morphological rules form a class i.e. they are defined on the basis of some properties which they share, while the BASES of other rules are made up of elements which make up a set, rather than a class, since, as they have nothing in common they must be enumerated (i.e. listed). Now, the two types of constraints seem to coincide in all the cases discussed so far in the following manner: - all stem-based morphological processes seem to be class-based processes at the same time e.g. the stem-based verbalizing suffix //ej// attaches to all stems which share the assignment to the same major syntactic class ADJECTIVE (with further qualifications pointed out in section 3) - all root-based morphological processes seem to be set-based e.g. the verbalizing suffix //a// attaches to a random assemblage of roots. An interesting question is whether the above convergence of rule characteristics is of logical or empirical nature i.e. do root-based processes operate on sets ex deflnitione or does it just so happen in some (perhaps even most) cases but is not necessarily so? One more morphological process must be considered before we return to this issue.

79 The verbalizing suffix //nin// attaches to a number of adjectival stems to form verbal stems with predictable meaning. Consider the examples in (23). (23)

slaby "weak" : stabnqc "grow weak" blady "pale" : blednqc "become pale" gtuchy "deaf' : gtuchnqc "get deaf' brzydki "ugly" : brzydnqc "become ugly" mi^kki "soft" : mi^knqc "get soft" chudy "lean" : chudnqc "get lean" rzadki "scarce" : rzednqc "become scarce"

The meaning relations between the adjectival and the corresponding verbal stems are recurrent and could be accounted for by a simple formula, which could serve as the SEMANTIC PARAPHRASE (in the sense of Aronoff 1976) of the relevant stem-based morphological rule. The SEMANTIC PARAPHRASE characterises the meaning of the derived verbal stem in terms of the meaning of the corresponding adjectival stem as "to be in the process of acquiring the features denoted by AD J" (Laskowski et. al 1984). Note that the resultant verbs form a uniform group as far as their grammar is concerned as they are invariably intransitive and imperfective. Recall that the semantic formula given above is identical to the semantic paraphrase suggested for the stem-based process of verb derivation by means of the suffix //ej // (section 3). However, there is a crucial difference between these two stem-based morphological rules. It will be remembered that //ej// attachment is a class-based process in that it seems to apply to all morphological units fulfilling certain conditions. The co-functional process of //nin// attachment, on the other hand, seems to be a typical example of a set-based morphological rule. The generalization at hand, although fully transparent on both formal and semantic plane applies to a limited number of adjectival stems; the ONLY idiosyncratic feature of the complex stem slabnq(c) "grow weak" is that it exists. The only way to characterise the BASE of the rule in question is to list its members. Now we can return to the problem (posed in section 5.4) of the correlation between the two types of constraints on the base of morphological rules discussed in the present chapter. Recall, that we argue for the validity of the twofold division of all morphological rules: set-based versus class-based, root-based versus stem-based. As indicated above there is a strong empirical correlation between the first two constraints in each pair i.e. root-based processes tend also to be set-based. This is due to the a priori limited possibilities of roots to form classes-groups of units characterized by common properties. Roots are purely formal units-sequences of phonological segments and the only properties they could conceivably share are elements of their phonological shape. The other two constraints from each pair tend to correlate as well i.e. it is probably true of the majority of stem-based morphological rules that their bases form a class as it is the case with //ej// attachment. However, the morphological process of

80 //nin// attachment to adjectival stems discussed above shows the contingent nature of this correlation. The table below sums up our discussion of the correlation between the two types of constraints on the BASE of morphological rules. stem-based process root-based process

the BASE is a set attested, but probably not very numerous typical correlation

the BASE is a class typical correlation not inconceivable but non-attested

Chapter IV Constraints on the operation of morphological rules - deverbal suffixation in Polish 4.1. The constraints The most interesting constraints on the operation of morphological rules that have been proposed in the literature concern: - the sensitivity of morphological rules to the morphological structure of the BASE, - the ability of morphological rules to induce morphologically conditioned changes to the BASE. Aronoff and Sridhar (1987) claim that all morphological rules which apply to members of (major) syntactic classes (i.e. stem-based morphological rules in our terms) are "blind to the internal morphology of the words to which they attach" (Aronoff and Sridhar 1987: 13). It is an interesting, although extremely strong hypothesis. A more relaxed version of this claim is implicit in the Adjacency Condition advanced in Siegel (1977) and defended in Allen (1978) and in the Atom Condition proposed in William (1981). The Adjacency Condition allows the morphological operations to refer to the outmost formative in the hierarchical structure of the BASE while the Atom Condition states that they be allowed access solely to the head formative. What the two constraints have in common is that they allow morphological rules to be sensitive to at most one formative in the BASE. Aronoff and Sridhar (1987) advance the claim with regard to all morphological rules that apply to members of (major) syntactic classes to the effect that they never trigger readjustment rules (i.e. allomorphy rules or rules of truncation) in the BASE. In the present chapter we intend to show that, contrary to Aronoff and Sridhar (1987) claim, stem-based morphological processes can be sensitive to the non-morphological make-up of the BASE. What is more, unlike predicted by both Atom Condition and Adjacency Condition they are often conditioned by information associated with more than one BASE formative. Moreover, stem-based morphological processes systematically fail to obey the other constraint advanced by Aronoff and Sridhar (1987) in that they repeatedly induce readjustment rules of both kinds i.e. allomorphy and truncation. Our data come from contemporary standard Polish. The following morphological processes shall be discussed at length: - the process of the attachment of the suffix (1) whereby one of the participles is formed (e.g. zgnic "rot": zgniiy "rotten"). The participle is referred to in Tokarski (1978) as "imieslow przesziy przymiotnikowy" (adjectival past participle). For the sake of convenience we shall be referring to it as the (1) Participle, - the process accounting for the distribution of the suffixal exponents of the value "imperfective" of the category ASPECT attached to verbal bases exhibiting the feature "perfective". In the generative tradition resultant complex word-forms are usually referred to as Derived Imperfective.

82

4.2. The (I) Participle (1) Participle is the least known of many participles in modern Polish. As pointed out in Tokarski (1978: 196): "The adjectival past participles are not merely remnants of a no longer productive morphological rule (...) a substantial number of these word-forms have preserved transparent relations with corresponding verbs so they should be considered participles even from the purely synchronic modern perspective" (translation mine). The meaning of such participles is given in Grzegorczykowa and Puzynina (1981: 135) "They denote the states of the subject resulting from the process referred to by means of the verb" (translation mine). This semantically uniform group should be kept distinct from two other cases of deverbial suffixal derivatives with the suffix (1). First and foremost it should be kept distinct from the complex word-forms in which the suffix (1) is the exponent of the value "preterite" of the category TENSE. In other words, we claim, following Tokarski (1978) that the two complex word-forms (zgnil) exemplified in the two sentences in (1) are derived via two separate processes of suffixation. (1)

Jabtka zgnily, bo Fryderyk zostawil je na deszczu. (The apples have gone bad because Fyderyk left them in the rain) Fryderyk lubi zgnile jabtka. (Fryderyk likes rotten apples)

The process of the Preterite ( 1 ) attachment differs from the process of the Participle ( 1 ) attachment with regard to readjustment rules they trigger in their hosts. Notice that in the preterite forms of verbs the stem-final / e / alternates with / a / depending on the value of the following consonant e.g. wylysiec "grow bald"; wylysieli: wylysialy (preterite). As pointed out in Laskowski (1975) this alternation is accounted for by positing the underlying //e// in the verbalizing suffix //ej//, which is turned to / a / in a non-palatal context (e.g. before ( 1)). Yet, no such alternations are attested in the case of ( 1 ) Participles, e.g. wytysiate kobiety "bald women", wytysieli m$zczyzni "bald men". Apparently, the simplest solution is to set up a rule of readjustment whereby prior to the application of phonological rules the underlying tense [e] of the verbalizing suffix //ej// is laxed. This rule would be triggered by the suffix of (1) Participle but obviously NOT by the suffix of (1) Preterite. Secondly, although both (1) Participle and (1) Preterite trigger the readjustment rule truncating the verbalizing suffix (ng.) there are crucial differences in the application of the rule in the two cases. The verbalizing suffix (n^) is automatically deleted in the context before the (1) Participle i.e. a sequence (nq)(l) (where (1) is the exponent of the (1) Participle) is ungrammatical. Consider some examples of the relevant participial forms of verbs with the verbalizing suffix (nq): (2)

opuchnqc "swell": opuchly "swollen" ostabnqc "weaken": oslably "weak"

83 oslepnqc "go blind": osleply "blind" zarosnqc "grow over": zarosly "overgrown' zmoknqc "get soaked": zmokly "soaked" On the other hand, the truncation of the verbalizing suffix (n^) seems to be lexically governed in the context before the Preterite (1). Verbal stems with the verbalizing suffix (n^) seem to fall into three groups with regard to the application of the (nq.) truncation in this context. The first group: the (nq) truncation is inapplicable: stuknqc "knock": stuknql, *stukl "he knocked". Also consider: si^gnqc "reach", krzyknqc "shout", zagadnqc "speak up", skrzypnqc "squeak", zamknqc "shut", wrzasnqc "cry". The second group: the (n^) truncation is obligatory: rosnqc "grow": rosi, *rosnql "he grew". Similarly: ochlonqc "cool down", przysi$gnqc "swear", chudnqc "grow lean", slepnqc "go blind", zgadnqc "guess". The third group: the (nq.) truncation is optional: wyschnqc "dry up" wysechl, "he dried". Similarly: schnqc "dry", kisnqc "rot", wi$dnqc "wither".

wyschnql

In the diachronic perspective there is a lot of variation in the assignment of verbal stems to the first and the second group. Many verbs that used to belong to the first group are nowadays members of the second group e.g. ucichnqc "hush" kwitnqc "blossom". The opposite cases (i.e. once members of the second group now in the first one) are also well-attested e.g. dotknqc "touch", zacisnqc "fasten" rozciqgnqc "stretch" (Klemensiewicz et al. 1981:383). The other group of complex word-forms from which outputs of the (1) Participle suffixation should be kept distinct are the numerically insignificant adjectives with the unproductive adjectivizing suffix (1). All the members of the latter group are given below. (3)

dbaiy "careful" niedbaly "careless" udaty "successful" smialy "bold" niesmiaiy "shy" zrozumialy "clear" wytrzymaly "strong" zapami^taly "stubborn' trwaiy "durable"

ulegiy "mild" rosly "strong" dorosly "adult" upadly "immoral" czuly "tender" rozwiqzly "promiscuous' podniosly "lofty" wyniosly "haughty" staly "stable"

The differences between this group of adjectives and the class of (1)Participles concern their semantic and morphological properties. It has already been pointed out that the meaning of (l)Participles is regular i.e. predictable on the basis of the meaning of the base. On the other hand, the meaning relations between the adjectives with the suffix (1) and the corresponding verbal stems are highly erratic. It is true not only with regard to pairs such as rozwiqzac "solve": rozwiqzly "promiscuous", stac "stand": staly "stable", udac si§ "succeed": udaty "successful", ζαρατηίςίαέ "remember": zapami$taly "stubborn", whose meanings are hardly related at all any more, but also pairs such as e.g. zrozumiec "understand" zrozumialy

84 "understandable". The meaning of zrozumialy is irregular since it does not refer to the subject of the action denoted with the verb zrozumiec "understand" as is regularly the case with (1) Participles. The closest to the regular pattern are formations such as upadly "immoral", which divert from it merely by the restricted scope of their usage, upadly being related to only one of the meanings of the verb upasc "get corrupted". The morphological differences between the two patterns concern both the bases and the outputs. First, notice that (l)Participles are regularly formed only from verbs marked for the value "perfective" of the category ASPECT, while some of the adjectives with the suffix (1) are related to imperfective verbal stems: rosnqc "grow" imperf.: rosiy "grown", dbac "take care" imperf.: dbaiy "careful", smiec "dare" imperf.: smialy "daring", trwac "endure" imperf.: trwaly "enduring", czuc "feel" imperf.: czuly "sensitive". Second, only the adjectives with the (!) suffix (but never the (1) Participles) are subject to two fully productive and regular morphological processes: adverb formation (all Polish adjectives are subject to adverbalization) and the attachment of the nominalizing suffix (ose) which derives names of qualities. Consider a typical (l)Pariciple wyzdrowialy (wyzdrowiec "recuperate") and a typical adjective with the suffix (1) smialy "bold" *wyzdrowiale *wyzdrowialosc

smialo "boldly" smialosc "boldness"

Many of the (ose) derivatives based on adjectives with the suffix (I) are highly lexicalized e.g. czulosc "tenderness", trwalosc "durability", ulegiosc "docility", wynioslosc "haughtiness", stalosc "immutability", wytrzymalosc "perseverance". All in all, the exponent of the value "preterite" of the category TENSE, the exponent of one of the Participles and one of the adjectivizing suffixes are three distinct suffixes, which, although homophonous vary in the morphological properties to such an extent as to require account in terms of three distinct morphological rules. Now let us consider the problem of the BASE of the (1) Participle attachment. Tokarski (1978) points out that the suffix is found with three groups of verbal stems. Although predominantly it attaches to verbal stems with the verbaliziling suffixes //ej// and (n^) it also sporadically joins verbal stems derived with other verbalizing suffixes. Consider the examples in (4) (after Tokarski 1978:196) (4) a)

b)

osiwiec "get grey": odmlodniec "get young": postarzec si§ "get old": potlusciec "get fat": sparszywiec "get rotten": wyzdrowiec "recover": zbielec "grow pale":

osiwiaiy "grey" odmlodnialy "young" postarzaly "old" potluscialy "fat" sparszywialy "rotten" wyzdrowialy "recovered" zbielaly "pale"

ogluchnqc "go deaf': ogluchly "deaf' odwyknqc "get out of habit": odwykty "free of habit" oslepnqc "go blind": osleply "blind" przywyknqc "get used": przywykly "used to" skisnqc "go sour": skisly "sour"

85 c)

oblezc "flake": oblazty "flaked" osiq.sc "settle": osiadly "sedentary" umrzec "die": umariy "dead" The two big subgroups of the BASE of (1) attachment exemplified in (4a) and (4b) should not be treated on a par, since they differ fundamentally with regard to productivity. Two observations help to make this point clear. First, there are a number of verbal stems with the verbalizing suffix (ng,) which do not form grammatical complex words with the suffix of the (l)Participle e.g. p^knqc "burst", wypoczqc "rest", ominqc "avoid" uplynqc "pass", zaginqc "get lost" wytchnqc "sigh". It seems that the verbal stems with the verbalizing suffix (nq.) which are subject to suffixation via the (l)Participle attachment form a closed set rather than an open class. On the other hand, whenever the (l)Participle suffix attaches to a "perfective" verbal stem with the suffix //ej// the resultant complex word is grammatical. Particularly convincing examples are to be found in Nowotny-Szybistowa (1973) who list the otherwise unattested participles with (1) suffix employed by the Polish writer Stanislaw Ignacy Witkiewicz. Some of her data are quoted in (5). (5)

okretynialy "imbecile" przeinteligentniaty "too clever" przemozgowialy "over-brainy" skociaty "cat-like"

skorkowacialy "cork-like" splciowialy "very sexy" zjesienniaty "autumn-like" znowotworzaly "cancerous"

Note that all of the participles in (5) are derived from verbal stems with the verbalizing suffix //ej// e.g. okretynialy "imbecile": okretyniec "become imbecile". The morphological structure of the latter is as follows: [o[[kretyn]N.ej]V]V The verbal stems with the verbalizing suffix //ej// also constitute the derivational BASE for the four participles with the suffix (1) quoted as attested in Smolkowa and Tekiel (1981) - the record of novel words in contemporary Polish: (6)

zakompleksialy "complexed" zjesienialy "autumn-like" skompleksialy "complexed" porudziaty "russet"

The verbal stems with the verbalizing suffix //ej// which are subject to (1) suffixation form a class rather than a set; the common class-defining feature is the morpheme //ej//. Secondly, the function of the (1) Participle is gradually being taken over by the exponent of the passive participle (Bartnicka 1970) as exemplified by the following cases of synchronic variation Verb infinitive zmoknq6 "get soaked" zmarznqi "freeze" zwi%dnqc "whither" zgnic "rot"

(1) Participle zmokiy zmarzly zwi^dly zgnily

Passive Participle zmokni^ty zmarzni^ty zwiqdni^ty zgnity

Notice that the encroachment of the new exponent of this category is restricted to verbal stems with verbalizing suffixes other than //ej//.

86 Thus, the participle forming suffix (I) seems to attach productively only to perfective verbal stems with the verbalizing suffix //ej// and unproductively to a number of perfective verbal stems the bulk of which contain the verbalizing suffix (nq). Note that the ej-verbs concerned are only those derived by the stem-based rule of / / e j / / attachment but not those produced by the root-based rule of //ej// suifixation. 1 This, incidentally, provides additional evidence for the validity of the distinction between stem-based and root-based morphological processes. Summing up, the productive attachment of the deverbal (1) Participle suffix is sensitive to two factors: - the verbal base must be marked for the value "perfective" of the category ASPECT, - the verbal base must contain a STEM followed by the verbalizing suffix //ej//. These apparently undeniable facts are hardly compatible with the familiar restrictions on the operation of morphological rules. Given that the (1) Participle attachment is a stem-based process, the constraint proposed by Aronoff and Sridhar (1987) effectively forbids its access to the morphological make-up of the base. As shown above, however, the rule seems to be conditioned by some essential aspects of the morphological structure of complex word-forms it applies to. Furthermore, the assumption shared by the less restrictive conditions on the rule operation advanced by Siegel (1977) (i.e. the Adjacency Condition) and Williams (1981) (i.e. the Atom Condition) that morphological rules have access to a single base formative at most is also disconfirmed by the process at hand. Notice that the unprefixed verbs with the verbalizing suffix / / e j / / are systematically "imperfective". It seems that the rule of //ej// suffixation determines the value for the category A S P E C T of its outputs as "imperfective". The feature "perfective" on the presence of which the application of (1) attachment hinges crucially is invariably introduced by the rules of prefixation. Consider the verb rudziec "get red-haired" and some derivatives thereof: rudziec: [[rud] ADJ ej]V "imperfective" zrudziec: [z[[rud] ADJ ej]V]V "perfective" porudziec: [po[[rud] ADJ ej]V]V "perfective" *rudzialy vs. zrudzialy, porudzialy "red-haired" The (1) attachment is only successful in the case of "perfective" verbal stems. From both the Atom Condition and the Adjacency Condition it follows that the operation of (l)Participle attachment may have access to the P R E F I X only. Note that the P R E F I X is the head formative (it determines both the syntactic class 2 and the 1

Out of the very few root-based formations with the verbalizing suffix //ej// only dojrzec "ripen" and zgorzec "burn down" seem to allow the (1) Participle attachment while the remaining ones clearly do not. Some of the root-based ej-verbs are subject to adjective-forming suffixation with the suffix (1) e.g. zrozumialy "clear" smiaiy "brave" etc. 2 Alternatively, one may claim that zrudziec has no head (i.e. single formative determining all the properties of the whole complex word) since the prefix determines the value of the category ASPECT and the verbalizing suffix - the syntactic class. This interpretation is strengthened by the fact that many of the prefixes can appear with members of various syntactic classes e.g. pod.patrzec "to peep" pod.ziemny "subterranean", pod.ziem.ie "an underground".

87 value of the category ASPECT) and simultaneously the outermost formatives in the hierarchical structure of hosts. Yet, as demonstrated above the productive (l)Participle attachment requires access to information associated with two base-formatives: not only the PREFIX (the feature "perfective") but also the VERBALIZING SUFFIX (the phonological shape of //ej//). Thus, not only Aronoff and Sridhar's (1987) claim but also both the Adjacency Condition and the Atom Condition are too strong restrictions on the operation of the morphological process of (l)Participle formation in modern Polish.

4.3. Derived Imperfective Rules which account for the distribution of the exponents of Derived Imperfective in Polish provide further evidence against all the constraints on the operation of morphological processes under consideration. The so-called Derived Imperfective is a well-known morphological category in Slavic languages. Comrie (1976:88) points out that "in Slavonic (...) there exists (...) a process of imperfectivization whereby imperfective verbs can be derived by suffixation from prefixed perfectives". Comrie's statement discloses the first factor which governs the operation of the rules of Derived Imperfective formation: the Derived Imperfective suffix(es) attach(es) only to bases which axe marked for the value "perfective" of the category ASPECT. We claim that in modern Polish the category Derived Imperfective has two exponents, whose distribution is governed by the morphological structure of the verbal base, in particular by the phonological shape of the verbalizing suffix. The two exponents of the Derived Imperfective can be seen in the following examples: rozbijac "break up", wyrqbywac "cut out". rozbijac "break up" bic "break": rozbic "break up" perf.: rozbijac "break up" imperf. The morphological structure of the stem in rozbijai "break up" is given in (8): (8)

[[roz[[bij]R]V imperf]V perf.aj]V imperf.

The root is preceded by the perfectivizing prefix and followed directly by the exponent of the Derived Imperfective since bic "break" (and all its prefixal derivatives) belong to the group of root-verbs i.e. the verbalizing suffix is not realised phonologically. wyrqbywac "cut out" rqbac "cut": wyrqbac "cut out" perf.: wyrqbywac "cut out" imperf. In wyrqbywac the root is followed directly by the sequence (ywa) at least part of which is apparently an exponent of Derived Imperfective. The morphological structure of what the graphemes (ywa) stand for requires further investigation. We shall also show what happens to the verbalizing suffix //a// present in both rqbqc "cut" and wyrqbac "cut out" perf.

88 Let us discuss the two exponents of the Derived Imperfective in turn. The first exponent of the Derived Imperfective - the suffix / / a j // attaches to two groups of verbal stems. The first group comprises root-verbs. Consider one example from each subgroup of root-verbs in (9) (9)

plesc "weave": klasc "put": pasc "graze": gryzc "bite": moc "can": drzec "tear" 3 : myc "wash" 4 :

wyplatac "weave" DI przektadac "put" DI wypasac "graze" DI rozgryzac "bite" DI przemagac "can" DI rozdzierac "tear" DI przemywac "wash" DI

The second group comprises i-verbs i.e. verbal stems with the verbalizing suffix //i//. Recall that the distribution of the verbalizing suffix //i// is governed by several morphological rules (two of them were discussed in chapter 3). The resultant verbal stems invariably require / / a j // as the exponent of the Derived Imperfective. Compare the examples from the three main groups of i-stems in (10) (10) Perfective Verb Infinitive Derived Infinitive walczyc "fight" przewalczyc przewalczac "fight off" "fight off" pijaczyc "brawl" rozpijaczyc rozpijaczac "accustom to drink" "accustom to drink" — uwspotczesnic uwspolczesniac "modernise" "modernise" The linear morphological structure of the verbal stems in the right hand side column in (10) is as follows: pre+ROOT+i+aj

i: the verbalizing suffix aj: the exponent of the Derived Imperfective

The presence of the verbalizing suffix //i// at the underlying level in all such complex word-forms is manifested on the surface in two groups of verbal stems. After the root final labial consonant (i.e. / p / / b / jij / v / ) the verbalizing suffix //i// surfaces as the glide / j / . e.g. topic "melt": wytapiac "smelt" [vitapjatc] robic "make": wyrabiac "produce" [virabjatg] The root final coronal consonants are subject to the phonological rule of Iotation which applies before / j / inserted before a sequence of two vowels (Rubach 1984) e.g. natluszczac "make greasy": / / n a + t v u s t + i + a j / / . 3 4

The vowel (e) in the infinitive drzec "tear" is inserted by an allomorphy rule.

Jerzy Rubach has pointed out to me that (w) in cases such as przemywac "rinse" is introduced by some phonological rule.

89 The presence of the verbalizing suffix //i// is crucial for triggering the rule of i-insertion which feeds Iotation. 5 The other exponent of the Derived Imperfective appears in all the remaining groups of verbal stems. It is manifested on the surface as either /iva/ or /iva/ and it seems that the initial vowel is represented as //i// at the underlying level, while the front variant / i / is derived via the phonological rule of Fronting (Gussmann 1978, Rubach 1984). A much more intriguing problem is that of its morphological representation. The solution we would like to suggest is that the surface sequence (ywa) has a bimorphemic structure. The exponent of the Derived Imperfective is followed by the formative / a / or / a j / . The latter is the verbalizing suffix; in some groups of verbal stems it is subject to rules of re-adjustment. First, consider the data in (11): (11)

a. machac "wave": wymachiwac "brandish" DI the verbalizing suffix / / a j / / (machaj(^)) b. kopac "dig": wykopywac "dig out" DI the verbalizing suffix //a// (kopi(^)), c. malowac "paint": przemalowywac "repaint" DI the verbalizing suffix //a// obligatorily preceded by //ov//.

For all the verbs exemplified in (11) we may assume that the exponent of the category Derived Imperfective is the formative //iv// inserted immediately to the left' of the verbalizing suffix. We discuss the three groups exemplified in (11) in turn. (11a). The segment (a) in the surface sequence (iwa) is the affix-initial vowel of the verbalizing suffix / / a j / / . The affix-final segment / j / in this suffix is deleted as predicted by the phonological rule of j-deletion. The morphological structure of wymachiwac, which exemplifies the structure of the Derived Imperfective forms in this group is as follows: / / v i + m a x + i v + a j / / ( l i b ) . The segment (a) in the surface sequence (ywa) is a separate formative - the verbalizing suffix //a//. The morphological structure of wykopywac typical of Derived Imperfective forms of this group of verbal stems is as follows: //vi+kop+iv+a//. (11c).The morphological structure of the surface sequence (owa) is complex - the verbalizing suffix (a) is preceded by the formative (ow). The exponent of the Derived Imperfective splits this sequence up as it is inserted immediately before the verbalizing suffix (a). The morphological structure of przemalowywac which exemplifies the structure of Derived Imperfective forms in this group of verbal stems is as follows: //prze+mal+ov-Mv+a//. Thus, it turns out that the surface sequence (iwa) ((ywa)) is either //iv+a// (groups l i b and 11c) or / / i v + a j / / (group 11a) - the exponent of the category Derived Imperfec5

A small group of verbs whose unprefaced stems consist of ROOT followed by the verbalizing suffix //ej// also take //aj// in the Derived Imperfective. Incidentally, we have another instance of the significance of the distnction between root-based and stem-based //ej// attachment as the latter is exempt form the Derived Imperfective suffixation (Laskowski et. al. 1984). As a relevant example of root-based //ej// suffixation consider the verb siac "sow" (in the infinitive the vowel / a / displaces / e j / as a result of allomorphy): siac wysiac wysiewa£ (wy+siew+aj+3).

90 tive followed by a verbalising suffix. The assumption that the separate formative / / a j / / is present in the underlying structure of the surface sequence (ywa) ((iwa)) has serious consequences with regard to the morphological structure of the inflectional forms in the non-preterite paradigm. Above all, recall that the vowelless exponent of the 1 person singular (i.e. / / m / / ) appears only after the verbalizing suffix / / a j / / , while all the other verbalizing suffixes require //Vm// here. Two points are noteworthy in this respect: First, as shown in Smiech /1974) there are a number of verbs which used to have the verbalizing suffix / / a j / / but have //a// nowadays instead (e.g. kopac "dig"). Accordingly, the form of the 1 person singular non-preterite of such verbs used to contain / / m / / while now it contains //Vm// e.g. kopam "dig" lp.sing. / / k o p + a j + m / / historical form, kopiet "dig" lp.sing. / / k o p + a + V m / / modern form. Of primary importance for our purposes is the observation noted in Smiech (1974) that the parallel change of the inflectional endings in question occurred simultaneously in the related Derived Imperfective forms i.e. wykopywam "dig out" lp.sing.: historical form, wykopujq "dig out" lp.sing.: modern form. Notice that the redistribution of the exponents of the 1 person singular non-preterite in the case at hand constitutes a very strong argument in favour of the assumption that the formative //iν// (the exponent of the Derived Imperfective) is inserted in front of the verbalizing suffix. The vowelless allomorph of the 1 person singular non-preterite is sensitive exclusively to the presence of the verbalizing suffix / / a j / / , so once a given root (e.g. kop-) which, as shown in Smiech (1974) was paralleled in this respect by many others, began to couple with the verbalizing suffix //a// instead of / / a j / / , / / m / / was replaced by //Vm// in both kopac and wykopywac (i.e. koparn and wykopywam changed into kopi§ and wykopuj% simultaneously). Second, in modern Polish there are still verbal stems whose Derived Imperfective forms select / / m / / rather than //Vm// as the exponent of the 1 person singular non-preterite, which indicates the presence of the formative / / a j / / in the underlying structure of the Derived Imperfective forms. Consider the examples in (12). (12)

dokonywac "achieve" inf.: dokonywam lp.sing. pokonywac "overcome" inf.: pokonywam lp.sing. przekonywac "persuade" inf.: przekonywam lp.sing. wykonywac "carry out" inf.: wykonywam lp.sing. dorownywac "match" inf.: dorownywam lp.sing. porownywac "compare" inf.: porownywam lp.sing. wyrownywac "compensate" inf.: wyrownywam. lp.sing. wyjednywac "wheedle" inf.: wyjednywam lp.sing. zjednywad "win over" inf.: zjednywam lp.sing. rozpami$tywa6 "ponder" inf.: rozpamiqtywam lp.sing. As pointed out in Butler et al. (1986:273) in all the cases in (12) the ending / / m / / (e.g. dokonywam, porownywam) is still acceptable albeit rare. Notice that all the verbal stems here contain the verbalizing suffix / / a j / / both in the morphologically simple imperfective and the prefixed perfective forms. Now, forms such as dokonywam, zjednywam, etc. are best accounted for on the assumption that the //iv// is inserted between the verbalizing

91 suffix / / a j / / and the root and it is the former formative that is responsible for selecting the vowelless exponent of the 1 person singular non-preterite in all the three forms derived from the same root including Derived Imperfective e.g. rownam / / r u v n + a j + m / / "equate" lp.sing. Imperf., wyrownam / / v i + r u v n + a j + m / / "equate" lp.sing. Perf., wyrownywam //vi+ruvn+iv+aj+m// "equate" lp.sing. DI. The other property of the verbalizing suffix / / a j / / which sets it apart from all the other verbalizing suffixes is the fact that / / a j / / does not select either of the two stem-extensions ( / i / or / e / ) in the non-preterite paradigm. Note that the verbal stems which show up with / / m / / in the 1 person singular non-preterite do not exhibit the non preterite stem-extension throughout the non-preterite paradigm e.g.: zjednywamy "win over" lp.pl, zjednywacie "win over" 2p.pl., zjednywasz "win over" 2p.sing., zjednywa "win over" 3p.sing. Two groups of verbal stems seem to argue against the assumption that the formative / / i v / / as the exponent of the Derived Imperfective is inserted between the root and the verbalizing suffix. First of all, the overwhelming majority of aj-verbs behave unlike dokonywac "achieve" and other stems listed in (12) inasmuch as they opt exclusively for the / / V m / / rather than the / / m / / allomorph of the 1 person singular non-preterite and require a non-preterite stem-extension in the relevant inflected forms of the Derived Imperfective derivatives. Consider wydmuchiwac "blow out" as a typical non - preterite paradigm in this class of verbal stems (13) vi+dmux+uj+ew vi+dmux+uj+e+mi vi+dmux+uj+e+s vi+dmux+uj+e+tpe vi+dmux+uj+e vi+dmux+uj+öw Here the verbal stems with the Derived Imperfective formative / / i v / / behave in a parallel manner to stems with the verbalizing suffix (a) preceded by the formative (ow) (i.e. owa-verbs) - the formative (uj) appears throughout the non-preterite paradigm to be followed by the non-preterite stem-extension formative (e) in all forms with the usual exception of the 1 person singular and the 3 person plural. Thus, it seems reasonable to assume that the morphological structure of the surface sequence (ywa) ((iwa)) parallels that of (owa) i.e. is as follows: / / i v + a / / . For that end we must postulate a readjustment rule by virtue of which the verbalizing suffix / / a j / / is changed into / / a / / after the application of the rule of Derived Imperfective formation which inserts / / i v / / . (14) The rule (14) is lexically governed in the sense that all verbal stems listed in (12) are marked as optional exceptions to it. 6 6 Interestingly enough, a formally parallel change occurred in all verbal stems with (owa). As pointed out in Klemensiewicz et al. (1981) ALL Derived Imperfective forms of such verbs (e.g. przemalowywac "paint over" could until recently take the vowelless exponent of the 1 person singular non-preterite and show up without the non-preterite stem-extension in the remaining forms, which shows that the morphological structure of (owa) was / / o v + a j / / rather than / / o v + a / / .

92 Next, consider the verbal stems with the verbalizing suffix //e// exemplified in (15). IMPERFECTIVE krzyczec "shout" bzyczec "buzz" beczec "whine"

PERFECTIVE wykrzyczec "shout out" pobzyczec "buzz" rozbeczec "start whining"

DERIVED IMPERFECTIVE wykrzykiwac "shout out" pobzykiwac "buzz" rozbekiwac "start whining"

In sharp contrast to morphologically simple imperfective stems and the prefixed perfective forms the Derived Imperfective derivatives systematically fail to show traces of the verbalizing suffix //e//. Apart from the obvious observation that (e) is not to be seen on the surface in the infinitival forms such as wykrzykiwac "shout out" two points are worthy of note in this respect. First, notice that the root-final consonants are systematically subject to rule-governed morphophonological changes in both simple and prefixed perfective stems but NOT in the Derived Imperfective derivatives. The phonological rules responsible for the consonantal changes in question are triggered by the verbalizing suffix //e//. Second, the non-preterite stem-extension formative which systematically shows up throughout the non-preterite paradigm of the relevant verbal stems is (e) rather than (i). Recall that, as shown in Rubach (1984), the phonological value of this extension formative is crucially conditioned by the phonological shape of the verbalizing suffix. The (i) extension appears only after the verbalizing suffixes realised by front vowels i.e. after //e// and //i//. Apparently the verbalizing suffix which conditions the appearance of the non-preterite extension formative throughout the non-preterite paradigm of verbal stems such as wykrzykiwac "cry out" is NOT //e// e.g. wy+krzyk+uj+e+sz 2p.sing wy+krzyk+uj+e+my lp.pl.

wy+krzyk+uj+e 3p.sing. wy+krzyk+uj+e+cie 2p.pl.

The assumption that the exponent of the Derived Imperfective is inserted between the root and the verbalizing suffix //e// must be coupled with the postulation of a rule of readjustment replacing the verbalizing suffix //e// with //a//. Two further processes take place in the non - preterite inflectional paradigm: - //iv// is replaced by //uj// (discussed later) - the verbalizing suffix //a// triggers off the appearance of the (e) variant of the non-preterite stem extension formative (if the verbalizing suffix //e// were not converted to //a// this variant here would incorrectly be (i) rather than (e)). The readjustment rule replacing the .verbalizing suffix //e// with //a// is given in (16) (16)

/ / e / / —• // a// j

//iv//—

coupled with (14) rule (16) gives (17)

(17)

{ //V/ } ^

//a// //iv//

/



(17) is a typical allomorphy rule in the sense of Aronoff (1976) by virtue of which a formative changes its phonological shape in a morphologically defined context.

93 Incidentally, note that both exponents of the Derived Imperfective, although they are attached via stem-based morphological processes trigger readjustment rules defying AronofF and Sridhar (1987) claim. All in all, the formative //iv// appears as an exponent of the Derived Imperfective in four groups of verbal stems: a-verbs, owa-verbs, aj-verbs, e-verbs. //iv// is inserted right in front of the verbalizing suffix. In aj-verbs and e-verbs the Derived Imperfective formation is followed by the application of a readjustment rule whereby verbalizing suffixes are changed into the verbalizing suffix //a//. The presence of the formative //a// rather than //aj// or //e// is necessary to ensure the correct application of morphological processes in the non-preterite paradigm of iwa-verbs. Thus it seems that the two exponents of the category Derived Imperfective in modern Polish are accounted for in terms of stem-based morphological processes, which, contrary to the claim advanced in AronofF and Sridhar (1987) are crucially sensitive to the morphological structure of the base. First of all, the two allomorphs of Derived Imperfective are in complementary distribution and their appearance in a given stem is crucially conditioned by the phonological value of a given verbalizing suffix. Compare the table (18). The exponent of Derived Imperfective aj iv

correlates with verbalising suffixes o,i aj, e, (ov)a

Next, while allomorph //aj// is directly attached to the verbal stem (i.e. it follows the verbalizing suffix) //iv// is inserted immediately in front of the verbalizing suffix. The other factor which crucially conditions the application of both rules introducing the exponents of the Derived Imperfective i.e. the presence of the feature "perfective" in the host has a bearing on the other familiar constraints on morphological processes: the Atom Condition and the Adjacency Condition. Recall that their common assumption is that a morphological rule may be sensitive to no more than one formative in the base. Thus, with regard to the rules of the Derived Imperfective formation it implies that both factors to which they are sensitive are associated with one formative. Notice that this prediction is borne out by merely a handful of i-stems such as strzelic "shoot" rzucic "throw" skoczyc "jump". In such cases the value "perfective" of the category A S P E C T is determined by the verbalizing suffix. The Derived Imperfective form is derived via the attachment of the formative //aj// selected because of the phonological value of the verbalizing suffix. Thus, the formation of the Derived Imperfective stem is conditioned by only one base formative. For all the other verbal stems in modern Polish, however, the morphological rules responsible for the formation of the Derived Imperfectives must crucially refer to information introduced by two formatives: the verbalizing suffix, whose phonological shape determines which of the two exponents of the Derived Imperfective is applicable for a given verbal stem and the prefix, which determines the value of the category ASPECT for the whole verbal stem.

94 Conclusion: stem-based processes axe sensitive to the morphological structure of the host. A closer scrutiny of some productive morphological processes of deverbal suffixation in modern Polish has revealed that Aronoff and Sridhax's (1987) claim to the effect that the morphological structure of the base is non-transparent to the operation of stem-based morphological rules would stand in the way of expressing some apparently valid generalizations. It seems evident that to the extent that the morphological rules under consideration are sensitive to morphological make-up of their bases they disconfirm the Aronoff and Sridhax's constraint. Likewise, neither of the well-known attempts to pinpoint those aspects of the morphology of the base to which a morphological operation may be sensitive (with the concomitant exclusion of all the others) seems to fare any better in view of the empirical data considered here. The restriction of the sensitivity of rules to one composite base formative only - the common core of the Adjacency condition and the Atom Condition seems to be empirically too strong inasmuch as the attachment of Polish suffixes requires information associated with several (sometimes all) base formatives. The morphological rules considered in this chapter lead to rather disappointing conclusions: morphological rules are notoriously conditioned by various aspects of the morphological structure of their bases in what seems to be an unprincipled manner.

Concluding remarks

Although the main thrust of the present work has been the criticism of constraints on morphological rules proposed by others I have also ventured some modest proposals of my own. Chief among those was the claim that morphological rules of suffixation may be restricted to apply to morphologically simple units devoid of meaning and grammatical properties labelled as ROOT or, alternatively, to members of a syntactic category i.e. STEM. In Chapter 3 we were at pains to demonstrate that root-based and stem-based morphological processes have different characteristics. The validity of this distinction was demonstrated with a number of Polish verb-forming rules. The body of the book has been devoted to a number of constraints on stem-based morphological rules of suffixation in both Polish and English. The constraints inherent in major tenets of the theory of Lexical Phonology have been examined in Chapter 2 against the background of the vast majority of English processes of suffixation. First, it has been shown how the distribution of English inflectional allomorphy follows from overall language-specific constraints on syllable structure. Next, the constraints on suffixation inherent in the major hypotheses of Lexical Phonology BRACKET ERASURE CONVENTION and AFFIX ORDERING GENERALIZATION have been shown to be empirically inadequate. The Saxon Genitive formative, which, of necessity attaches outside the Lexicon crucially must have access to the morphological make-up of his Lexicon-generated hosts although the BRACKET ERASURE CONVENTION would have otherwise. The AFFIX ORDERING GENERALIZATION stipulates that phonological and distributional properties of English derivational suffixes go hand in hand. The stress-shifting suffixes are free to attach to stems unless the latter contain a stress-neutral affix. The stress-neutral affixes axe allegedly fully blind to the morphological make-up of their hosts. Yet, it turns out that these predictions are clearly false. Only a handful (i.e. 3) of stress-neutral suffixes behave as the AFFIX ORDERING GENERALIZATION would have it. Over a dozen (i.e. 14) English derivational suffixes, irrespective of their phonology, are clearly sensitive to the morphological structure of their hosts in that they attach to some suffixed stems but not to others. The bulk of English suffixes, regardless of their phonological properties fail to attach to an already suffixed stem. As the latter two constraints are also relevant to the vast majority of stress-neutral suffixes they clearly falsify the BRACKET ERASURE CONVENTION. Further constraints on the operation of stem-based morphological processes advanced by leading generative morphologists have been checked against a sample of relevant Polish data in Chapter 4.

96 Both the strong constraint proposed by Aronoff and Sridhar (1987) whereby all stems remain opaque to all suffixes that attach to them as well as the more relaxed Atom Condition and Adjacency Condition have been shown to be inadequate to the extent that the productive deverbal morphological processes in Polish systematically fail to conform to them. (The common assumption of Atom Condition and Adjacency Condition is that suffixes are allowed access to no more than one component formative in the base) The other important constraint advanced by Aronoff and Sridhar (1987) - that no stem-based suffix induces readjustment rules in the host has also been disqualified as the Polish processes of deverbal suffixation notoriously trigger off both allomorphy and truncation rules.

References

Aitchison, Jean (1981): Language Change: Progress or Decay. - London: Fontana. Anderson, Stephen (1974): The Organization of Phonology. - New York: Academic Press. - (1982): "Where is Morphology". - In: Linguistic Inquiry 13, 571-612. - (1985): "Inflectional Morphology". - In: T.Shopen(ed.): Language Typology and Syntactic Description. Vol. 3: Grammatical Categories and the Lexicon (Cambridge: CUP) 150-201. - (1987): "Morphological Theory". - In: Nemeyer (ed.): Linguistics. The Cambridge Survey. Vol. 1 (Cambridge: CUP) 146-191. Allen, Margaret (1978): Morphological Investigations. - PhD Thesis, University of Conneticut. Aronoff, Mark (1976): Word Formation in Generative Grammar. - Cambridge, Ma: MIT Press. - (1988): Review of: On the Definition of Word, by A. Di Scullio, E. Williams (Cambridge, 1987). Language 64, 766-770. Aronoff, Mark and S.N. Sridhar (1987): "Morphological Levels in English and Kannada". - In: E.Gussmann (ed.): Rules and the Lexicon. (Lublin: WKUL) 9-22. Bartnicka, Danuta (1970): Adjektywizacja imieslowöw w j^zyku polskim. - Warszawa: PWN. Bauer, Laurie (1983): English Word Formation. - Cambridge: CUP. - (1988): Introducing Linguistic Morphology. - Edinburgh: EUP. Bloomfield, Leonard (1933): Language. - New York: Holt, Rinehart, Winston. Booij, Geert (1977): Dutch Morphology. - Lisse: De Ridder. - (1986): "Form and Meaning in Morphology: the Case of the Dutch Agent Nouns". - In: Linguistics 24, 503-517. Booij, Geert and Jerzy Rubach (1987): "Postcyclic versus Postlexical Rules in Lexical Phonology". - In: Linguistic Inquiry 18, 1-44. Borowsky, Toni (1989): "Structure Preservation and the Syllable Coda in English". - In: Natural Language and Linguistic Theory 7, 145-166. Bybee, Joan (1985): Morphology. - Amsterdam: Benjamins. Butler, Danuta (ed.)(1986): Kultura j^zyka polskiego. - Warszawa: PWN. Carstairs, Andrew (1987): Allomorphy in Inflection. - London: Croom Helm. Chomsky, Noam (1957): Syntactic Structures. The Hague: Mouton. - (1965): Aspects of the Theory of Syntax. - Cambridge, Ma: MIT Press. (1970): "Remarks on Nominalization". - In: R. Jacobs, P. Rosenbaum (eds.): Readings in English Transformational Grammar. (Waltham, Ma: Ginn. Chomsky, Noam and Morris Halle (1968): The Sound Pattern of English. - New York: Harper and Row. Clark, Eve and Henry Clark (1979): "When Nouns Surface as Verbs". - In: Language 55, 767-811. Clements, George and Samuel Keyser (1983): CV Phomology: A Generative Theory of the Syllable. Cambridge, Ma: MIT Press. Comrie, Bernard (1976): Aspect. - Cambridge: CUP. Di Sculio, Anna and Edwin Williams (1987): On the Definition of the Word. - Cambridge, Ma: MIT Press. Doke, Charles (1935): Bantu Linguistic Terminology. - London: Longman. Doroszewski, Witold (ed.) (1958-1969): Slownik wsptSlczesnego j^zyka polskiego. - Warszawa: PWN. Dressier, Wolfgang (1977): "Phono-morphological dissimilation". In: W. Dressler (ed.): Phonologica 76 (Innsbruck), 41-48. - (1986): "Explanation in Natural Morphology, Illustrated with Comparative and Agent Noun Formation". - In: Linguistics 24, 519-548. - (1988): Inflection and Derivation. - ms.

98 Fabb, Nigel (1988): "English Suffixation is Constrained Only by Selectional Restrictions". - In: Natural Language and Linguistic Theory 6, 527-539. Pudge, Eric (1984): English Word Stress. - London: Allen and Unwinn. Gawronski, Andrzej (1930): Podr^cznik sanskrytu. - Krakow: PAU. Gold, Dwight (1975): "Rabbit-Foot vs. Rabbit's Foot - a Trend in English Compounds". - In: American Speech 50, 149-155. Gorska, Elzbieta (1982): "Formal and Functional Restrictions on the Productivity of Word Formation Rules". - In: Folia Linguistica XVI/1-4, 149-162. Grzegorczykowa, Renata and Jadwiga Puzynina (1981): Slowtwörstwo wspolczesnego j^zyka polskiego. Warszawa: PWN. Gussman, Edmund (1978): Contrastive Polish-Engish Consonantal Phonology. - Warszawa: PWN. - (1980): Studies in Abstract Phonology. - Cambridge, Ma: MIT Press. - (1988): "The Lexicon of English De-adjectival Verbs". In: E.Gussman (ed.): Rules and the Lexicon. (Lublin: WKUL) 79-103. - (1988): Review of: The Theory of Lexical Phonology, by K.Mohanan (Dodrecht, 1986). - Journal of Linguistics 24, 232-239. Halle, Morris (1962): "Phonology in Generative Grammar". - In: Word 18, 54-72. - (1973): "Prolegomena to a Theory of Word Formation". - In: Linguistic Inquiry 4, 3-16. - (1978): Formal and Functional Considerations in Phonology. - ms. - (1987): On the Phonology-Morphology Interface. - ms. Halle, Morris and K. Mohanan (1985): "Segmental Phonology of Modern English". - In: Linguistic Inquiry 16, 57-116. Harms, Robert (1972): Some Non-rules of English. - ms. Hoard, James and Charles Sloat (1970): "The Inflectional Morphology of English". - In: Glossa 5, 47-56. Hockett, Charles (1958): A Course in Modern Linguistics. - New York: Mcmillan. - (1987): Refurbishing Our Foundation. - Amsterdam: John Benjamins. Hyman, Larry (1978): "Word Demarcation". - In: J. Greenberg (ed.): Universals of Human Language. Vol.1: Phonology, 443-480. Jackendoff, Roy (1972): Semantic Interpretation in Generative Grammar. - Cambridge, Ma: MIT. - (1975): "Morphological and Semantic Regularities in the Lexicon". - In: Language 51, 639-671. Jakobson, Roman (1948): "Russian conjugation". - In: Word 4, 155-167. Jensen, John (1990): Morphology. Word Structure in Generative Grammar. - Amsterdam: John Benjamins. Jensen, John and Mary Stong-Jensen : "Morphology is in the Lexicon!". - In: Linguistic Inquiry 15, 474-478. Jespersen, Otto (1909-1949): Modern English Grammar on Historical Principles. - London: Allen and Unwinn. Joos, Martin (1957): Readings in Linguistics. - Chicago: UCP. Karlsson, Fred (1983): Finnish Grammar. - Parvoo: WSO. Kastovsky, Dieter (1980): "Zero in Morphology. A Means for Making up for Phonological Losses". - In: J. Fisiak (ed.): Historical Morphology.(The Hague: Mouton) 213-250. Kiparsky, Paul (1973): "How Abstract is Phonology?". - In: O. Fujimura (ed.): Three Dimensions of Linguistic Theory. (Tokyo: TEC Company) 57-86. - (1982a): "From Cyclic to Lexical Phonology". - In: H. van der Hulst, N. Smith (eds.): The Structure of Phonological Representation. Part 1 (Dodrecht: Foris) 131-176. - (1982b): "Word Formation and the Lexicon". - In: F. Ingemann (ed.): Proceedings of the 1982 Mid-America Linguistic Conference. (Lawrence: University of Kansas). - (1982c): "Lexical Morphology and Phonology". - In: Linguistics in the Morning Calm. (Seoul: Hanshin) 3-91. Kisseberth, Charles (1970): "On the Functional Unity of Phonological Rules". - In: Linguistic Inquiry 1, 291-306.

99 Klemensiewicz, Zenon et al. (1981): Gramatyka historyczna j^zyka polskiego. - Warszawa: PWN. Klimek, Zbigniew (1983): Historia odmiany czasowniköw typu kopa6, kopam II kopi§. - Krakow: WUJ. Kramsky, Jan (1969): The Word as a Linguistic Unit. - The Hague: Mouton. Lang, Martin (1990): Spanish Word-formation. - London: Routledge. Laskowski, Roman (1975): Studia nad morfologiq, wspdlczesnego jQzyka polskiego. - Wroclaw: Ossolineum. Laskowski, Roman et al. (1984): Gramatyka wspölczesnego j^zyka polskiego. Morfologia. - Warszawa: PWN. Lass, Roger (1974): "Linguistic Orthogenesis? Scots Vowel Quantity and the English Length Conspiracy". - In: J. Anderson, Ch. Jones (eds.): Historical Linguistics. Vol. 2 (Amsterdam: North Holand Publishing Company) 311-353. Linde, Boguslaw (1807-1814): Slownik j^zyka polskiego. - Warszawa. Lieber, Rochelle (1981): On the Organization of the Lexicon. Ph.D. Dissertation. MIT. - (1982): "Allomorphy". - In: Linguistic Analysis 10, 27-51. Lyons, John (1977): Semantics. 2 vols. - Cambridge: CUP. - (1981): Language and Linguistics. - Cambridge: CUP. Magee, Bryan (1990): Popper. - London: Clarendon. Marchand, Hans (1960): The Categories and Types of Present Day English Word Formation. - München: C.K. Beck. Mascaro, Juan (1976): Catalan Phonology and the Phonological Cycle. Ph.D. Dissertation. MIT. Milewski, Tadeusz (1973): Introduction to the Study of Language. - The Hague: Mouton. Miner, Paul (1975): "English Inflectional Endings and Unordered Rules". - In: Foundations of Language 12, 339-365. Mohanan, K. (1986): The Theory of Lexical Phonology. - Dodrecht: Reidel. Mortimer, Halina (1982): Logika indukcji. - Warszawa: PWN. Myers, Samuel (1985): "The Long and the Short of It". - In: Papers from the 21 Regional Meeting of the Chicago Linguistic Society. Nowotny-Szybist, Magdalena (1973): Neologizmy slowotwörcze w tworczosci Stanislawa Ignacego Witkiewicza. - Wroclaw: Ossolineum. Pike, Kenneth (1967): Language in Relation to a Unified Theory of the Structure of Human Behaviour. - The Hague: Mouton. Peplowski, Franciszek (1974): Nazwy wykonawcöw czynnosii w polszczyznie XVI w. - Wroclaw: Ossolineum. Pesetsky, David (1979): Russian Morphology and Lexical Theory, ms. Plank, Karl (1936): English Genitive Case. ms. Popper, Karl (1936): Logik der Forschung. - Vienna: Springer Verlag. Reszkiewicz, Alfred (1977): Correct your English Pronunciation. - Warszawa: PWN. Rice, Paul (1985): "On the Placement of Inflection". - In: Linguistic Inquiry 16, 155-161. Rubach, Jerzy (1977): Consonant Changes in English and Polish. - Wroclaw: Ossolineum. - (1982): Analysis of Phonological Structures. - Warszawa: PWN. - (1984): Cyclic and Lexical Phonology. The Structure of Polish. - Dordrecht: Foris. - (1985): "Segmental Rules of English and Cyclic Phonology". - In: Language 60, 21-54. Ruszkiewicz, Piotr (1985): "Constraining a Rule of Truncation in English". - In: Studia Gramatyczne VII, 179-215. - (1990): "Constraining the Rule of Lower in Polish". - In: Folia Linguistica XXIII/3- 4, 317-326. Scalise, Sergio (1984): Generative Morphology. - Dordrecht: Foris. - (1988): "The Notion of Head in Morphology". - In: Morphology Yearbook 1, 229-244. Schultink, Henk (1988): "Morphological Heads: Evidence from Swahili". - In: Morphology Yearbook 1, 247-258. Selkirk, Elisabeth (1982): The Syntax of Words. - Cambridge, Ma: MIT Press. Shibatani, Mayoshi (1972): "The Phonological Representations of English Inflectional Endings". - In: Glossa 6, 117-127.

100 Siegel, Dorothy (1974/1979): Topics in English morphology. Ph.D. Dissertation, MIT. - New York: Garland. - (1977): The Adjacency Condition in Morphology, ms. Smolkowa, Maria and Teresa Tekiel (1983): Nowe slownictwo polskie. - Wroclaw: Ossolineum. Spencer, Andrew (1991): Morphological Theory. - Cambridge: CUP. Sproat, Richard (1985): On deriving the Lexicon. Ph.D. Dissertation, MIT. Stemberger, John (1981): "Morphological Haplology". - In: Language 57, 791-817. Strauss, Stephen (1979): "Against Boundary Distinctions in English Morphology". - In: Linguistic Analysis 5, 387-419. - (1982): Lexicalist Phonology of English and German. - Dordrecht: Foris. Szpyra, Jolanta (1987): "Inputs to WFR - Phonological, Intermediate or Phonetic. The Case of Verbs and Deverbal Nouns in Polish". - In: E. Gussman (ed.): Rules and the Lexicon. (Lublin: WKUL) 169-205. Smiech, Witold (1967): Rozwöj czasu terazniejszego czasownika w j§zyku polskim. - Lödz. - (1974): "Uwagi ο fleksji czasownikow koniugacji na -ac w j^zyku polskim". - In: Rozprawy komisji jezykowej LTN t.XX, 167-172. Taylor, John (1989): Linguistic Categorization. - Oxford: Clarendon Press. Tokarski, Jan (1951): Czasowniki polskie. - Warszawa: PWN. - (1978): Fleksja polska. - Warszawa: PWN. Trommelen, M. and W. Zonneveld (1986): "Dutch Morphology: Evidence for the Right Hand Rule". - In: Linguistic Inquiry 17, 147-169. Urbaiiczyk, Stanislaw (1953-1989): Slownik staropolski. - Warszawa: PWN. Vachek, Jan (1976): Selected Writings in English and General Linguistics. - The Hague: Mouton. Williams, Edwin (1981): "On the Notions 'Lexically Related' and 'Head of a Word'". - In: Linguistic Inquiry 12, 145-274. Wells, John (1982): Accents of English. 3 Vols. - Cambridge: CUP. Zwicky, Arnold (1975): "Settling on an Underlying Form: the English Inflectional Endings". - In: D. Cohen, J. Wirth (eds.): Testing Linguistic Hypotheses. (Washington: Hemsphere) 129-185. - (1986): "Supressing the Zs". - In: Ohio State University Working PapersLinguistics 32, 149-156. Zirmunskij, Victor (1966): "The Word and its Boundaries". - In: Lingusitics 27, 65-91.