Beyond Words: Content, Context, and Inference 9781614512776, 9781614513865

Citation preview

Beyond Words

Mouton Series in Pragmatics 15

Editor Istvan Kecskes

Editorial Board Reinhard Blutner Universiteit van Amsterdam The Netherlands N. J. Enfield Max-Planck-Institute for Psycholinguistics Nijmegen The Netherlands

Ferenc Kiefer Hungarian Academy of Sciences Budapest Hungary Lluı´s Payrato´ University of Barcelona Spain

Raymond W. Gibbs University of California Santa Cruz USA

Franc¸ois Recanati Institut Jean-Nicod Paris France

Laurence R. Horn Yale University USA

John Searle University of California Berkeley USA

Boaz Keysar University of Chicago USA

De Gruyter Mouton

Deirdre Wilson University College London Great Britain

Beyond Words Content, Context, and Inference

edited by

Frank Liedtke Cornelia Schulze

De Gruyter Mouton

ISBN 978-1-61451-386-5 e-ISBN 978-1-61451-277-6 ISSN 1864-6409 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at http://dnb.dnb.de. ” 2013 Walter de Gruyter GmbH, Berlin/Boston Printing: Hubert & Co. GmbH & Co. KG, Göttingen ⬁ Printed on acid-free paper Printed in Germany www.degruyter.com

Contents Introduction: Beyond Words Frank Liedtke and Cornelia Schulze . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Section I.

General concepts

Short introduction (Frank Liedtke) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 1.

Communication in the narrower and broader sense. A reconstruction in terms of Sign Theory Rudi Keller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.

Pragmatics in Optimality Theory Reinhard Blutner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

Section II. Acquiring inferential abilities Short introduction (Cornelia Schulze) . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 3.

Word learning by exclusion - pragmatics, logic and processing Susanne Grassmann . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

4.

Children’s knowledge of scales in the acquisition of almost Patricia Amaral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

5.

Relevance inferences in young children: 3-year-olds’ understand a speaker’s indirectly expressed social intention Cornelia Schulze . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

6.

Early pragmatics with words Nausicaa Pouscoulous . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

VI

Contents

Section III. Grammar, meaning, and enrichment Short introduction (Frank Liedtke) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147 7.

Procedures and prosody: Weak encoding and weak communication Billy Clark . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151

8.

Pragmatic templates and free enrichment Frank Liedtke . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183

9.

Pragmatic enrichment in adjectival passives: The case of the post state reading Helga Gese . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207

10.

Pragmatic inferencing and expert knowledge Petra B. Schumacher and Jörg Meibauer . . . . . . . . . . . . . . . . . . 231

Section IV. Constraints, memes, and constructions Short introduction (Frank Liedtke) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249 11.

Empirical and theoretical evidence for a model of quantifier production Chris Cummins and Napoleon Katsos . . . . . . . . . . . . . . . . . . . . . 253

12.

Construction as memes – Interactional function as cultural convention beyond the words Elke Diedrichsen . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283

13.

A pragmatic Pandora’s box: Regularities and defaults in pragmatics Marco Mazzone . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307

Contributors to the volume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335

Introduction: Beyond Words In pragmatics, it is widely accepted that the overall meaning of an utterance performed as part of a verbal interchange (in the sense of what is said) is basically underdetermined by the meaning of the sentence uttered. The bare sentence meaning yields only a propositional schema, which does not even suffice for the assignment of a truth-value. Sentences containing indexicals (I am here) and other context-sensitive expressions are the most salient examples for this sort of underdetermination. Truth conditions or fulfillment conditions in a broader sense can only be upset if contextual information is added. Beyond these contextual ingredients, which are triggered by (indexical) expressions in the sentence itself, further elements of utterance meaning have been assumed, which do not correspond to any element which is articulated in the sentence, but which are added in a top down process as free enrichments. These enrichments encompass for instance temporal relationships between conjuncts (and ! and then), strengthening of conditionals (if ! only if), restrictions of quantifier domains (all ! all of the guests) as well as the narrowing of predicative constructions (The flower is red ! the blossom is red). Assuming purely pragmatic enrichment strategies means that pragmatic interpretation may come, and in many cases does come prior to the semantic interpretation of an utterance in terms of truth/fulfillment conditions. Thus, if one wants to explain how people (as addressees) achieve a full understanding of a performed utterance, one has to presume manifold enrichment strategies on the part of the language users. The information needed for the ascription of a full proposition as well as of a definite illocution of an utterance by an audience must rely on aspects beyond words, resulting from pragmatic inferences which contain the words uttered as one of their premises. The sentence producers on the other hand are supposed to anticipate these interpretive strategies, so that it is assumed that they will construe their utterance in a parsimonious way accounting for enrichment activities as part of utterance interpretation. The traditional picture of utterance meaning was that pragmatic aspects do not come into play on the level of what-is-said. It was only in connection with conversational implicatures that inferences based on conversational maxims had been presumed. This picture has changed, so that for many authors pragmatic processes are accepted not only as part of conversational implicatures, but as part of what speakers say with their utterances

2 Frank Liedtke and Cornelia Schulze

or as part of some intermediate level of utterance meaning. If this idea is correct – and there is much evidence that it is – then what counts as having been said for most contemporary authors goes far beyond sentence meaning. Rather, it has to be considered as a complex utterance level combining semantic knowledge and context-driven, pragmatic information as an integrated whole. Thus, the semantics/pragmatics distinction is not coextensive with the distinction between what-is-said and what-is-implicated by an utterance, rather it is either situated within the realm of what-is-said or part of an utterance level, neither consisting of what-is-said nor of what is implicated, rather being localized in between as a third, intermediate level. One of the first authors who assumed that contextual information plays a decisive role in the interpretation of literal meaning was J.R. Searle. He held that background knowledge, i.e. knowledge about our physical and cultural environment, is constitutive for interpreting even such harmless seeming sentences like “The cat is on the mat” (Searle 1978). Without such basic assumptions as the existence of gravity we are not able to decide whether such sentences are true or false. Charles Travis also denied that there is a semantics of words of a language which is independent of the occasion-meaning of their utterance. Any semantics which words might contain is one that they contain only in an occasion-sensitive manner. (Travis 1981). From this starting point, different approaches have been developed in order to account for the pragmatic character of what-is-said as a level between sentence meaning on the one side and conversational implicatures on the other. One of the first approaches was Relevance Theory (Sperber and Wilson; Blakemore; Carston) whose proponents have coined the term explicature in order to characterize the inferential aspect of utterance meaning. What is explicit in their account is the semantic information within the sentence and the results of the inferential procedures leading towards the development of the logical form. Thus, they see “the explicit side of communication as richer, more inferential, and hence more worthy of pragmatic investigation than do most pragmatists in the Gricean tradition“ (Sperber and Wilson 1986: 183). Bach, as a non-relevance theoretic author, assigned the role of pragmatic enrichment to the intermediate level of implicitures, which he delineated against what-is-said on the one hand and implicatures on the other. His notion of what-is-said or the explicit is rather restricted in a Gricean manner in that it contains only the sentence meaning, whereas pragmatic enrichments belong to the realm of implicitures, which are by nature non-explicit (cf. Bach 1994).

Introduction: Beyond Words

3

Récanati (2004, 2010) is more radical in his conception insofar as he subsumes under what-is-said not only the sentence meaning and saturations which are required in order to achieve a full proposition, but also free enrichments. This leads towards a very wide – or as Bach says liberal – conception, which may even acknowledge metaphorical speaking as a case of what-is-said. Concerning the semantic side of the whole utterance, he proposes the Availability Principle, stating that we have to preserve our pre-theoretic intuitions, if we are constructing the different levels of utterance meaning (Récanati 2004: 20). This prevents counterintuitive semantic readings of sentences from being part of the analysis (All were leaving – All people in the world were leaving). The so-called neo-Gricean approach, on the other side, has enhanced the Gricean concept of conversational implicatures and its different subtypes, namely particularized and generalized implicatures. The latter have been recognized to contribute essentially to the propositional content of the utterance. Gazdar, Horn and Levinson are the protagonists of this pragmatic line of theory. Levinson has mainly developed the paradigm in postulating three fundamental principles, which replace the number of maxims in Grices system. These principles function as heuristics, which allow addressees to ascribe presumptive meanings to the utterances perceived. Different enrichment strategies are assumed as being based on generalized conversational implicatures, i.e. inferences, which take place by default and which are cancelled only if the presumptive interpretation turns out to be false (cf. Levinson 2000). Thus, Levinsons’ theory is based on utterance types and generalized inferences, in contrast to Relevance Theory, which emphasizes the ad hoc character of pragmatic inferencing (cf. Carston 2002). Irrespective of the theory which is being pursued concerning enrichments, the context-based nature of literal utterance meaning is generally accepted among non-minimalist authors. From a minimalist point of view, however, enrichment strategies are ruled out of the determination of an utterance’s truth conditions, processes of inferential enrichment of an utterance being relegated to the level of the performed speech act. (cf. Cappelen and LePore 2005, 2006 and the resulting controversy in Mind and Language 21 (1), 2006). This book is focused on the non-minimalist accounts of utterance meaning, but even within this paradigm – which some may call a contextualist one – central questions about the nature, the function and the acquisition of pragmatic inferencing strategies remain open. Mainly the question of the relation between the explicit and the implicit side of verbal communication and its mutual delimitation has not been

4 Frank Liedtke and Cornelia Schulze

answered in a final way. Do we have to assume an intermediate level of pragmatic inferencing, leaving what-is-said to the semantic side of the utterance, or do we have to conceive the notion of what-is-said as a pragmatic one encompassing context-based inferences? What is the character of these pragmatic inferences, wherever they may be situated in a descriptive model? Are they nonce inferences arising anew in each act of communication, or do we have to conceive of them as based on regularities and conventions? Even if these central questions have been answered at a provisional basis, another question remains open, namely that of the acquisition of the skills which are relevant for mastering the inferential processes leading to an adequate interpretation of utterances. Apparently, these skills, despite their complexity, are acquired at a very early stage of the development of the communicative competence. The inferential character of pragmatic enrichment is, however, also under debate. Another model which is potentially rival to the inferential picture is one which assumes contextual knowledge being associated with the utterance perceived. Récanati makes this assumption explicitly in his writings (cf. 2004; 2010), and it is the basis of Marco Mazzone’s paper in this book, who assumes that associative processes based on the co-occurrence of stimuli are responsible for the storage of linguistic and non-linguistic information. Defaults and generalized conversational implicatures are seen as cases of the storage of information in the presence of regular patterns of experience. Within the associative model, the paradigm of Optimality Theory (OT) is adopted by some pragmatically oriented linguists as a strategy for dealing with implicatures of different kinds. One component of OT is a list of tendencies which hold for properties of a language, and these tendencies have the form of constraints which may be violated. It is in principle possible that specific Gricean maxims may be reformulated as constraints for optimal form-meaning pairings in terms of OT. This strategy is chosen by Blutner in this volume. Another approach which aims at reconciling OT with Gricean pragmatics is also undertaken by Cummins and Katsos (see their contribution in section IV). The present book structures the contributions into four areas. In the first section, general questions concerning human communication, be it of a linguistic or a non-linguistic type, are addressed. What exactly does it mean that communicators use signs in order to communicate? Based on Peirce’s distinction of signs, Rudi Keller asks in his contribution how human communication - understood as symbolic communication - emerges out of other sign uses, namely those of the symptomatic and iconic type.

Introduction: Beyond Words

5

With respect to the ongoing debate concerning the relation between sentence meaning and utterance meaning, this distinction is being reformulated in terms of a metamorphosis of signs, i.e. as a transition from a symbolbased to a symptom-based inference on the side of the addressees. The relation between form and meaning of utterances is also addressed in the second contribution of this section. Reinhard Blutner investigates, as previously mentioned, the relationship between current pragmatic paradigms, that of relevance theory and the theory of presumptive meanings in connection with further neo-Gricean approaches on the one hand, and optimality theory on the other. He shows that OT is compatible with these pragmatic paradigms in a synchronic as well as in a diachronic perspective. The contributions of the second section are devoted to the acquisition of inferential skills such as relevance inferences, scalars and metaphorical interpretation. They show that young children are able to perform pragmatic inferences, leading from the utterances perceived to the ascription of utterance meanings. The general issue is at what age exactly are children able to produce and understand utterance meanings based on inferential processes as described above – and in what way do they make use of them. This is also relevant for word learning as part and parcel of the acquisition of communicative behavior. Susanne Grassmann investigates the strategies two-year-old children use if they hear a novel word and she studies the inferences these children perform in order to disambiguate this word. Patrícia Amaral examines the mechanism of scalar implicatures connected with the adverb almost, whose meaning depends on a contextually given scale. In her paper, Cornelia Schulze shows that three-year-old children are able to understand indirectly conveyed requests. Since this pragmatic ability is based on the maxim of relevance, this demonstrates that young children obviously are able to understand relevance implicatures. Nausicaa Pouscoulous prooves that children are able to process fully novel metaphors at the age of three. The third section deals with the relation of pragmatics to the more structural aspects of language, i.e. phonology and grammatical relations. Billy Clark pursues the question as to what extent intonational meaning, which is conceived as weak procedural meaning, can effect the derivation of higher level explicatures in the sense of relevance theory. Frank Liedtke introduces the notion of a pragmatic template, which is defined as a structured whole containing information about the form of the utterance and central elements of the standardized context in which they are uttered. Helga Gese demonstrates in her contribution how pragmatic processes can influence differences in the reading of adjectival passives. Distinguish-

6 Frank Liedtke and Cornelia Schulze

ing target state and post state reading, she may show that the latter can serve as the basis for pragmatically enriched interpretations. Petra Schumacher and Jörg Meibauer rely on data of event-related brain-potentials in order to identify the role and the importance of encyclopedic knowledge, which is central for utterance interpretation beyond lexical competence and the knowledge of Gricean maxims. The nature of regularities and conventions in pragmatic interpretation is presented in the fourth section. Notions like memes, constructions and constraints are investigated, and it is asked whether they enable us to describe the pragmatic competence better than ad hoc strategies in interpretation. Chris Cummins and Napoleon Katsos examine quantifiers, numerical expressions and negators from an optimality theoretic standpoint, and they claim that, for the question of isomorphism between syntax and semantics, contextual aspects do not play a role. Elke Diedrichsen also works at the intersection of two paradigms, that of construction grammar and that of memetics (as part of a theory of cultural evolution). Considering three examples of linguistic and nonlinguistic behavior she demonstrates that an analysis in terms of cultural memes is able to capture novel forms of this behavior. Marco Mazzone highlights the role of regularities in pragmatics. Starting with the notion of generalized conversational implicatures, he develops an associative account of defaults which are not based on inferences but rather on the storage of regularities of past utterances. This is the content of this book in a nutshell. More detailed information about the sections and the papers within them is contained in the short introductions, which can be found at the beginning of each section. The contributions contained in this book are a selection of revised papers held at the International Leipzig Conference on Pragmatics “Beyond the Words”, which took place in May 2010. The editors wish to thank Julia Protze, who supported the editorial work of the book. Frank Liedtke and Cornelia Schulze Leipzig, August 2012

Introduction: Beyond Words

7

References Bach, Kent 1994 Conversational Impliciture, Mind and Language 9: 124–162. Cappelen, Herman, and Ernie LePore 2005 Insensitive Semantics. A Defense of Semantic Minimalism and Speech Act Pluralism. Oxford: Blackwell. 2006 Précis of Insensitive Semantics, Philosophy and Phenomenological Research 73 (2): 425–434. Carston, Robyn 2002 Thoughts and Utterances. The Pragmatics of Explicit Communication. Oxford: Blackwell. Levinson, Stephen C. 2000 Presumptive Meanings. The Theory of Generalized Conversational Implicatures. Cambridge / Mass.: MIT Press. Récanati, François 2004 Literal Meaning. Cambridge: Cambridge UP. 2010 Truth-conditional pragmatics. Oxford: Clarendon Press. Searle, John R. 1978 Literal Meaning, Erkenntnis 13 (1): 207–224. Sperber, Dan, and Deirdre Wilson, 1986 Relevance. Communication and Cognition. Oxford: Blackwell. Travis, Charles 1981 True and the False. The Domain of the Pragmatic. Amsterdam: Benjamins.

Section I. General concepts

Short introduction: General concepts The contributions of this section are centered around the question of how a pragmatic foundation of language is possible, especially with regard to the use of signs in order to realize communicative goals on the one hand, the description of language use in terms of constraints on the other. The role of pragmatic inferences in language interpretation by the addressees and their anticipation by language producers is analyzed from different directions. The starting point is based on an anthropological perspective, dealing with the use of symbolic signs by humans, either in a phylogenetic or an ontogenetic perspective. Rudi Keller analyzes communication generally as an activity which consists of influencing other members of the own species. The specific way human beings perform acts of communication is explicated using the terminology of Peirce; in this sense Keller distinguishes symptoms, icons and symbols. Human communication by means of symbols is the result of a developmental chain which consists of several processes of sign metamorphosis. If some higher animals are able to communicate with each other by simulating a symptom, e.g. expressing fear without there being any danger, this can be classified as an image of the specific symptom, i.e. an icon. Humans in turn are able to communicate not only by simulating a symptom, e.g. by yawning in order to express their own fatigue, but also to do this in a transparent manner, i.e. by letting the addressee know that this is a simulation. This sort of transparent communicative behavior is an instance of what Grice called non-natural meaning: openly letting someone know that one wants him to believe or to do something. If the iconization of symptoms is conventionalized, this may result in a new symbol. Symptoms, icons and symbols are not distinguished by different types of standing-for-relations, but by different types of inference from the perceived to its interpretation. There are causal (for symptoms), associative (for icons) and rule-based (for symbols) inferences. For a theory of communication by means of signs this means that one has to distinguish between different types of inferences from the perceived sign-material to the interpretation of this material. In distinguishing systematically the meaning of an utterance from its sense, Keller distinguishes two basic inferences: a symbol-based inference from what-is-uttered to what-is-said, which is the meaning of the utterance, and a symptom-based inference from what-issaid to what-is-meant, which he calls the sense of the utterance. The ra-

12

Frank Liedtke

tionale of the symptom-based inference is the means-end-relation, that is the relation between the uttered sign and the intention the utterer has in performing it. Thus the sense of the production of a sign is to be conceptualized as a symptom-based inference towards what the producer of this signs might have meant with it. Communication in the narrower sense is then defined as swaying another person to believe, to do or to feel something. In the case of linguistic communication, the addressee of an utterance uses the symbolic mode in order to determine what was said, the meaning, and the symptomic mode based on a means-end-analysis in order to determine what was meant by the utterance, the sense. Keller thus reformulates the relation between sentence meaning and utterance meaning in terms of sign metamorphosis, especially as a case of “symptomization” of symbols. Reinhard Blutner’s contribution too is dedicated to a principled account of what it means to communicate by means of specific linguistic devices with circumscribed properties. For this purpose he considers firstly the main approaches of contemporary pragmatics, Relevance Theory (Sperber and Wilson), the Theory of Presumptive Meanings as presented by Levinson, and the neo-Gricean framework (Horn). The second approach Blutner considers is Optimality Theory (OT), which had been applied successfully to the domains of phonology, morphology and syntax. The aim of the paper is to examine the contribution OT might make to pragmatics in the sense of an OT pragmatics. The author tries to reconcile the so called normative pragmatics as it is condensed e.g. in Grices cooperative principle with the explanatory theory of natural language represented by OT. One characteristic of OT pragmatics which brings it close to e.g. Relevance Theory (RT) is its naturalistic stance. A further point of agreement is that both, OT and RT share the basic assumption of semantic underdetermination on the one hand and contextualism on the other. Beneath the hearer-oriented pragmatics as represented by RT, also speaker-oriented normative pragmatics in the Gricean sense might be reconstructed by OT pragmatics. Eventually, also the neo-Gricean approaches based on two principles, the Q- and the Rprinciple, may be reconciled with OT pragmatics. The aim of the contribution lies in the discussion of the main paradigms in contemporary pragmatics, RT, Presumptive Meanings and the neoGricean accounts. It is then demonstrated that the theory of optimal interpretation is compatible with these three pragmatic approaches, and furthermore it is shown that the theory of bidirectional optimization is able to account for the synchronic and diachronic perspective in pragmatics.

Short introduction: General concepts

13

Both contributions, that of Keller and of Blutner, are to be seen as attempts to clarify the fundamentals of pragmatic interpretation of utterances on the basis of different types of inferences, on the basis of the Peircian theory of signs on the one hand, of the reconciliation of Optimality Theory and contemporary pragmatics on the other. They therefore form the first section of the present book. Frank Liedtke

Communication in the narrower and broader sense A reconstruction in terms of Sign Theory

Rudi Keller

1. Prologue “In the beginning was the Word”, we read in Luther’s translation of the Gospel according to St. John (John 1, 1). I lack all the necessary prerequisites for providing a theological interpretation of this passage, but I should like to suggest one correction from a linguistic point of view: in the beginning was not the Word, but rather the ability to communicate. The same applies to the matter of trade and money. We humans are not able to carry on trade because we have the institution of money at our disposal, rather, we have the institution of money at our disposal because we have the ability to carry on trade (cf. Keller 1994: 14). The analogy holds true: we are not able to communicate because we have “the word”, i.e., language, at our disposal, rather, we have our languages at our disposal because we have the capability of communicating in a very specific way. Language is not the prerequisite for our communicative ability, but rather a consequence of the latter. “Communicating” is a multifaceted concept. Animals communicate, humans communicate with machines, machines with one another, and then there are also the communicating tubes. It is not my intention to address all these forms of communication. I am only concerned with communication in the human sense of the term. In the first part of my lecture, I would like to clarify what that, in fact, means. 2. Communicating in a human sense Communicating, as Erica García has put it (1994: 344), is intelligent guesswork, a game that allows us as human beings to influence our fellow human beings. Humans have obviously discovered a trick as to how they can influence other members of their species in a very specific way. Other animals are not capable of such a trick. Let us consider the following three types of ability that (can be) at the disposal of animals. To be sure, only

16

Rudi Keller

humans are capable of the third type and it is this type that we call the ability to communicate in a human sense. 1.

All animals are in a position to perceive and interpret events that transpire in their environment. Some species of animals, for example, can perceive scents and “conclude” from the latter that there is prey nearby or, on the other hand, a dangerous hunter. Baboons can interpret as a sign of danger a particular form of movement in the blades of the steppe grass in their vicinity. This is a pattern of movement that typically occurs when a predatory cat is moving stealthily through the grass. In terms of sign theory, this means that animals are in a position to interpret events as symptoms that they can use to their benefit.

2.

Some animals are capable of carrying out an act with the intention of having another animal derive a particular interpretation from the latter. For example, the following behavior has been observed in baboons (Sommer 1989: 151). When a baboon senses danger and presumes that a cheetah or lion is moving through the brush, he will sometimes stand up straight on his hind legs in order to get a better view. Other baboons that observe his behavior recognize that their fellow baboon, the one who is standing up straight, senses danger and, if necessary, they will seek refuge in safer fields. Thus, these baboons themselves can interpret the reaction of their fellow baboon to a perceived symptom in turn as a symptom. (“I interpret your standing on your hind legs as a symptom of the fact that you have perceived a symptom indicating that a predatory cat is approaching.”) Now it occasionally occurs even in baboon families that a younger, adolescent member of the family gets into a row with a stronger member, but subsequently is forced to recognize that he is the weaker of the two and takes to flight. In cases like this, the following scenario has been observed: the younger baboon prepares to flee, but then abruptly halts, stands up, and looks off into the distance with a fixed stare. His stronger pursuer likewise stops, wheels around and flees. He obviously interprets the behavior of the younger baboon as a symptom of the fact that the latter has detected a symptom for danger. The point, however, is this: there was, in fact, no predatory cat anywhere in the vicinity. The adolescent baboon had “lied”. Baboons have, then, registered the experience that, as baboons, they are in a position to interpret a specific mode of behavior as a symptom for the perception of a symptom for danger. Moreover, they are also capable of utilizing this experience for their own benefit. They can simulate a symptom in order to convince a fellow baboon to arrive at a specific interpretation. Behavior of this type has been observed in a variety of ways in the animal world. Naturally, we humans also act

Communication in the narrower and broader sense

17

in this manner, but we go one step further and that is what lends the specifically human aspect to our ability to communicate. 3.

We human beings are in a position to make it clear to our fellow humans that we behave in a particular way in order to convince another person to adopt a specific interpretation. Thus, whenever we communicate in a human sense, we not only intend to have the other person adopt a particular interpretation, we also intend to have him recognize that we want to have him adopt that particular interpretation. In order to be able to communicate like a human being, the baboon would have to acquire the ability to go to a fellow baboon, stand in front of him on his hind legs, and look off in the direction of danger in order to make it clear to him: “See here, I am standing up in front of you so that you will recognize that there is danger approaching from the direction in which I am gazing.” That is, he would have to be able to make the simulation of the symptom public as a simulation; were that to happen, he would have taken the decisive step from deception to communication in the human sense using the Grice model. “One person informing the other about things”, as Karl Bühler puts it. The simulation of a symptom is, in a certain sense, an image of the symptom and thus an icon. Simulation, i.e., the icon, would, with time, only have to be performed through suggestion or insinuation. In this way, an arbitrary gesture of warning about danger would gradually emerge. Briefly put, communicating in a human sense means then, that one openly gives the other person a means of recognizing the point to which one would like to bring him (cf. Grice 1969). And how does the person addressed recognize the point to which the communicator would like to bring him? In our example, he recognizes this first and foremost through the fact that the means still has sufficient similarity with the original symptom. In the degree to which the application of the means becomes a convention, the aspect of similarity becomes obsolete, and an unmotivated means of communication arises. A symbol has emanated from a symptom via the intermediate step of the icon. I have given the name of “sign metamorphoses” to such processes (see Keller 1998: 143ff.). Precisely those people can interpret a symbol who are familiar with the convention of its usage. Those are people who know under what circumstances it is used for what purposes. Whenever we call that which allows the interlocutor to interpret a symbol its meaning, then we can say: the meaning of a symbol is its convention of usage.

18

Rudi Keller

3. Symptoms, icons, and symbols Up until now, I have used the terms symptom, icon, and symbol without intimating how I wish to use them. That point will now be addressed. The distinction between three types of signs can be attributed to Charles Sanders Peirce. He used the terms index, icon, and symbol (Peirce 1965). I prefer the term symptom over index since the former – at least for German ears – sounds more natural and more familiar. Moreover, there is a fundamental difference between the perception of signs that I prefer when compared to the way in which Peirce sees them. Peirce’s theory of signs belongs within the tradition of a representationalist concept of signs, while I have attempted to design a consistently instrumentalist theory. Simply stated, representationalists pose the question – in accordance with the scholastic definition of signs “aliquid stat pro aliquo” – what does a sign represent, while instrumentalists ask how a sign is used, i.e., how it is interpreted. And there is a good reason for that. Even given the most benevolent interpretation of the predicate “to stand for”, it is completely implausible to assume that all signs might stand for something. The famous spots associated with measles are a symptom (i.e., an index) for measles. Naturally, however, they do not represent this illness, but are rather a part of the syndrome. And the German word tschüs (‘good-bye’) does not stand for the departure (for otherwise tschüs would have the same meaning as the word Abschied, which, after all, also represents the departure). Rather, tschüs is a means for a person to take his leave in an informal manner. Those scholars who consider the “stands-for” relationship to be the decisive criterion assume that symptoms, icons, and symbols stand for something in diverse ways in each case and attempt to define these ways in a useful manner: ― ― ―

Naturalness is what characterizes the relationship of the symptom to whatever it designates; Similarity is what characterizes the relationship of the icon to whatever it designates; Arbitrariness is what characterizes the relationship of the symbol to whatever it designates.

I prefer at this time to avoid a detailed criticism of this attempt to arrive at definitions (included in Keller 1998: 102). Let me just say this: not every sign has, as already mentioned, a denotative function. The criteria of naturalness and similarity are extremely unspecific. The accent of my dialect is a symptom of my origins, but it is not natural in the common sense

Communication in the narrower and broader sense

19

of the term, and the stick figures to be seen on the doors of toilets have no similarity to men’s restrooms; nonetheless, they are icons. Moreover, similarity and arbitrariness are not mutually exclusive (see Keller 1998: 130ff.), and the three categories of naturalness, similarity, and arbitrariness are not on the same logical level. In keeping with my suggestion, the distinction between these three types of signs is defined solely through the method of interpretation. It is this that I would now like to clarify. Every sign possesses two basic characteristics: it can be perceived, and it can be interpreted. The question of how something is to be interpreted that cannot be perceived does not arise. And what cannot be interpreted does not have the character of a sign. Both of these points can certainly be disputed. There are people who hear divine voices, and they are confronted with the question of how to interpret them. I myself cannot hear them; but I can, for example, perceive the constellation of the stars in the skies. However, I doubt their ability to be interpreted. For me, then, divine voices and the constellation of the stars are incapable of being expressed as signs, while the contrary holds true for other people. In every case, interpreting what has been perceived means drawing conclusions. I notice that smoke is billowing out of the window of my neighbor’s house and conclude that there is a fire. During a flight with Lufthansa, I notice the image of crossed-out pig on my dining plate and conclude that the food is for Muslims. I hear my interlocutor say “Tschüs” and conclude that he now wishes to depart. In the case of the fire, I arrive at a causal conclusion: from the effect I derive the cause. In the second case, I make use of my gift of association: the little image on my plate reminds me that Muslims are prohibited from eating pork and so I can assume that the meals to be served will conform to the Koran. In the third case, I activate my knowledge regarding a rule of usage in the German language. Naturally, the information that I can derive from the signs themselves does not, in any of these cases, suffice – with respect to icons and symbols – to determine what is meant and, in the case of symptoms, what is relevant. The signs themselves are only one factor of the premises that enable me to make a relevant, i.e., intended, interpretation. They are augmented by a (shared) knowledge of the world, assessment of the interlocutor, shared cultural knowledge or even shared emotional judgments, in short, a common background, as Michael Tomasello (2008) calls it. As I have said: communicating is an intelligent guesswork. In short: we are in a position to draw exactly three types of conclusions: causal, associative, and rule-based. Quartum non datur. These are the basic methods of interpretation that we apply, among other things, for communi-

20

Rudi Keller

cating. I have elsewhere (Keller 1998: 99ff.) explained how icons or symbols are derived from symptoms and symbols from icons. Linguistic signs, i.e., symbols, do not fall from the skies. “Symbols grow”, as Charles Sanders Peirce (1965) said. In the interest of time, I will not go here into detail on the processes of these sign metamorphoses. But I would like to add that we can once again employ these three methods, the symptomic, the iconic, and the symbolic, on a higher level. When we employ the symptomic method on the linguistic level, we produce metonyms and the application of the iconic method on the linguistic level results in metaphors. Metonyms are thus meta-symptoms and metaphors meta-icons. When the symbolic method is applied on the level of symbols, it produces meta-symbols and those are citations. That is, we can employ these three basic methods (and only these three) on different levels and, among other things, derive new meaning from old material. In the iconic mode, for example, we communicate “with our hands and feet” when communicating with someone whose language is unfamiliar to us. But that is of no further interest to us here. I will be dealing primarily with the symptomic and the symbolic mode. The symptomic mode plays a much more important role in our communication than is generally assumed. Let me now demonstrate how this is the case. 4. Sense versus meaning As stated previously, I refer to signs that we interpret with the aid of causal conclusions as symptoms. We can differentiate three sub-types of causal conclusions: 1. When a detective discovers fingerprints at a crime scene, he interprets these as a symptom for the fact that the owner of these fingerprints was present at the crime scene. He derives the cause from the effect. 2. If an archeologist comes across a skull while excavating, he may possibly draw the conclusion that a person has been buried in this spot. He moves from the part to the whole. 3. And if the victim of an attack sees that the perpetrator is aiming his weapon at him, he runs for cover. He derives the aim from the means.

Communication in the narrower and broader sense

21

The cause-effect relation, the part-whole relation, and the means-aim relation are the relations that are used with the symptomic method. When one attempts an analysis based on sign theory, it is often not clear whether one can assume, in a given case, the cause-effect relation or the part-whole relation. If the physician diagnoses measles based on the spots on a person’s face, then one could assume that he is determining the cause from the effect since, in the final analysis, it was the illness measles that caused these spots to appear. But it would also not be at all wrong to assume that he is moving from the part to the whole. For the doctor perceives a part of the syndrome of the illness and from the latter draws his conclusions with respect to the complete clinical picture. However, this lack of certainty shall not concern us any further. The role of the means-aim analysis is particularly important for understanding the very normal acts of communication that we employ when we speak with one another. To show this, however, I must return once again to the concept of “meaning”. I refer to “meaning” in the tradition of Wittgenstein, namely, as the aspect of a sign that makes interpretation possible. In the case of linguistic symbols, this refers to the fact that their use in communication follows generally accepted rules. Simply put: the meaning of a linguistic sign is the rule governing its use. Knowing the meaning of tschüs is to know that this expression is generally used to take one’s leave in informal situations. Knowing the meaning of bachelor is to know that this expression is used to refer to unmarried, adult men, or to indicate of adult men that they are unmarried. With expressions such as bachelor, the rule parameters (unmarried, adult, etc.) are simultaneously the conditions of truth. Conditions of truth constitute a special case of the conditions of use. There are other types of conditions of use (rule parameters), but I do not intend to go into them in detail here (cf. Keller 2006). In the act of communicating we attempt to bring the person we are addressing to a point where he understands what we mean. What we mean by uttering a linguistic expression (a word or a sentence) is not its meaning! The aim of the communicative act is not that the addressee may recognize the meaning, but rather it is a means utilized by the communicator to achieve his aim, namely, to let the addressee recognize what he means by this act. And, in order to find that out, we must determine three aspects of each respective act of communication: the “what”, the “what for”, and – something that is often overlooked – the “how”. Let us take a look at one example. I find a piece of paper lying on the street on which is written: “I have a headache.” What do I know from this? Do I know that the author had a headache? No, for he might just have

22

Rudi Keller

quoted the remark, he could have meant it ironically or metaphorically. Naturally, I also do not know to whom “I” refers; that is, I cannot establish any fixed form of reference. I have absolutely no idea what the person who wrote this had in mind. What I do know, thanks to my linguistic competence, is the meaning of this sentence. That is to say, I know the rules that govern the usage of the words and the syntax. The language itself does not provide anything else. And there is no difference if I hear this sentence spoken aloud. The meaning of the word “I” also does not permit me to establish a fixed form of reference. For that, I have to see who said “I” or I have to hear it spoken and recognize the voice. I would like to propose the following stipulation: the meaning of an expression is not something that is supposed to be understood. One knows the meaning of an expression or one does not know it. If one knows its meaning, one possibly has a good chance of understanding what the speaker means by uttering this expression. Thus, one knows the meaning and one understands what is meant. I call the sense of the utterance that which is meant by the utterance. That is: one knows the meaning of an expression and one understands (or does not understand) the sense of an utterance. The meaning and the sense are two completely different categories. The sense is not a special case of a meaning or a meaning that is “restricted” by the context. Meanings are rules, the sense of something is an aim, an intention. In part, one understands the intentions of a speaker on the basis of the knowledge of the meanings, the rules of use (including syntax). And how does one find the bridge from the meaning of an expression to the sense of an utterance? Through the context, it is often maintained. But that does not tell us very much. My answer, based on sign theory, is as follows: with the aid of a meansaim analysis. In other words: on the one hand, I interpret the sentence in the symbolic mode – the “what” – and, on the other hand, I interpret the fact that it was uttered under given conditions, namely, in the symptomic mode – the “what for”. I understand the utterance of the symbol as a symptom for the fact that the communicator desires to effect a communicative intention towards me. In order to find out what the fact of the utterance might be a symptom for, I need to make a whole series of assumptions that do not constitute a part of my linguistic competence, but which are certainly part of my communicative competence: knowledge of the subject, the partner’s hypotheses, shared knowledge of the world, and shared knowledge regarding attitudes and emotions. The fact that all of these assumptions are of a hypothetical nature makes communicating, in principle, a rather risky business, a guessing game. The most important, basic as-

Communication in the narrower and broader sense

23

sumption, however, is the principle of rationality: regard the contributions that your interlocutors make to the conversation as rational acts. For, without this assumption, means-aim analyses are basically not possible. By no means does a communicative act always have to be cooperative! I can certainly hang up the receiver during a telephone conversation because I have become angry. Naturally, the act of hanging up the receiver is inspired by the wish to be understood. But if it is understood, this is not because it is a good example of cooperation, but rather an example for a rational act, or, more precisely: for a rational refusal of cooperation. In summary, it can be said that, by means of symbolic conclusions, I recognize what the other person has said (or done), by means of symptomic conclusions I recognize, to what purpose he has said (or done) something. It is also important to emphasize that it is not the symbols that are the object of a symptomic analysis, but rather the fact that they were uttered by the given communicator under the given circumstances. What happens when I am not successful in achieving a plausible means-aim analysis on the basis of the assumption of rationality? I could say: well, he’s simply acting irrationally, he’s crazy. This conclusion is not out of the question, but we normally hold off on it until the very end, when nothing else is to any avail. Normally, we try to preserve the assumption of rationality – as Grice has described it – by revising the interpretation of the symbol; we might try, for example, to interpret the utterance metaphorically or as something that was meant ironically. In other words, the failure of a means-aim analysis is a symptom for the fact that the utterance cannot be meant literally. So much for the “what” and “what for”. Let us now come to the third aspect, the “how”. 5. The symbolic mode and the symptomic mode of interpretation We have seen that communication in the sense in which we are interested in here is not the solution to a logistics problem (how do I transfer my thoughts from my head to yours?), but rather the solution to a problem of influencing. The person who communicates would like to convince the other person of something in particular. To clarify this, let us take a look at two situations which, on the surface, may appear to be very similar: Situation 1: A person is sitting in an auditorium listening to a lecture and has to yawn out of boredom.

24

Rudi Keller

Situation 2: A person turns to his companion during a lecture and pretends to yawn, indicating to her that he finds the lecture deathly boring. What is the difference between these two situations? In the first situation, the person demonstrates behavior that can certainly be interpreted by observers as a symptom of boredom, but which has not taken place with the intention of evoking this interpretation. In the second situation, however, that was precisely the intention. In this instance, we are dealing with a communicative act: a person does something in order to influence another person in a very specific way. The person yawning wishes to make his companion realize that he his deathly bored. The following conditions are fulfilled in Situation 2, in contrast to Situation 1: 1. The person communicating makes it clear that his yawning is not simply a reflex, but rather an intentional act. (He can achieve that, for example, by making his yawning somewhat “exaggerated”, so that it can be differentiated from sincere yawning.) 2. The communicator intends to have the addressee recognize that he is bored. 3. The communicator intends to have the addressee recognize that it is his intention to have her recognize through his yawning that he is bored. (cf. Grice 1969) I would call this form of communicating in a narrower sense. That is, something is done with the intention of having the addressee recognize something specific and also that he may take note of the fact that the person who does this wants him to be aware of his intention. The third condition is necessary to prevent the following scenario, for example, from also being included in the definition: I have guests, it is already two-thirty in the morning and I would be happy if they would gradually make an effort to go home. Since I would consider it impolite just to come out and ask them to go home, I yawn, in the hope that they might draw the conclusion that I am tired and use this as an excuse to depart. This is a human version of the aforementioned baboon trick. Now, one can justifiably ask: why should this case be excluded from the definition of

Communication in the narrower and broader sense

25

“communicating”? After all, this is also a method aimed at getting the other person to recognize something. The answer is that it makes sense first of all to define what communicating in a narrower sense means, specifically, in the sense in which we communicate when we speak in a normal way with a person. All other forms of communicating – for example, forms of clandestine, i.e., manipulative communicating – can be more specifically described under the mantle of the definition of open and clear communicating. In terms of sign theory, the person yawning in Situation 2, the simulated yawning, employs an icon as a means of communication. He uses the knowledge of the addressee that yawning is a sign of fatigue. By means of an artificial simulation he simultaneously creates an image of this symptom, i.e., an icon of the symptom of fatigue. Authentic yawning is a symptom of fatigue; openly simulated yawning is an icon of the symptom of fatigue. If our yawner had said, “Good Lord, this lecture is boring”, instead of yawning, his companion would have had to draw a symbolic conclusion. In both cases, the three aforementioned conditions are fulfilled, i.e., in both cases we are dealing with an example of flawless communication in the narrower sense. In our second example – getting rid of guests who stay late in a less than open way – the simulated gesture is not intended to be recognized as such. The guests are supposed to draw their conclusions on the basis of a (feigned) symptom: “He has to yawn, so, he must be tired. Then we ought to get going.” Since, in this case, the guests are not intended to recognize the intention behind the yawning – they should not even recognize that there is any intention at all behind the yawning – it is not a matter of authentic communication in the narrower sense. Rather, as we shall presently see, it is more of a borderline case. Frank Liedtke pointed out to me that, alternatively, it is quite possible that another, likewise symptomic, conclusion may play a role here. In our culture, yawning in public is frowned upon, or, put another way, it is advisable to suppress a yawn. That allows my guests to draw the following conclusion: “In public, a person makes an effort not to yawn openly. He has refrained from making any effort to suppress his yawning. The fact, that he has refrained from doing so is a symptom for me that he wants to make it clear to me that we should go home.” “One cannot not communicate” (Watzlawick et al. 1967: 51; emphasis in original) – this nicely formulate thesis is generally known. It is taken from the famous book, Pragmatics of Human Communication by Watzlawick, Beavin, and Jackson. Is it really the case that one cannot

26

Rudi Keller

avoid communicating? When communication is, by necessity, bound up with the performing of an act by the communicator – this is precisely what the first condition of our definition states – then this thesis cannot be correct, for one can refrain from acting. (To be sure, under certain conditions, refraining itself can be an act for which a person can even be called to account.) Watzlawick et al. take here as their basis a more extensive concept of communication. In its present sententious form, I consider this thesis overstated, but it does have a kernel of truth that is also relevant for our topic. I would now like to clarify this point. In order to assess this thesis adequately, one must know from the outset that Watzlawick, Beavin, and Jackson were not linguists, but rather psychologists and therapists. They use the following example to illustrate the plausibility of their thesis: “The man sitting in the overcrowded waiting room with his eyes closed, or staring at the floor in front of him, is conveying to others that he does not wish to talk or have anyone talk to him. Usually his neighbors will react in the correct manner and leave him in peace. This is no less an exchange of communication than an animated conversation” (Watzlawick et al. 1967: 49). Therapists would be ill-advised to confine their attention during their therapy sessions to what their patients said. All other signs from the patient could be just as important. And to a lesser degree, we do not act any differently in our daily life. We take note of how our interlocutor expresses himself, how he acts, etc, and not just what he says. So far, so good. But, is it appropriate to assume that the man in the overcrowded waiting room is “imparting” something to others through the way he is sitting, and that this is “no less an exchange of communication” than “an animated conversation”? This assumption erases the difference between the two, aforementioned yawning scenarios (cf. Alston 1967: 20f.). Let us consider an even more obvious example: I wear glasses for everyday use and speak with a Palatinate accent. This may be interpreted by many an interlocutor as a sign that my eyesight is weak and that am from the Palatinate. But have I really “imparted” that to him? Watzlawick’s thesis on the impossibility of not communicating rests, in the final analysis, on the following, false conclusion: everything that can be revealed through interpretation must also have been communicated. From Watzlawick’s theses that “behavior does not have an opposite” and “behavior is communication”, the conclusion can be drawn: “communication does not have an opposite”. By doing so, one has defined the concept of communication so broadly that it is no longer suited for making distinctions. However, I have already stated that this thesis has, in its properly

Communication in the narrower and broader sense

27

understood sense, a thoroughly reasonable kernel that I now propose to extrapolate. The conclusion that I am short-sighted and come from the Palatinate, a conclusion arrived at by my interlocutor based on the fact that I am wearing glasses and speak with a certain accent, can be clearly differentiated from my remark: “I am short-sighted and come from the Palatinate.” The latter represents a communicative statement, undertaken intentionally and to be interpreted in the symbolic mode. In the former case, however, we are dealing with the interpretation of unintentional aspects of my person, which, if necessary, are (or can be) interpreted in the symptomic mode. Naturally, the accrual of information on the part of the person doing the interpreting can be the same in both cases, and that is most certainly the reason why Watzlawick (et al.) include both cases under the concept of communication. I would call the first case “communication in the broader sense” and the second “communication in the narrower sense”. Communication in the narrower sense makes use of the symbolic mode of interpretation; the interpreter of communication in the broader sense uses the symptomic mode, for my glasses are, in the sense of sign theory, not a symbol of my shortsightedness, but rather a symptom of the latter. The reason why even communication in the broader sense is relevant for comprehending our normal, everyday processes of communication is as follows. We human beings are in a position to reflect on the fact that every aspect of a person (such as the glasses), and every aspect that accompanies his communicative actions (such as the accent), can be the object of symptomic interpretation. I cannot have any influence over most aspects of my person that could assist the other person as a basis for his symptomic interpretation. However, there are some that I can modify in a way that will allow me to derive the greatest possible advantages from these interpretations. In other words: we are in a position to exploit the symptomic mode in a communicative manner. For example, I can attempt to suppress my Palatinate accent if I wish to conceal my regional origins; or I can also emphasize it, if I intend to create the impression that we share a common geographical heritage. One of the most important accompanying factors of our communicative actions that can be – and generally are – made the object of symptomic interpretation – is the style in which a (written or oral) remark is composed. The person to whom the remark is addressed not only interprets the What and the What For of the matter communicated, but also the How. With reference to Watzlawick one could say: one cannot write without style. “Every remark has style”, writes Barbara Sandig (Sandig 2006: 2).

28

Rudi Keller

But what is style? In her comprehensive study, Barbara Sandig offers the following answer: “Style is the varied use of language that contains relevance for the community. Style is the HOW, the significant variation – with respect to both function and context – in the use of language and other communicatively relevant sign types” (Sandig 2006: 1). The key concepts in her characterization are variation and relevance. Style is predicated upon the possibility of variation. It is only in his sermon that the minister can shine, not in a prayer! That significance is accorded a choice with respect to the possible variants demonstrates that style has the character of a sign. I should now like to demonstrate in what sense this is the case. Let us consider a simple example of a stylistic alternative. I can write “wegen des [genitive case] schlechten Wetters” or, alternatively, “wegen dem [dative case] schlechten Wetter” (both meaning ‘on account of the bad weather’). Most speakers of German will share my opinion that the first variation is more “aesthetically pleasing”. Why is that the case? Some of you will think that it is simply incorrect to use the dative after the preposition “wegen”, but that is wrong. We write: “wegen ihm haben wir länger warten müssen” (‘we had to wait longer on his account’), not “wegen seiner”. (In his Die Leiden des Jungen Werthers (Sorrows of Young Werther), Goethe wrote: “wegen dem Gegensazze” (Goethe [1774] 1997: 32) and not “wegen des Gegensazzes” [‘because of the contrast’].) So, let us assume that both versions are, in principle, possible, and that they are also, in fact, quite frequently used. What constitutes the beauty of a grammatical case, in this instance, the genitive? Is a case endowed with an inherent aesthetic? Are there aesthetic and ugly cases? Such an assumption would most assuredly be absurd. The correct answer, in my opinion, is that the aesthetic evaluation does not concern the case itself, for the genitive is not something that can be the object of an aesthetic judgment. Rather, we draw very specific conclusions from the fact that an author has chosen this case in this particular situation. We assess the choice of the genitive as a symptom for the fact that the author has particular traits that we appreciate: as a symptom for his level of education, enjoying a certain social standing, for linguistic diligence, etc. Conversely, we assess the choice of the dative (in this situation) as a symptom for the fact that the author evinces certain traits that we frown upon. In truth, then, we do not subject the genitive itself to an aesthetic evaluation, but rather what we conclude from the fact that it was chosen by a person. Stated colloquially: being educated, not the genitive case, is “aesthetically pleasing”. The genitive is simply a symptom for the latter, and that is what we appreciate about it.

Communication in the narrower and broader sense

29

Let us consider a second example, a stylistic phenomenon with which everyone in Germany is familiar: the German of bureaucracy. The following sentence has been lifted from a business report of a well-known firm: Apart from insuring economic viability, XY Ltd. guarantees the successful implementation of the projects through to final completion through total project management and in its capacity as the general contractor. It is easy to “de-bureaucratize” a monstrosity such as this with relatively little effort. As the General Contractor, XY Ltd. assumes responsibility for the entire management of the project. In this way, we can guarantee not only the economic viability of the projects, but also their successful implementation from the planning phase through to completion. Here, as well, the majority of readers will agree with me that the “debureaucraticized” version reads better and sounds “more appealing”. What is it, however, that makes the bureaucratic variant “unappealing”? It is often said that bureaucratic German is more difficult to understand than texts that are not written in a bureaucratic mode. I cannot concur with this view in all instances. Whether a text is easy or difficult to understand depends on many factors, for example, the vocabulary, the transparency of the sentences, or the manner in which the text has been structured. However, a diction that is heavily laden with nouns and written in an impersonal style does not per se make a text incomprehensible. I believe that here, as well, the answer lies in the symptomic assessment. In Germany, administrative authorities and other bureaucracies do not enjoy a reputation for being sophisticated, customer-oriented, friendly, and efficient. Thus, it is easy to draw the conclusion: whoever writes like a bureaucrat, also behaves like a bureaucrat. This conclusion may be incorrect in any number of ways. There are, after all, bureaucrats who are quite friendly and efficient, and there are companies that tend to formulate their correspondence in a bureaucratic style, but which in no way behave in a bureaucratic manner. It is not important whether a symptomic conclusion is right or wrong. The only thing that is important is that the conclusion is drawn in the first place. Here, as well, it is true that it is not words or syntactical constructions that are pleasing or not unpleasing, but rather the human traits that can be derived from the use of words and syntactical constructions by means of the symptomic conclusion. In general it may be said that the way in which a person expresses himself can be interpreted as a symptom of his

30

Rudi Keller

education, his political convictions, his social or regional origins, his intelligence, his charm, his wittiness and humor, his sincerity, his professional competence, etc. Moreover, the How can also influence the interpretation of the What and the What For. For example, a highly unusual mode of expression can be a symptom designed to have the utterance interpreted ironically. What we perceive to be the style of a text is nothing other than a confused mixture of all the symptomic conclusions that we can draw from the way in which the text (whether written or spoken) has been put together. The person who writes meines Vaters Auto (‘my father’s car’) allows people to draw quite different conclusions regarding his social origins or his level of education as they will of the person who uses the expression meinem Vater sein Auto (‘my dad, his car….’). The two expressions are not different with respect to their comprehensibility, but solely with regard to their symptomic potential. Their stylistic aesthetics are something that the recipient has deduced. Let us sum up briefly: communication in the narrower sense of the word means to do something intentionally with the aim of swaying another person to do, to think, or to feel something by letting him recognize of what one would like to do, to think, or to feel him. In a normal case, i.e., in the case of linguistic communication, the addressee employs the symbolic mode to determine the What, to discover what the communicator has said. Once he knows that, he subjects the fact that the communicator has said precisely this in the given situation and context to a means-aim analysis in the symptomic mode. He does this in order to find out what the communicator meant by what he said. Apart from that, the addressee has the opportunity of interpreting the choice of stylistic alternatives as a symptom of what the addressee can derive from the communication. That an author chooses to communicate in the manner that he does can be interpreted as an expression of his education, his vanity, his competence, his stupidity, his social background, his credibility, and much more. A person who reflects on this while communicating will attempt to construct his remarks in such a way that the symptomic interpretation of the How emerges in his favor. It goes without saying that this applies to both oral and written communication. Symptoms that are initiated are, to be sure, no longer real symptoms – but there is a continuum. Were I to place particular emphasis on my Palatinate accent, I would still accept this as a symptom. However, were I to imitate a Swabian accent in order to deceive my interlocutor with respect to my regional origins, then that is just as little a symptom as the aforementioned yawning that was aimed at getting guests

Communication in the narrower and broader sense

31

to depart. A healthy person who simulates illness during a physical examination does not evince any true symptoms. Nonetheless, he communicates in the symptomic mode. In conclusion: communicating in the narrower sense of the word normally means to work with symbols, specifically, in the Grice mode. Even then, however, we have to draw symptomic conclusions when we want to explain the What For of a communicative act, when we want to understand what the communicator meant by what he said. Apart from this, the How plays a prominent role, above all, in connection with what one refers to as the style of an utterance. Since (up to a point) we have at our disposal a choice in the means of communication, we can also exploit the choice of the How in a communicative manner. This form of communicating – I have called it “communicating in the broader sense” – usually does not take place within the Grice mode. For normally I would not at all want the other person to recognize that I am only expressing myself in this way, and not in another, so that he – let us say – might consider me to be clever and credible. Translation by Winder McConnell References Alston, William P. 1967 Expressing. In Philosophy in America, Max Black (ed.), 15–34. Ithaca N.Y.: Cornell University Press. Bühler, Karl 1990 Theory of Language. The representational function of language, engl. trans. by Donald Fraser Goodwin, Achim Eschbach (ed.). Amsterdam and Philadelphia: Benjamins. Original edition, Sprachtheorie. Die Darstellungsfunktion der Sprache, Jena: Fischer, 1934. García, Erica 1994 Reversing the Status of Markedness. Folia Linguistica 28: 329–362. Goethe, Johann Wolfgang 1997 Die Leiden des jungen Werthers. Nachdruck der Erstausgabe von 1774. München: DTV (Bibliothek der Erstausgaben). Grice, Herbert Paul 1969 Utterer’s Meaning and Intentions. The Philosophical Review 78: 147–177.

32

Rudi Keller

Keller, Rudi 1994 On Language Change. The Invisible Hand in Language. London and New York: Routledge. 1998 A Theory of Linguistic Signs. Oxford, New York: Oxford University Press. 2006 On Evaluating. Values and Evaluating, Rudi Keller and Winder McConnell (eds.), 23–34. Tübingen, Basel: A. Francke. Peirce, Charles Sanders 1965 Elements of Logic. Collected Papers of Charles Sanders Peirce. Vol. 2. (CP), Charles Hartstone and Paul Weiss (eds.). Cambridge, Mass.: Belknap Press of Harvard University Press. Sandig, Barbara 2006 Textstilistik des Deutschen. 2., völlig neu bearbeitete und erweiterte Auflage. Berlin, New York: de Gruyter. Sommer, Volker 1989 Lügen haben lange Beine. GEO Wissen 2: 149–152. Tomasello, Michael 2008 Origins of Human Communication. Cambridge, Mass and London: MIT Press. Watzlawick, Paul, Janet Helmick Beavin, and Don D. Jackson 1967 Pragmatics of Human Communication. A Study of Interactional Patterns, Pathologies, and Paradoxes. New York: Norton.

Pragmatics in Optimality Theory Reinhard Blutner

1. Introduction Optimality theory is an integrated approach to cognition that combines the advantages of symbolic, constraint-based models with the advantages of subsymbolic, neuron-style models of cognition (Smolensky and Legendre 2006). In the study of natural language, OT was successfully applied to the main linguistic disciplines phonology, morphology and syntax, and also to the explanation of natural language acquisition and other performance traits. OT pragmatics is an application of the integrated approach to the domain of Gricean pragmatics. It has its origin in the attempt to explain certain phenomena of lexical pragmatics (Blutner 1998) and is inspired by the optimal interpretation approach proposed by Hendriks and de Hoop (2001). The view of seeing OT pragmatics within the scope of a naturalistic (explanatory) approach to cognition (as represented by the main proponents of OT) is not without problems. This has to do with the normative character that is attributed to the Gricean setting. Speakers, as Grice puts it, must make their contribution such as is required, at the stage at which it occurs, by the accepted purpose or direction of the talk exchange in which (they) are engaged. (Grice 1975: 45)

It’s obvious that this principle of cooperation is normative, and so are Grice’s conversational maxims. If a person acts in a particular situation in a particular way we can ask why she did it the way she did; alternately, we can ask if it was reasonable what the person did, and if other options were possibly more reasonable in the given situation. Good Griceans are expected to ask the second type of questions whereas the first question is expected to be asked by cognitive scientists. While the normative and the naturalistic aspects of understanding human actions can be clearly separated from each other that does not mean they predict different action patterns in most cases. The idea of a rational world isn’t so irrational to be excluded in ordinary affairs. Evolutionary game theory has presented us with many

34

Reinhard Blutner

examples demonstrating that the reasonable is naturally arising (Axelrod 1984). In other words, though there is a philosophical gap between Gricean pragmatics as a normative theory and OT as a scientific, explanatory theory of natural language there is not a deep empirical conflict between an interpretation oriented pragmatics and a speaker ethics. It seems the speaker better be cooperative or pretend to be cooperative if she wants to use language to bring about effects in hearers. The naturalistic stance taken by OT pragmatics is one characteristic that brings it close to relevance theory (Sperber and Wilson 1986/1995; Sperber and Wilson 1995). Another point of agreement has to do with the way OT pragmatics views the division of labour between semantics and pragmatics. Taking the lead from Atlas (e.g. Atlas 2005), both relevance theory (RT) and OT pragmatics reject the doctrine of literal meaning. And both approaches conform to the ideas of (i) (ii)

semantic underdetermination contextualism (the suggestion that the mechanism of pragmatic interpretation is crucial both for determining what the speaker says and what he means).

In the broad view of OT, this framework can be seen as a general scheme that is useful for expressing many different and possibly diverging views. For instance, it is possible to give optimality-theoretic reconstructions of a speaker-oriented normative pragmatics like the one developed by Grice. It is also possible to reconstruct hearer-oriented naturalistic pragmatics as in RT (Hendriks and de Hoop 2001; Zeevat 2007b). These systems are important for online synchronic accounts of speaking and interpretation. But – perhaps most surprisingly – it is also possible to reconstruct the neo-Gricean systems of Horn (1984), Atlas and Levinson (1981) and Levinson (2000). In contrast to RT where there is only one fundamental pragmatic principle (the presumption of optimal relevance), the neo-Gricean systems have two opposing optimization principles, the Qand the I-principle (Atlas and Levinson 1981; Horn 1984 who writes R instead of I) by two simultaneous optimization directions (the speaker and the hearer direction) and so obtain a bidirectional OT pragmatics. OT pragmatics in the narrower sense will start from this system and will show that Levinson’s M-principle (iconicity) can be reduced to it. The system can also explain the emergence of mono-directional pragmatic systems that

Pragmatics in Optimality Theory

35

can account for online incremental interpretation in the style of RT. Given the divergences within the neo-Gricean camp1, it cannot be expected that a coherent theory like bidirectional OT-pragmatics can reconstruct all the views of all representatives of this camp. The present chapter aims to give an overview of the application of OT to the domain of pragmatics. The assumptions (i) and (ii) are essential requisites for a natural theory of pragmatic interpretation. In section 2, I will introduce the three main views conforming to these assumptions: (a) RT, (b) Levinson’s (2000) theory of presumptive meanings, and (c) the neoGricean approach. In section 3, I explain the general paradigm of OT and the idea of bidirectional optimization. I show how the idea of optimal interpretation can be used to restructure the core ideas of these three different approaches. Further, I argue that bidirectional OT has the potential to account both for the synchronic and the diachronic perspective of pragmatic interpretation. Section 4 lists relevant examples of using the framework of bidirectional optimization in the domain of pragmatics. Section 5 provides some general conclusions. It argues that OT pragmatics has the potential to account both for the synchronic and the diachronic perspective in pragmatics. This bolsters the way for a deeper understanding of the idea of naturalization and (cultural) embodiment in the context of natural language interpretation. 2. The naturalization of pragmatics: three variations on Grice The naturalization of pragmatics refers to a research program that aims to provide a cognitively realistic picture of utterance interpretation and production. Hence, the proponents of this program such as relevance theorists take the stance of seeing natural language interpretation as a cognitive phenomenon and thus considering the basic principles of communication as a consequence of the nature of human cognition. A prerequisite of this program deals with the levels of cognitive representations and the boundary between semantics and pragmatics. There is a strong tendency among current researchers to follow the tradition of radical pragmatics and to accept the following three claims (Jaszczolt 2010):

1 For instance, Horn (2005) points out that Levinson’s (2000) view is very close in important respects to that of RT.

36 1.

2.

3.

Reinhard Blutner

There is a level of logical form or semantic representation. The representations of this level do not necessarily provide truth conditions. Rather, they underspecify truth-conditional content in a number of ways. There is a mechanism of enriching underspecified representations; sometimes this mechanism is called development of logical form. The result of this development is propositional content. It expresses the utterance meaning of the expression under discussion. There is a level of implicatures proper, understood as separate thoughts implied by the utterance. It is implicit propositional content that can be inferred from the explicit content mentioned in 2.

Obviously, the consensus is about rejecting the Gricean doctrine of literal meaning (logical form conforms to literal meaning), accepting the role of underspecification (logical forms are underspecified with regard to the expressed semantic content) and acknowledging that implicature is a graded category (some implicatures are closer to LF than others). Before I come to a discussion of three variations on Grice and the naturalization of conversational implicatures in utterance interpretation it is useful first to introduce the distinction between global and local approaches to conversational implicatures (Chierchia 2004). According to the global (neo-Gricean) view one first computes the (plain) meaning of the sentences; then, taking into account the relevant alternatives, one strengthens that meaning by adding in the implicature.’ (Chierchia 2004: 42). This contrasts with the local view, which first introduces pragmatic assumptions locally and then projects them upwards in a strictly compositional way where certain filter conditions apply. Representatives of the global view are Atlas and Levinson (1981), Gazdar (1979), Horn (1984), Soames (1982), Krifka (1995), Blutner (1998), Sauerland (2004), Sæbø (2004), and Geurts (2010); the local view is taken by Chierchia (2004), Levinson (2000), and relevance theory (e.g. Sperber and Wilson 1986/1995; Carston 2002). Usually, the globalists argue against the local view and the localists against the global view. I will argue, instead, that proper variants of both views are justified if a different status is assigned to the two views: global theories provide the standards of rational discourse and correspond to a diachronic, evolutionary scenario; local theories account for the shape of actual, online processing including the peculiarities of incremental interpretation. In this way, I will argue that seemingly conflicting approaches such as relevance theory and the neo-Gricean approach are much closer related than expected by its opponents. In section 3, once more OT will

Pragmatics in Optimality Theory

37

prove his power of unification in giving hints how to relate these different frameworks in a systematic way. RT assumes the representational/computational view of the mind, and, on this basis, gives a naturalization of pragmatics adopting Jerry Fodor’s language of thought hypothesis (Fodor 1975). The central thesis of RT is the communicative principle of relevance, according to which utterances convey a presumption of their own optimal relevance. In other words, any given utterance can be presumed: (i) (ii)

to be at least relevant enough to warrant the addressee’s processing effort to be the most relevant one compatible with the speaker’s current state of knowledge and her personal preferences and goals.

From these two assumptions relevance theorists derive the following general procedure that the cognitive system follows in comprehending an utterance (Sperber, Cara, and Girotto 1995: 95): (a) test possible interpretations in their order of accessibility, (b) stop once the expectation of (optimal) relevance is satisfied (i.e. a certain context-dependent threshold value of relevance is reached). The procedure makes sure that the wanted effect (a certain value of relevance) is reached with the minimal cognitive effort. Levinson’s (2000) theory of presumptive meaning is a chameleon that in a certain sense adapts general assumptions of RT and in another sense crucially conflicts with RT, for instance in assuming more than one basic principle (maxim) for formulating the interpretational mechanism. In short, these are the general assumptions: (i)

(ii)

Differing from both RT and the standard neo-Gricean view, Levinson assumes three levels of meaning corresponding to sentence meaning, utterance-type meaning and utterance-token meaning utterance-type meanings are in correspondence with Grice’ generalized conversational implicatures. They are a matter of preferred interpretation calculated by a particular default mechanism. Basically, there are three such defaults or heuristics: Q-heuristic: What isn’t said is not the case I-heuristic: What is expressed simply is stereotypically exemplified M-heuristic: What’s said in an abnormal way isn’t normal

38

Reinhard Blutner

(iii)

In contrast to Grice’ generalized conversational implicatures, which are calculated in a global manner, presumptive meanings are local, i.e. they arise at the point at which they are triggered (for instance, the word some triggers the default interpretation NOT ALL via the Q-heuristics). The feature of local pragmatics is essential to artificial intelligence pragmatics (e.g. Hobbs and Martin 1987) and likewise to RT.

Presumptive meanings are very useful for understanding natural language interpretation, especially for explaining the predominantly incremental character of utterance comprehension. Neo-Griceans (Atlas and Levinson 1981; Horn 1984; Blutner 1998; e.g. Atlas 2005; Horn 2005) are assuming two countervailing optimization principles: the Q-principle and the R-principle.2 The first is oriented to the interests of the hearer and looks for optimal interpretations; the second is oriented to the interests of the speaker and looks for expressive optimization. Here is a standard presentation of the two principles (Horn 1984, 1989, 2004, 2005): The Q-Principle (Hearer-based): Make your contribution sufficient! Say as much as you can! (modulo R) (Grice’s first quantity maxim and the first two manner maxims) The R-Principle (Speaker-based): Make your contribution necessary! Say not more than you must! (modulo Q) (Grice’s second quantity maxim, relation maxim and the second two manner maxims) It is tempting to identify the Q-principle with Levinson’s Q-heuristic and the R-principle with the I-heuristics. However, they are not identical though there is a correspondence between them. The difference has to do with the different status of principles in the global, neo-Gricean pragmatics on the one hand and heuristics (defaults) in Levinson’s local pragmatics on the other hand. According to the neo-Gricean picture the principles consti2 In OT, these ‘principles’ correspond to different directions of optimization where the content of the optimization procedure is expressed by particular OT constraints. This will be pointed out in more detail in the following section.

Pragmatics in Optimality Theory

39

tute a kind of communication game – either between real speakers and hearers or between fictive speakers and hearers in the mind of a language user. In this game both principles are applied in a recursive way (corresponding to the modulo-clause in the formulation of the principles). In Levinson’s theory, no such interaction between real or fictive speakers/hearers takes place. Instead, presumptive meanings are default interpretations and they are processed in a nearly automatic way. No ‘mind reading’ facilities or other mechanisms of controlled processing are required.3 The difference will become quite clear in the following section when I give formalization in terms of bidirectional OT. Sometimes it is stressed that there is a fundamental difference in perspective and goals between the neo-Gricean and the RT approaches to pragmatics. For instance, Horn (2005) claims that neo-Griceans typically take the normative stance whereas relevance theorists take the naturalist stance. However, I agree with Carston (2005) who argues that RT (as does Horn’s theory) makes some predictions about why the speaker, given her communicative intention, utters what she utters. Further, the difference between the normative stance and the naturalistic stance should not be overestimated because in practice there is seldom a deep empirical conflict between the two stances (Spohn 1993). A much more important question concerns the status of the theory with regard to synchrony versus diachrony. Both RT and Levinson’s theory of presumptive meaning takes the synchronic view where neo-Griceans take both views (and, sometimes, confuse them). In the following section, I will discuss how OT relates both views/perspectives.

3 However, presumptive meanings can demand a lot of effort as soon ‘conflicts’ arise and the corresponding assumption has to be cancelled. Conflict resolution can be very resource demanding. Hence, for the overall mechanism we have to take into account the peculiarities of controlled processing. Of course, this does not refer to any mind reading facilities.

40

Reinhard Blutner

3. The framework of OT OT can be seen as a general framework that systematizes the use of optimization methods in linguistics.4 One component of OT is a list of tendencies that hold for observable properties of a language. These tendencies take the form of violable constraints. Because the constraints usually express very general statements, they can be in conflict. Conflicts among constraints are resolved because the constraints differ in strength. Minimal violations of the constraints (taking their strength into account) define optimal conflict resolutions. Standardly, OT specifies the relation between an input and an output. This relation is mediated by two formal mechanisms, GEN and EVAL. GEN (for Generator) creates possible output candidates on the basis of a given input. EVAL (for Evaluator) uses the particular constraint ranking of the universal set of constraints CON to select the best candidate for a given input from among the candidate set produced by GEN. In phonology and syntax, the input to this process of optimization is an underlying linguistic representation. The output is the (surface) form as it is expressed. Hence, what is normally used in phonology and syntax is unidirectional optimization. Obviously, the point of view of the speaker is taken. This contrasts with OT semantics where the view of the hearer is taken (de Hoop and de Swart 2000; Hendriks and de Hoop 2001). Bidirectional optimization (Blutner 1998, 2000) integrates the speaker and the hearer perspective into a simultaneous optimization procedure. In pragmatics, this bidirectional view is motivated by a reduction of Grice’s maxims of conversation to two principles: the R-principle, which can be seen as the force of unification minimizing the speaker’s effort, and the Qprinciple, which can be seen as the force of diversification minimizing the auditor’s effort. In a slightly different formulation, the R-principle seeks to select the most coherent interpretation and the Q-principle acts as a blocking mechanism which blocks all the outputs which can be expressed more economically by an alternative linguistic input. This formulation makes it quite clear that the neo-Gricean framework can be conceived of as a bidirectional optimality framework which integrates the speaker and the hearer perspective. Whereas the R-principle compares different possible interpretations for the same syntactic expression, the Q-principle compares differ4 An overview is given in Smolensky and Legendre (2006). For OT pragmatics the reader is referred to Blutner and Zeevat (2004) and Blutner, de Hoop, and Hendriks (2005).

Pragmatics in Optimality Theory

41

ent possible syntactic expressions that the speaker could have used to communicate the same meaning. I will give a very schematic example in order to illustrate some characteristics of the bidirectional OT. Assume that we have two forms f1 and f2 which are semantically equivalent. This means that GEN associates the same interpretations with them, say m1 and m2. We stipulate that the form f1 is less complex (less marked) than the form f2 and that the interpretation m1 is less complex (less marked) than the interpretation m2. This is expressed by two markedness constraints: F for forms and M for interpretations – F prefers f1 over f2 and M prefers m1 over m2. This is indicated by the two leftmost constraints in table (1). Table 1. Markedness and bias constraints in a 2-forms u 2-interpretations design F

M

FoM

*

*

*Fo*M

Fo*M

F*oM

*



*

*

* *

*

Besides the markedness constraints, four so-called linking constraints can be formulated. There are precisely four independent linking constraints in the present example. The linking constraint FoM says that simple (unmarked) forms express simple interpretations. Hence, this is a straightforward formalization of Levinson’s (2000) I-heuristics as an OT constraint. The constraint *Fo*M says that complex forms express complex interpretations, and this is an expression of Levinson’s M-heuristics5. The two remaining bias constraints express the opposite restrictions. In the present case linking constraints can be seen as lexical stipulations that fix a forminterpretation relation in an instance-based way. With only two forms and two meanings, the substance of the Q-heuristics is not really different from that of the M-constraint.6 In the so-called strong version of bidirectional OT, a form-interpretation pair is considered to be (strongly) optimal iff 5 Levinson’s M-principle should not be confused with the markedness constraint M introduced in Table 1. 6 Harmonic alignment (Prince and Smolensky 1993; Aissen 2003) is precisely the fact that these two linking constraints hold.

42 ― ―

Reinhard Blutner

Interpretive Optimization: no other pair can be generated that satisfies the constraints better than and Expressive Optimization: no other pair can be generated that satisfies the constraints better than .

From the differences of markedness given by the constraints F and M the ordering relation between form-meaning pairs can be derived as shown in Figure 1. The preferences are indicated by arrows in a two-dimensional diagram. Such diagrams give an intuitive visualization for the optimal pairs of (strong) bidirectional OT: they are simply the meeting points of horizontal and vertical arrows.7 The optimal pairs are marked with the symbol $ in the diagram.

f1

$

q

f2

q m1

q m2

Figure 1. Diagram to illustrate strong bidirectionality

The scenario just mentioned describes the case of total blocking where some forms (e.g., *furiosity, *fallacity) do not exist because others do (fury, fallacy). However, blocking is not always total but may be partial. This means that not all the interpretations of a form must be blocked if another form exists. McCawley (1978) collects a number of examples demonstrating the phenomenon of partial blocking. For example, he observes that the distribution of productive causatives (in English, Japanese, German, and other languages) is restricted by the existence of a corresponding lexical causative. Whereas lexical causatives (e.g. (1a)) tend to be restricted in their distribution to the stereotypical causative situation (direct, unmediated causation through physical action), productive (periphrastic) causatives tend to pick up more marked situations of mediated, indirect causation. For 7 Dekker and van Rooy (2000), who introduced these diagrams, gave bidirectional OT a game theoretic interpretation where the optimal pairs can be characterized as so-called Nash equilibriums.

Pragmatics in Optimality Theory

43

example, (1b) could have been used appropriately when Black Bart caused the sheriff’s gun to backfire by stuffing it with cotton. (1)

a. b.

Black Bart killed the sheriff Black Bart caused the sheriff to die

To make things concrete we can take f1 to be the lexical causative form (1a), f2 the periphrastic form (1b), m1 direct (stereotypic) causation and m2 indirect causation. Typical cases of partial blocking are found in morphology, syntax and semantics. The general tendency of partial blocking seems to be that "unmarked forms tend to be used for unmarked situations and marked forms for marked situations" (Horn 1984: 26) – a tendency that Horn (1984: 22) terms "the division of pragmatic labour". There are two ways of avoiding total blocking within the bidirectional OT framework and to describe Horn’s division of pragmatic labour. The first possibility makes use of linking constraints and fits the intended forminterpretation relation by stipulating the appropriate ranking of the constraints such that partial blocking comes out. Let us assume that the two bias-constraints FoM and *Fo*M are higher ranked than the rest of the constraints. This can be depicted as in Figure 2a. Hence, strong bidirectionality can be taken as describing Horn’s division of pragmatic labour when the appropriate linking constraints are dominating.

f1

$

[FoM]

q

f1

q m1

$ m2

q [M] [F]

[*Fo*M]

f2

$

f2

q m1

($) m2

Figure 2. Two ways of describing Horn’s division of pragmatic labour: (a) by assuming two dominant bias constraints; (b) by assuming markedness constraints and weak bidirectionality

The second possibility is to weaken the notion of (strong) optimality in a way that allows us to derive Horn’s division of pragmatic labour by

44

Reinhard Blutner

means of the evaluation procedure and without stipulating particular bias constraints. Blutner (2000) proposes a weak version of two-dimensional OT, according to which the two dimensions of optimization are mutually related: Super-Optimality A form-interpretation pair is called super-optimal iff Interpretive Optimization: no other super-optimal pair can be generated that satisfies the constraints better than ; Expressive Optimization: no other super-optimal pair can be generated that satisfies the constraints better than . This formulation looks like a circular definition, but Jäger (2002) has shown that this is a sound recursive definition under very general conditions (well-foundedness of the ordering relation). The important difference between the weak and strong notions of optimality is that the weak one accepts super-optimal form-meaning pairs that would not be optimal according to the strong version. It typically allows marked expressions to have an optimal interpretation, although both the expression and the situations they describe have a more efficient counterpart. Figure 2b shows that the weak version of bidirectionality can explain the effects of partial blocking without the stipulation of extra bias constraints; especially it can explain why the marked form f2 gets the marked interpretation m2. This is a consequence of the recursion8 implemented in weak bidirectionality: the pairs and are not super-optimal. Hence, they cannot block the pair and it comes out as a new superoptimal pair. In this way, the weak version accounts for Horn’s pattern of the division of pragmatic labour. The two parts of Figure 2 describe the same set of solution pairs but the calculation of the solutions is completely different in the two cases. In the first case unidirectional optimization (either hearer or speaker perspective) is sufficient to calculate the solution pairs. It is plausible to assume that this kind of OT systems can be used to construct cognitively realistic models of online, incremental interpretation (Blutner 2006).

8 In the original formulation given in section 2, the recursion is indicated by the modulo-clause.

Pragmatics in Optimality Theory

45

The second case – using the recursion of weak bidirectionality (superoptimality) – has a completely different status. Because of its strictly nonlocal nature the proposed algorithm that calculate the super-optimal solutions do not even fit the simplest requirements of a psychologically realistic model of online, incremental interpretation (Zeevat 2000; Beaver and Lee 2004)9. The proper understanding of weak bidirectionality relates best to an off-line mechanism that is based on bidirectional learning (Blutner et al. 2002; Benz 2003b; Van Rooy 2004). In these approaches the solution concept of weak bidirectionality is considered as a principle describing the results of language change: super-optimal pairs emerge over time in language change. This relates to the view of Horn (1984) who considers the Q- and the I-principle as diametrically opposed forces in language change. This conforms to the good old idea that synchronic structure is significantly informed by diachronic forces. For the sake of illustration let us go back to our example illustrated in (1). Let us assume a population of agents who realize speaker- and hearer strategies based exclusively on the markedness constraints F and M. In this population each content is expressed in the simplest way (f1) and each expression is understood in the simplest way (m1). Let us assume further that these agents communicate with each other. When agent x is in the speaker role and intends to express m1, then expressive optimization yields f1. Agent y is a hearer who receives f1 and, according to interpretive optimization, he gets the interpretation m1 – hence the hearer understands what the speaker intends: successful communication. Now assume the speaker wants to express m2. With the same logic of optimization he will produce f1 and the agent y interprets it as m1. In this case, obviously, the communication is not successful. Now assume some kind of adaptation either by iterated learning or by some mutations of the ranked constraint system (including the bias constraints). According to this adaptation mechanism the expected ‘utility’ (how well they understand each other in the statistical mean) can improve in time. In that way a system that is evolving in time can be described including its special attractor dynamics. In each case there 9 There are several arguments why bidirectional OT cannot yield an online mechanism of linguistic competence. Beaver and Lee (2004) argue that if more rounds of optimization are allowed, the bidirectional OT-model severely overgenerates in the sense that in later rounds peculiar new form-meaning pairs will emerge as winners. Before the Beaver and Lee paper, Zeevat (2000) argued against the symmetric view of OT pragmatics starting from the famous rat/rad problem and its pragmatic counterparts.

46

Reinhard Blutner

is a stabilizing final state that corresponds to the system of Figure 2a where the two Levinsonian (2000) constraints I (= [FoM]) and M (= [FoM]) outrank the rest of the constraints. It is precisely this system that reflects Horn’s division of pragmatic labour. The only condition we have to assume is that the marked contents are less frequently expressed than the unmarked contents.10 Hence, the important insight is that a system that is exclusively based on markedness constraints such as in Figure 2b is evolutionary related to a system based on highly ranked bias constraints such as in Figure 2a. I will use the term fossilization for describing the relevant transfer.11 Now I come back to the earlier goal of giving an OT reconstruction of the three variations on Grice (section 2). For reconstructing Levinson’s (2000) presumptive meaning theory, unidirectional optimization is sufficient where a system of OT constraints has to be formulated conforming to his I, Q and M heuristics and Levinson’s putative ranking Q > M > I. The unidirectional optimization procedure (interpretive optimization) is conform with a local approach to conversational implicatures, one which satisfies the requirements of incremental interpretation. The neo-Gricean approach, on the other hand, is globalist in nature. Hence, the idea of (weak) bidirectional optimization fits best to this theory and can be used for a straightforward formalization. Unsurprisingly, this conception can be seen best from a diachronic perspective, at least so far we take a naturalistic stance towards Gricean pragmatics. As a model of actual language interpretation (or production) this approach does not make real sense and never was designed for this purpose. Like Levinson’s (2000) approach, RT conforms to the localist approach and can be formulated in terms of unidirectional optimization. Let us stipulate a constraint EFFECT for describing the wanted effect (a certain value of relevance) and a constraint EFFORT for describing the cognitive effort. Then the stipulation EFFECT > EFFORT makes sure that the wanted effect is reached with the minimal cognitive effort. Obviously, there are many questions left concerning the concrete content of the constraints EFFECT and 10 For more discussion of the role of frequencies in an evolutionary setting cf. Stalnaker (2006). The general conclusion is that the solution concept of weak bidirection can be seen as a rough first approximation to the more adequate solution concepts of evolutionary game theory that describe the results of language change. 11 In a somewhat different context, Cole (1975) calls it “lexicalization of contextual meaning”.

Pragmatics in Optimality Theory

47

EFFORT, the RT literature contains a number of specifications. These specifications typically have the character of linking constraints. It might be interesting to investigate recent OT models of pragmatics (see section 4) in the light of the general structure of RT – a task that goes beyond what can be done in the present paper. I have mentioned already that there is a relation between diachronic and synchronic systems, and I have introduced the term fossilization for describing the relevant transfer.12 Taken the existence of this transfer, it can be demonstrated that the three discussed variations on Grice are much closer related than the occasional polemics let us expect. 4. Some applications of bidirectional OT in pragmatics OT pragmatics has been used for describing a series of phenomena and observation in the domain of natural language interpretation. This section gives an overview of some of these applications without going into any technical or empirical details. ― ―

―

―

Centering theory. Beaver (2004) is using bidirectional OT as framework for the reformulation of centering models of pronoun resolution. Disambiguation. Gärtner (2004b, 2004a) analyses Icelandic objectshift and differential marking of (in-)definites in Tagalog addressing the issue of disambiguation and partial iconicity in natural language. Differential object marking: Aissen (2003), Nilsenova (2002), and Jäger and Zeevat (2002) discuss the relevant correlation between grammatical functions and semantic/pragmatic properties. Binding theory. Mattausch (2004a, 2004b) introduces the influential work of Levinson on the origin and typology of binding theory (summarized in Levinson 2000) and reformulates the different historical stages assumed by Levinson in bidirectional optimality theory. Mattausch’s work is of essential importance as one of the first in-depth studies showing the importance of the diachronic view for bidirectional OT. For early work on discourse anaphora in a bidirectional framework I refer to Buchwald, Schwartz, Seidl, and Smolensky (2002).

12 Concerning the debate whether certain pragmatic inferences are truly conversational or whether they have become lexically encoded the reader is referred to Cole (1975) and Potts (to appear).

48 ―

―

―

―

―

―

―

Reinhard Blutner

Pragmatics for propositional attitudes. Aloni (2001, 2005b) has argued that a number of seeming paradoxes emerging from logical analyses of attitudes and questions can be explained in terms of shifts in perspective over the universe of discourse. Shifts in perspective have a cost and, therefore, are generally avoided. However, under certain circumstances such shifts are required to comply with general principles of rational conversation, which, for example, disallow vacuous or inconsistent interpretations. Aloni’s work suggests a formulation of a perspective selection procedure in the framework of bidirectional OT. Discourse particles and presupposition. Zeevat (2002, 2004) treats discourse particles within an extended OT reconstruction of presupposition theory and concludes that more particles can be treated and the analysis becomes simpler if one starts from the fact that discourse particles are obligatory if the context of utterance and the current utterance stand in one of a number of special relations, like adversativity, additivity, contrast, etc. In another paper, Zeevat (2007a) provides a full solution to the projection problem for presuppositions. Jäger and Blutner (2000, 2003) suggest an bidirectional analysis of the different reading of German ‘wieder’ (again). Complex implicatures. Blutner (2007) gives an OT account of implicature projection and explains the relevance theoretic distinction between implicatures and explicatures in terms of a neo-Gricean framework. Interpretation of stress and focus. Several articles deal with a bidirectional perspective for stress on anaphoric pronouns and the interpretation of focus (Beaver 2004; de Hoop 2004; Hendriks 2004; Aloni, Butler, and Hindsill 2007) Marking and Interpretation of negation. Henriëtte de Swart (2004) provides a bi-directional OT approach to the syntax and pragmatics of negation and negative indefinites (see also, de Swart 2010). Scalar implicatures and exhaustification. Exhaustivity implicatures and also scalar implicatures depend on the issue under discussion which can be formalized using Groenendijk and Stokhof’s theory of question and answers. Combining this framework with that of bidirectional OT, Aloni (2005b, 2005a; de Swart 2010) explains several puzzles in this area. Permission sentences. A series of other articles deals with the interpretations of permission sentences and the analysis of the particular conditions which constitute a so-called free choice interpretation (Sæbø 2004; Aloni 2005a, 2005b; Blutner 2006).

Pragmatics in Optimality Theory ―

―

―

―

49

Stage level/individual level contrast: Maienborn (2004, 2005) argues against the popular view that the distinction between stage level predicates and individual level predicates rests on a fundamental cognitive division of the world that is reflected in the grammar. Instead, Maienborn proposes a pragmatic explanation of the distinction, and she gives, inter alias, a discourse-based account of Spanish ser/estar. Aspectual interpretation of the Dutch past tenses. Van Hout (2007) applied bidirectional reasoning about tense forms and their aspectual meanings. Lexical pragmatics: Lexical Pragmatics investigates the processes by which linguistically-specified (‘literal’) word meanings are modified in use. Well-studied examples include narrowing (e.g. drink used to mean ‘alcoholic drink’), approximation (e.g. square used to mean ‘squarish’) and metaphorical extension (e.g. battleaxe used to mean ‘frightening person’). Lexical Pragmatics can be formulated by using the formal instruments of OT-based pragmatics (Blutner 1998; Blutner et al. 2005). Prototypical applications include the pragmatics of dimensional adjectives (Blutner and Solstad 2000), the analysis of Dutch om/rond (Zwarts 2006), the pragmatics of negated antonyms (Blutner 2004; Krifka 2007b), gender opposition of animate nouns (Zwarts et al. 2009), the approximate interpretation of number words (Krifka 2007a), several examples of semantic change (Eckardt 2002). Language acquisition and learning: There are several studies that test the role of weak bidirectionality in developing interpretation and production preferences in connection with (in)definite NPs (deHoop and Kramer 2005/2006; van Hout, Harrigan, and Villiers 2010) and pronominal anaphors (Hendriks and Spenader 2005/2006; Hendriks, van Rijn, and Valkenier 2007; Mattausch and Gülzow 2007; Hendriks et al. 2010; van Rij, van Rijn, and Hendriks 2010). From a theoretical perspective, the problem of learning is investigated by Benz (2003a, 2003b).

5. Conclusions The error in many formulations of pragmatic inferences is that synchrony and diachrony are confused. OT pragmatics accounts both for the synchronic perspective – by formulating a localist, incremental model based on unidirectional optimization using an emerging system of linking constraints – and the diachronic perspective – using the solution concept of

50

Reinhard Blutner

weak bidirectionality which conforms to a strictly global view. The perspectives are connected by the idea of fossilization. Many patterns in language have been proposed to be directly or indirectly influenced by the conflict between multiple influences on output form. Within phonology for example, the notion that conflict between minimization of articulatory effort and maximization of perceptual distinctiveness has an influence on grammatical patterns has held currency at least since Baudouin de Courtenay (1895). Contemporary work grounding phonological patterns in optimization of conflicting influences on output form include work done within natural phonology (Stampe 1973), grounded phonology (Archangeli and Pulleyblank 1994), and optimality theory (Prince and Smolensky 1993/2004; Boersma 1998) to name but a few. Weak bidirectionality describes the interaction of these forces in an approximate but simply to understand way. However, for fully understanding the bidirectional game that leads to the resolution of the conflict between the opposite forces, evolutionary game theory provides a more adequate model (e.g. Van Rooy 2004). Relevance theory and Levinson’s theory of presumptive meaning account for the resolution of the conflict between effort minimization and effect maximization in different ways. In a certain sense, the crux of both approaches can be translated in OT pragmatics by making use of particular linking constraints. This translation makes the advantage of both approaches visible: both conform to the incremental, online character of natural language interpretation.13 I have argued that OT pragmatics has the potential to account both for the synchronic and the diachronic perspective in pragmatics and for the way both are related to each other. I further have pointed out that the concepts of fossilization can help to understand the idea of naturalization and (cultural) embodiment in the context of natural language interpretation. However, there are also important open questions regarding the status of 13 In discussing processing characteristics, incrementality and automaticity of processing have to be discriminated. Whereas automaticity of processing implicates the incremental character of processing the opposite is not true: incrementality does not implicate automatic processing. RT explains the incremental character of processing and has good reasons for assuming controlled processing in order to account for the processing of conversational implicatures. That’s different from Levinson’s account which assumes automatic processing for generalized conversational implicatures. It seems that RT is better justified on empirical grounds (cf. Noveck and Sperber 2005).

Pragmatics in Optimality Theory

51

fossilization. In a by now classical paper Cole (1975) considered the following example of a true conversational implicature, where a girl called Pamela upon being asked (2) might reply (3): (2)

How are you doing in your new position at San Andreas Fault University?

(3)

Well, I haven’t been fired yet.

Although the logical content of (3) is roughly that of the proposition that Pamela has not yet lost her job, more than that is implicated, namely that Pamela is not doing well. In this example, the implicature is really novel. There is no construction involved whose frequent use could lead to the fossilization phenomenon (Cole’s term is ‘lexicalization’). Hence, this implicature is different from many other cases where a certain amount of fossilization is plausible. The important question is how to discriminate between offline implicatures that are not fossilized and their fossilized counterparts. Where is the boundary between aspects of interpretations that are truly conversational and aspects which have become lexically (or syntactically) encoded? I think the former aspect of interpretation can require some real mind reading capacities, requires conscious reflections and proceeds offline. So far I can see, none of the discussed pragmatic theories has an interesting answer for this long-standing and intriguing question. References Aissen, Judith 2003 Differential coding, partial blocking, and bidirectional OT. In Proceedings of the 29th Annual Meeting of the Berkeley Linguistics Society, Pawel Nowak and Corey Yoquelet (eds.), 1–16. Berkeley: Berkeley Linguistics Society. Aloni, Maria 2001 Pragmatics for Propositional Attitudes. In Proceedings of the Thirteenth Amsterdam Colloquium, Robert van Rooy and Martin Stokhof (eds.), Amsterdam: University of Amsterdam. 2005a Expressing ignorance or indifference. Modal implicatures in BiOT. Paper presented at the Batumi 2005 symposium Batumi. 2005b A Formal Treatment of the Pragmatics of Questions and Attitudes. Linguistics and Philosophy 28 (5): 505–539. Aloni, Maria, Alastair Butler, and Darrin Hindsill

52

Reinhard Blutner 2007

Nuclear Accent in Bidirectional Optimality Theory. In Questions in Dynamic Semantics, Maria Aloni, Alastair Butler and Paul Dekker (eds.), 253–269. Amsterdam: Elsevier. Archangeli, Diana, and Douglas Pulleyblank 1994 Grounded Phonology. Cambridge MA: MIT Press. Atlas, Jay D. 2005 Logic, Meaning, and Conversation: Semantical Underdeterminacy, Implicature, and Their Interface. Oxford: Oxford University Press. Atlas, Jay D., and Stephen C. Levinson 1981 It-clefts, informativeness and logical form. In Radical Pragmatics, Peter Cole (ed.), 1–61. New York: Academic Press. Axelrod, Robert 1984 The evolution of co-operation. London: Penguin. Baudouin de Courtenay, Jan 1895 Versuch einer Theorie phonetischer Alternationen: Ein Capitel aus der Psychophonetik. Strassburg/Crakow: Trubner. Beaver, David 2004 The optimization of discourse anaphora. Linguistics and Philosophy 27: 3–56. Beaver, David, and Hanjung Lee 2004 Input-output mismatches in OT. In Optimality Theory and Pragmatics, Reinhard Blutner, and Henk Zeevat (eds.), 112–153. Houndmills, Basingstoke, Hampshire: Palgrave/MacMillan. Benz, Anton 2003a Partial blocking and associative learning. Unpublished manu-script. Berlin and Kolding. Available from http://www.antonbenz.de/Paper.html). 2003b Partial Blocking, associative learning, and the principle of weak optimality. In Proceedings of the Stockholm Workshop on Variation within Optimality Theory, Jennifer Spenader, Anders Eriksson, and Östen Dahl (eds.), 150–159. Stockholm. Blutner, Reinhard 1998 Lexical pragmatics. Journal of Semantics 15: 115–162. 2000 Some aspects of optimality in natural language interpretation. Journal of Semantics 17: 189–216. 2004 Pragmatics and the lexicon. In Handbook of Pragmatics, Laurence Horn and Gregory Ward (eds.), 488–514. Oxford: Blackwell. 2006 Embedded implicatures and optimality theoretic pragmatics. In A Festschrift for Kjell Johan Sæbø: In partial fulfilment of the requirements for the celebration of his 50th birthday, Torgim Solstad, Atle Grønn and Dag Haug (eds.), 11–29. Oslo.

Pragmatics in Optimality Theory 2007

53

Optimality Theoretic Pragmatics and the Explicature/Implicature Distinction. In Pragmatics, Noel Burton-Roberts (ed.), 67–89. Houndmills, Basingstoke, Hampshire: Palgrave/MacMillan. Blutner, Reinhard, Erik Borra, Tom Lentz, Jasper Uijlings, and Reinier Zevenhujzen 2002 Signalling games: Hoe evolutie optimale strategieen selecteert, Handelingen van de 24ste Nederlands-Vlaamse Filosofiedag. Amsterdam: University of Amsterdam. Blutner, Reinhard, Helen de Hoop, and Petra Hendriks 2005 Optimal Communication. Stanford: CSLI Publications. Blutner, Reinhard, and Torgim Solstad 2000 Dimensional designation: a case study in lexical pragmatics. In Studies in Optimality Theory, Reinhard Blutner, and Gerhard Jäger (eds.), 30–40. Potsdam: University of Potsdam. Blutner, Reinhard, and Henk Zeevat (eds.) 2004 Optimality Theory and Pragmatics. Houndmills, Basingstoke, Hampshire: Palgrave/Macmillan. Boersma, Paul 1998 Functional phonology. The Hague: Holland Academic Graphics. Buchwald, Adam, Oren Schwartz, O, Amanda Seidl, and Paul Smolensky 2002 Optimality Theory: Discourse Anaphora in a Bidirectional framework citeseer.ist.psu.edu/buchwald02optimality.html. Carston, Robyn 2002 Thoughts and Utterances: The Pragmatics of Explicit Communication. Oxford: Blackwell. 2005 Relevance Theory, Grice and the neo-Griceans: a response to Laurence Horn’s ‘Current issues in neo-Gricean pragmatics’. Intercultural Pragmatics 2: 303–319. Chierchia, Gennaro 2004 Scalar implicatures, polarity phenomena, and the syntax/pragmatics interface. In Structures and Beyond, Adriana Belletti (ed.), 39–103. Oxford: Oxford University Press. Cole, Peter 1975 The synchronic and diachronic status of conversational implicature. In Syntax and Semantics, Volume 3: Speech Acts, Peter Cole, and Jerry L. Morgan (eds.), 257–288. San Diego, Cal.: Academic Press, Inc. de Hoop, Helen 2004 On the interpretation of stressed pronouns. In Optimality Theory and Pragmatics, Reinhard Blutner, and Henk Zeevat (eds.), 25–41. Houndmills, Basingstoke, Hampshire: Palgrave/Macmillan.

54

Reinhard Blutner

de Hoop, Helen, and Henriette de Swart 2000 Temporal adjunct clauses in optimality theory. Rivista di Linguistica 12: 107–127. de Hoop, Helen, and Irene Kramer 2005/2006 Children’s Optimal Interpretations of Indefinite Subjects and Objects. Language Acquisition 13: 103–123. de Swart, Henriette 2004 Marking and Interpretation of negation: A bi-directional OT approach. In Negation, Tense and Clausal Architecture: Crosslinguistic Investigations, Raffaella Zanuttini, Héctor Campos, Elena Herburger, and Paul Portner (eds.), 199–218. Washington: Georgetown University Press. 2010 Expression and interpretation of negation: An OT typology. Dordrecht: Springer. Dekker, Paul, and Robert van Rooy 2000 Bi-directional optimality theory: An application of game theory. Journal of Semantics 17: 217–242. Eckardt, Regine 2002 Semantic Change in Grammaticalization. In Proceedings of the Sixth Annual Meeting of the Gesellschaft für Semantik, Sinn & Bedeutung VI, Graham Katz, Sabine Reinhard, and Philip Reuter (eds.), 53–67. Osnabrück: University of Osnabrück. Fodor, Jerry 1975 The Language of Thought. New York: Thomas Crowell. Gärtner, Hans-Martin 2004a On Object-Shift in Icelandic and Partial Iconicity. Lingua 114: 1235–1252. 2004b On the OT-status of ‘unambiguous encoding’. In Optimality Theory and Pragmatics, Reinhard Blutner, and Henk Zeevat (eds.), 154– 172. Houndmills, Basingstoke, Hampshire: Pal-grave/Macmillan. Gazdar, Gerald 1979 Pragmatics. New York: Academic Press. Geurts, Bart 2010 Quantity implicatures. Cambridge: Cambridge University Press. Grice, Herbert Paul 1975 Logic and conversation. In Syntax and Semantics, 3: Speech Acts, Peter Cole, and Jerry L. Morgan (eds.), 41–58. New York: Academic Press. Hendriks, Petra 2004 Optimization in Focus Identification, Optimality Theory and Pragmatics. Houndmills, Basingstoke, Hampshire: Palgrave/Macmillan. Hendriks, Petra, and Helen de Hoop 2001 Optimality theoretic semantics. Linguistics and Philosophy 24: 1–32.

Pragmatics in Optimality Theory

55

Hendriks, Petra, Helen de Hoop, Irene Kramer, Henriette de Swart, and Joost Zwarts 2010 Conflicts in Interpretation. London: Equinox Publishing. Hendriks, Petra, and Jennifer Spenader 2005/06 When production precedes comprehension: An optimization approach to the acquisition of pronouns. Language Acquisition 13 (4): 319–348. Hendriks, Petra, Hedderik van Rijn, and BeaValkenier 2007 Learning to reason about speakers’ alternatives in sentence comprehension: A computational account. Lingua 117: 1879–1896. Hobbs, Jerry, and Paul Martin 1987 Local Pragmatics, Proceedings, International Joint Conference on Artificial Intelligence, 520–523. Milan. Horn, Laurence 1984 Towards a new taxonomy of pragmatic inference: Q-based and Rbased implicature. In Meaning, form, and use in context: Linguistic applications, Deborah Schiffrin (ed.), 11–42. Washington: Georgetown University Press. 1989 A natural history of negation. Chicago: Chicago University Press. 2004 Implicature. In Handbook of Pragmatics, Laurence Horn, and Gregory Ward (eds.), 3–28. Oxford: Blackwell 2005 Current issues in neo-Gricean pragmatics. Intercultural Pragmatics (2): 191–204. Jäger, Gerhard 2002 Some notes on the formal properties of bidirectional optimality theory. Journal of Logic, Language and Information 11: 427–451. Jäger, Gerhard, and Reinhard Blutner 2000 Against lexical decomposition in syntax. In Proceedings of the Fifteenth Annual Conference, IATL 7, Adam Zachary Wyner (ed.), 113– 137. Haifa: University of Haifa. 2003 Competition and interpretation: The German adverb wieder ("again"). In Handbook of Adjuncts, Claudia Fabricius-Hansen, Ewald Lang, and Claudia Maienborn (eds.), 393–416. Berlin: Mouton de Gruyter. Jäger, Gerhard, and Henk Zeevat 2002 A reinterpretation of syntactic alignment. In Proceedings of the 3rd and 4th International Symposium on Language, Logic and Computation, Dick de Jongh, Henk Zeevat, and Maria Nilsenova (eds.), Amsterdam ILLC. Jaszczolt, Katarzyna 2010 Default Semantics. In The Oxford Handbook of Linguistic Analysis, Bernd Heine, and Heiko Narrog (eds.), 193–221. Oxford: Oxford University Press.

56

Reinhard Blutner

Krifka, Manfred 1995 The Semantics and Pragmatics of Polarity Items. Linguistic Analysis 25: 209–257. 2007a Approximate interpretation of number words: A case for strategic communication. In Cognitive foundations of interpretation, Gerlof Bouma, Irene Kramer, and Joost Zwarts (eds.), 111–126. Amsterdam: Koninklijke Nederlandse Akademie van Wetenschapen. 2007b Negated antonyms: Creating and filling the gap. In Presupposition and implicature in compositional semantics, Uli Sauerland, and Penka Stateva (eds.), 163–177. Houndmills, Basingstoke, Hampshire: Palgrave/Macmillan. Levinson, Stephen 2000 Presumptive meanings: The theory of generalized conversational implicature. Cambridge, Mass.: MIT Press. Maienborn, Claudia 2004 A pragmatic explanation of the stage level/individual level contrast in combination with locatives. In Proceedings of the Western Conference on Linguistics (WECOL), Vol. 15, Brian Agbayani, Vida Samiian, and Benjamin Tucker (eds.), 158–170. Fresno: CSU. 2005 A discourse-based account of Spanish ser/estar. Linguistics 43 (1): 155–180. Mattausch, Jason 2004a On the Optimization & Grammaticalization of Anaphora. Ph.D. Thesis, Humboldt University, Berlin. 2004b Optimality Theoretic Pragmatics and Binding Phenomena. In Optimality Theory and Pragmatics, Reinhard Blutner, and Henk Zeevat (eds.), 63–91. Houndmills, Basingstoke, Hampshire: Palgrave/Macmillan. Mattausch, Jason, and Gülzow, Insa 2007 A note on acquisition in frequency-based accounts of Binding Phenomena. In Frequency Effects in Language Acquisition: Defining the Limits of Frequency as an Explanatory Concept, Insa Gülzow, and Natalia Gagarina (eds.), 331–357. Berlin. New York: Mouton de Gruyter. McCawley, James D. 1978 Conversational implicature and the lexicon. In Syntax and Semantics 9: Pragmatics, Peter Cole (ed.), 245–259. New York: Academic Press. Nilsenova, Maria 2002 The Pragmatics of Differential Object Marking. In Proceedings of the ESSLLI ‘02 Student Session, Malvina Nissim (ed.), 221–232. Trento, Italy: University of Trento, revised version available from http://staff.science.uva.nl/~mnilseno/)

Pragmatics in Optimality Theory

57

Noveck, Ira, and Dan Sperber (eds.) 2005 Experimental Pragmatics. Houndmills, Basingstoke, Hampshire: Palgrave MacMillan. Potts, Christopher to appear Into the conventional-implicature dimension. Philosophy Compass. Prince, Alan, and Paul Smolensky 1993/2004 Optimality theory: Constraint interaction in generative grammar. Rutgers University and University of Colorado at Boulder: Technical Report RuCCSTR-2, available as ROA 537-0802. Revised version published by Blackwell, 2004. Sæbø, Kjell J. 2004 Optimal interpretations of permission sentences. In Proceedings of the 5th Tbilisi Symposium on Language, Logic and Computation, Dick de Jongh, and Paul Dekker (eds.), 137–144. Amsterdam and Tbilisi. Sauerland, Uli 2004 Scalar implicatures in complex sentences. Linguistics and Philosophy 27: 367–391. Smolensky, Paul, and Geraldine Legendre 2006 The Harmonic Mind: From neural computation to optimalitytheoretic grammar. Cambridge, Mass.: MIT Press. Soames, Scott 1982 How presuppositions are inherited: A solution to the projection problem. Linguistic Inquiry 13: 483–545. Sperber, Dan, Francesco Cara, and Vittorio Girotto 1995 Relevance theory explains the selection task. Cognition 57: 31–95. Sperber, Dan, and Deirdre Wilson 1986/95 Relevance: communication and cognition. Oxford: Basil Black-well. 1995 "Postface" to the second edition of Relevance. In Relevance: communication and cognition, Dan Sperber, and Deirdre Wilson, Oxford: Blackwell. Spohn, Wolfgang 1993 Wie kann die Theorie der Rationalität normativ und empirisch zugleich sein? In Ethik und Empirie. Zum Zusammenspiel von begrifflicher Analyse und erfahrungswissenschaftlicher Forschung in der Ethik, Lutz Eckensberger, and Ulrich Gähde (eds.), 151–196. Frankfurt a.M.: Suhrkamp. Stalnaker, Robert 2006 Saying and meaning, cheap talk and credibility. In Game Theory and Pragmatics, Anton Benz, Gerhard Jäger, and Robert Van Rooij (eds.), 82–101. Houndmills, Basingstoke, Hampshire: Palgrave MacMillan.

58

Reinhard Blutner

Stampe, David 1973 A Dissertation on Natural Phonology. Bloomington: Distributed 1979 by Indiana University Linguistics Club. van Hout, Angeliek 2007 Optimal and non-optimal interpretations in the acquisition of Dutch past tenses. In Proceedings of GALANA 2, Alyona Be-likova, Luisa Meroni, and Mari Umeda (eds.), 159–170. Som-erville, MA: Cascadilla Proceedings Project. van Hout, Angeliek, Kaitlyn Harrigan, and Jill de Villiers 2010 Asymmetries in the acquisition of definite and indefinite NPs. Lingua 120 (8): 1973–1990. van Rij, Jacolien, Hedderik van Rijn, and Petra Hendriks 2010 Cognitive architectures and language acquisition: A case study in pronoun comprehension. Journal of Child Language 37 (3): 731– 766. van Rooy, Robert 2004 Signalling games select Horn strategies. Linguistics and Philosophy 27: 493–527. Zeevat, Henk 2000 The asymmetry of optimality theoretic syntax and semantics. Journal of Semantics 17: 243–262. 2002 Explaining presupposition triggers. In Information Sharing, Kees van Deemter, and Roger Kibble (eds.), 61–87. Stanford: CSLI Publications. 2004 Presupposition Triggers, Context Markers, or Speech Act Markers. In Optimality Theory and Pragmatics, Reinhard Blutner, and Henk Zeevat (eds.), 91–112. Houndmills, Basingstoke, Hampshire: Palgrave/Macmillan. 2007a A Full Solution to the Projection Problem for Presuppositions. University of Amsterdam. 2007b Optimal Interpretation as an Alternative to Gricean Pragmatics. Unpublished manuscript, University of Amsterdam. Zwarts, Joost 2006 Om en rond: Een semantische vergelijking. Nederlandse Taalkunde 11: 101–123. Zwarts, Joost, Lotte Hogeweg, Sander Lestrade, and Andrej Malchukov 2009 Semantic markedness in gender oppositions, blocking and fossilization. Language Typology and Universals 62: 325–343.

Section II. Acquiring inferential abilities

Short introduction: Acquiring inferential abilities Seen from the vantage point of linguistics, the range of pragmatic phenomena resembles scraps on the edge of linguistic analysis. Deixis, discourse markers, referential choices, presupposition, implicature, speech acts, politeness: What these phenomena seem to have in common is that the regularities of their use cannot be fully explained by formal grammar theories alone. But seen from a broader aspect, they all have to do with the speaker’s meaning beyond that what is explicitly communicated. The general question is how the hearer makes sense of a speaker’s verbal, or more broadly, her communicative behaviour since doing so is crucial in order to fully take part in everyday communication where non-explicit verbal exchanges are common rather than the exception. Thus, the ability to draw inferences addressing the speaker’s meaning based on Gricean reasoning is the centrepiece of communication and the main topic of pragmatic research. In order to better understand the cognitive prerequisites of inferential communication one might want to have a look at how and at which age children make sense of a speaker’s verbal behaviour. Given that very young children only begin to take part in everyday verbal exchanges and that their cognitive abilities are, initially, rather limited, finding evidence for children’s success with pragmatic phenomena might tell us a lot about the nature of communication. The aim of this section is to summarize experimental evidence showing that young children do have the skills required to infer a speaker’s verbally conveyed meaning addressing different pragmatic phenomena. In a pragmatic approach to word learning, children infer the meaning of words by coordinating their attention and intentions towards an object with another person in a certain situation (Bruner 1983; Tomasello 2003). If then the interlocutor uses a label to direct the child’s attention to a particular aspect of the object or the situation, children try to make sense of that linguistic behaviour based on the assumption that the language used is somehow relevant to the situation (Sperber and Wilson 1986/1995). By doing so, they infer the speaker’s intention: did she want to name the object, label some part, refer to the action performed with the object and so on and so forth. Most word learning studies address the question whether children use previously acquired word knowledge about a familiar object to exclude this object as a referent of a novel word (Clark 1990; Diesen-

62

Cornelia Schulze

druck and Markson 2001). Success in those studies might be based on a combination of children’s own labeling-expectations towards a particular object and their ability to read the speaker’s intentions. Namely, children expect the speaker to use a familiar (conventional) label if she wants to refer to the familiar object. When they hear a novel word, children assume that the speaker most likely wanted to refer to another (unknown) object (Clark 1990). Lead by this assumption, children disambiguate the meaning of the novel word. In her paper, Susanne Grassmann will present a detailed description of the inference underlying the disambiguation phenomenon based on her experimental research with two-year-old children. She argues that the Modus Tollens described above must be seen as embedded in the disjunctive nature of the task, where the task itself evokes labeling-expectations on the part of the child. Those expectations in turn might or might not conflict with the label provided by the speaker. Thus, monitoring another person’s intentions and coordinating one’s own expectations and the other’s (verbal) behaviour are the major instances of the inferential skill required for word learning. But the importance of intention-reading is not only apparent when the child is confronted with novel words but also when she faces whole utterances. In Grice’s (1989) approach to communication, interlocutors work cooperatively in order to reach their goal of a successful communication. Doing so, they adhere to a set of principles that guide their conversational turns. The quantity maxim, for instance, states that a speaker will say as much as necessary (but not more) in order to provide the hearer with an informative utterance. Willingly saying less (or more) than necessary leads the hearer to re-interpret the speaker’s utterance under the assumption that there are other meanings that do not flout the quantity maxim. Some of those quantity implicatures are tied to certain linguistic expressions that form a scale and are therefore called scalar implicatures (Horn 1989; Levinson 2000). Especially those scalar implicatures tied to the quantifiers some and all are well-studied. Adults seem to make the necessary inferences quickly and relatively effortless (Breheny et.al. 2006; Geurts and Pouscoulous 2009). By contrast, Noveck (2001) found that 8- to 10-yearolds do not make the inference based on the use of a weak scalar term (such as ‘some’ and ‘might’) in an adult-like way. Yet, in a subsequent study, Pouscoulous et.al. (2007) showed that 4-to 5-year old children are able to assign meaning to scalar terms in a facilitated context and Katsos and Bishop (2011) found that children are sensitive to informativeness in more general terms.

Short introduction: Acquiring inferential abilities

63

Patrícia Amaral investigates another linguistic expression tied to a scale, namely the acquisition of almost; an adverb whose meaning depends on contextually-given elements of a scale (e.g., describing the path of an agent). She finds that 4.6-year-old children correctly understand almost, if they could identify the spatial, temporal or quantity-related path, the adverb refers to. Another Gricean maxim, namely the relevance maxim, is less wellstudied. Sperber and Wilson (1986/1995) claim, that relevance is a cognitive principle underlying the perception of linguistic as well as nonlinguistic behaviour. When confronted with an utterance, hearers try to relate the perceived content to the discourse-context (the question under discussion) in order to make sense of the speaker’s meaning. How much of the comprehension process is tied to language, Theory of Mind, and other social-cognitive factors and when and how children make use of this principle, is far from clear. Previous studies on relevance implicatures found that children must be at least 6 to 7 years old to derive the appropriate inference (Bucciarelli et.al. 2003; Loukusa et.al. 2007; Verbuk 2009, de Villiers et al. 2009). In her paper, Cornelia Schulze shows that younger children show the inferential abilities required to assess the relevance of a speaker linguistic behavior. Using a simpler task, she shows that children understand a speaker’s statement that indirectly conveys her desire to play with a certain object as a request to hand her that object; thus presenting evidence for three-year-old children’s abilities to make use of indirectly communicated content. Based on her previous findings in studies on scalar implicatures, presupposition, and discourse markers, that showed that children display inferential abilities relatively early in development, Nausicaa Pouscoulous presents an account of early pragmatic comprehension of metaphors. In the literature there is a widely held view that children do not understand metaphors until fairly late, maybe not before adolescence (see Nippold 1988/1998 and Winner 1988/1997 for reviews). But given that more recent studies on the above mentioned pragmatic phenomena found welldeveloped inferential abilities in quite young children, Nausicaa Pouscoulous aims to show that even 3-year-old children understand fully novel metaphors that correspond to their world knowledge and linguistic competence. Taken together, this section’s papers provide evidence that children display inferential abilities at a very young age. Given that pre-verbal children correctly infer the referential and social intention of a person pointing to an

64

Cornelia Schulze

object (Liebal et.al. 2009) and that they differentiate between ostensively and accidentally communicated cues to a person’s intentions (Behne et. al. 2005), it seems quite reasonable to assume that children use those early emerging skills in linguistic communication. Intention-reading as the cognitive prerequisite for understanding the speaker’s meaning beyond that what is explicitly communicated unifies the range of pragmatic phenomena presented in this section. Cornelia Schulze References Behne, Tanya, Malinda Carpenter, and Michael Tomasello 2005 One-year-olds comprehend the communicative intentions behind gestures in a hiding game. Developmental Science 8 (6): 492–499. Breheny, Richard, Napoleon Katsos, and John Williams 2006 Are generalised scalar implicatures generated by default? Cognition 100 (3): 434-463. Bruner, Jerome 1983 Child’s Talk. New York: Norton. Bucciarelli, Monica, Livia Colle, and Bruno G. Bara 2003 How children comprehend speech acts and communicative gestures. Journal of Pragmatics 35 (2): 207–241. Clark, Eve V. 1990 On the pragmatics of contrast. Journal of Child Language 17: 417431. de Villiers, Peter A., Jill G. de Villiers, D’Jaris Coles-White, and Laura Carpenter 2009 Acquisition of Relevance Implicatures in Typically-Developing Children and Children with Autism. In Proceedings of the 33th Annual Boston University Conference on Language Development, Jane Chandlee, Michelle Franchini, Sandy Lord, and Gudrun-Marion Rheiner (eds), 121–132. Somerville, Mass.: Cascadilla Press. Diesendruck, Gil, and Lori Markson 2001 Children’s avoidance of lexical overlap: A pragmatic account. Developmental Psychology 37 (5): 630-641. Geurts, Bart, and Nausicaa Pouscoulous 2009 Embedded implicatures?!? Semantics and Pragmatics 2 (4): 1-34. Grice, Herbert P. 1989 Studies in the way of words. Cambridge, MA: Harvard University Press.

Short introduction: Acquiring inferential abilities

65

Horn, Laurence R. 1989 A natural history of negation. Chicago: University of Chicago Press. Katsos, Napoleon, and Dorothy V.M. Bishop 2011 Pragmatic tolerance: implications for the acquisition of informativeness and implicature. Cognition 120 (1): 67-81. Levinson, Stephen C. 2000 Presumptive Meanings. The Theory of Generalized Conversational Implicature. MIT. Liebal, Kristin, Tanya Behne, Malinda Carpenter, and Michael Tomasello 2009 Infants use shared experience to interpret pointing gestures. Developmental Science 12 (2): 264–271. Loukusa, Soile, Eeva Leinonen, and Nuala Ryder 2007 Development of pragmatic language comprehension in Finnishspeaking children. First Language 27 (3): 279–296. Nippold, Marilyn A. 1988/98 Later Language Development: The School-Age and Adolescent Years. (2nd edition). Pro-Ed: Austin: Chapter 7, “Metaphors and Similes”, 89-103. Noveck, Ira A. 2001 When children are more logical than adults: Investigations of scalar implicature. Cognition 78 (2): 165-188. Pouscoulous, Nausicaa, Ira A. Noveck, Guy Politzer, and Anne Bastide 2007 Processing costs and implicature development. Language Acquisition 14 (4): 347-376. Sperber, Dan, and Deirdre Wilson 1986/95 Relevance: Communication and Cognition. Oxford: Blackwell. Tomasello, Michael 2003 Constructing a Language: A Usage-Based Theory of Language Acquisition. Cambridge, Mass.: Harvard University Press. Verbuk, Anna 2009 Acquisition of Relevance Implicatures and Modularity. In: Proceedings of the 33th Annual Boston University Conference on Language Development, Jane Chandlee, Michelle Franchini, Sandy Lord, and Gudrun-Marion Rheiner (eds), 575–586. Somerville, Mass.: Cascadilla Press. Winner, Ellen 1988/97 The Point of Words: Children’s Understanding of Metaphor and Irony. Harvard University Press.

Word learning by exclusion - pragmatics, logic and processing Susanne Grassmann

1. The Phenomenon Children learn the meaning of words in various situations by inferring the communicative intention of speakers (e.g., Akhtar and Tomasello 2000; Bloom 2000). A particularly interesting case of how children interpret novel words is that they exclude familiar objects as likely referents. This was first demonstrated by Vincent-Smith, Bricker and Bricker (1974). In that study, children were presented with a familiar object as well as with an unfamiliar one and heard an adult using a novel word in relation to those two objects. In response, children identified the unknown object as the one the speaker was talking about. But children exclude familiar objects as referents of novel words in situations other than such object choice scenarios. Specifically, children also tend to exclude familiar objects when there is no alternative novel object around to which a spoken novel word could refer (e.g., Markman and Wachtel 1988; Markman and Hansen 2009). A huge amount of research has been devoted to research and understand exclusion inferences in word learning. For example, it has been found that children draw exclusion inferences not only about the reference of nouns but also when interpreting novel adjectives (e.g., Carey and Bartlett 1978), verbs (e.g., Merriman et al. 1996) and utterances (Diesendruck and Markson 2001). Further, children as young as 16 months exclude familiar objects as referents of novel words (Graham, Poulin-Dubois, and Baker 1998; Halberda 2003; Markman, Wasow, and Hansen 2003; Mervis and Bertrand 1994) but older children, from about 4 years on, seem to do so more consistently and reliably (e.g., Merriman and Bowman 1989; Merriman and Schuster 1991). Correlations between the size of a child’s vocabulary and the strength of the exclusion effect have been demonstrated (Graham et al. 1998, Mervis and Bertrand 1994, but see Markman et al. 2003). The effect is smaller when the novel word sounds pretty much like the familiar object’s label (Merriman and Schuster 1991). Importantly, exclusion is also depended on who is speaking and which words that speaker can

68

Susanne Grassmann

or cannot be expected to use (Diesendruck 2005; Diesendruck, Carmel, and Markson 2010). In an attempt to explain the vast amount of empirical findings, a number of theoretical proposals have been put forward and spell out why children exclude familiar objects as referents of novel words. Generally, there are two camps of theoretical approaches, namely pragmatic accounts and constraint-based accounts. Constraint-based approaches assume that children’s word learning is cognitively constrained in a way to reduce processing load by limiting the potentially infinite number of possible meanings for each novel word.1 For example, Markman (1990; 1992; 1994) suggested that at the outset of lexical acquisition, children adhere to the following word learning biases: the Whole Object Assumption, the Taxonomic Principle and the Mutual Exclusivity Assumption. The Mutual Exclusivity Assumption accounts for the phenomenon at hand: Children’s exclusion of familiar objects as referents of novel words. It states that each object has only one label (because it belongs to only one of many mutually exclusive categories). Therefore, when children hear a novel word in the presence of an object for which they already know the category and a label, they exclude it as a likely referent. Pragmatic accounts on the other hand maintain that children, when interpreting novel words, draw inferences about the speaker’s communicative intention in the very situation a novel word is used. The general idea of pragmatic accounts to word learning is that interpreting the reference of a novel word relies on the same social cognitive skills and assumptions that children employ in any other situations when interpreting another person’s verbal and non-verbal behavior (Tomasello 2000; 2003). When interpreting 1 To illustrate the problem, proponents of constraints accounts appeal to Quine's (1960) gavagai example and claim it is analogous to word learning in children. Quine's problem, however, is specific to whether two words are translational equivalents (or synonymous). He describes (pp. 29 et seq) a linguist who wants to translate a foreign language and hears a speaker of that language say the word 'gavagai’ as a rabbit hops by. From his observations of the use of 'gavagai', the linguist can not conclude with any certainty that the word is translationally equivalent to a word in his own language (whether to rabbit, hopping, brown, or ears, etc.). This could apply to word learning when one assumes that children map novel words to pre-existing concepts. Quine's theory of meaning however is not compatible with such a view. Indeed, regarding children's word learning, Quine did not see any problem at all. Rather, he suggests: 'Each of us learns his language from other people, through the observable mouthing of words under conspicuously intersubjective circumstances' (Quine, 1960: 1).

Word learning by exclusion – pragmatics, logic and processing

69

novel words, children rely on an assumption that their cooperative interlocutors use language in a conventional manner and express their communicative intentions. Specifically, when interpreting novel words by exclusion, exclusion of familiar objects as referents of novel words results from inferences based on the assumption of Conventionality and Contrast (Clark 1987; 1990; 2009; Diesendruck and Markson 2001; Gathercole 1989; Grassmann and Tomasello 2010). Conventionality and Contrast are general pragmatic principles, not specific to language learning (maybe not even to language) and available to children very early2. The Principle of Conventionality states that speakers rely on certain (conventional) ways to express particular meanings. The principle of Contrast was formulated by Eve Clark, “[e]very two forms contrast in meaning” (Clark 1988: 318) or “Speakers assume that any difference in form signals a difference in meaning.” (Clark 2009: 133). Taken together, Conventionality and Contrast lead to the conclusion that, “when speakers choose an expression, they do so because they mean something that they would not mean by choosing some alternative expression” (Clark 1990: 417; see also Clark and Clark 1979). Starting out from the pragmatic account to word learning by exclusion I will now specifying the structure of the inference underlying exclusion in word learning. This analysis indicates that exclusion in word learning might not rely solely on general default assumptions such as Clark’s Principle of Conventionality and Contrast but on more specific expectations what conventional language use in a certain situation would be. 2. The logical structure of exclusion inferences in word learning As mentioned above, children exclude familiar objects as referents of novel words in two rather different situations: First, when a familiar object as well as a novel object are available, the familiar object is excluded and the novel identified as the referent of the novel word (I call that the two-way choice task3). Second, children exclude familiar objects as referents of 2 It so far remains open where these general principles come from and how they develop. Thus, the main difference to constraints approaches is the level of specificity at which a priori principles are assumed to operate. 3 In the literature one also finds the term disambiguation effect. However this is an unfortunate term as it is not clear in which sense to speak of an ambiguity in that situation.

70

Susanne Grassmann

novel words when only one single familiar object is present at the novel word’s utterance (I call this the one-object situation). This paper is by no means the first one that attempts a logical description of the inference underlying word learning by exclusion. A number of researchers have described and tried specifying that inference (e.g., Diesendruck and Markson 2001; Gathercole 1989; Halberda 2003; 2006). However, these descriptions focus on the two-way choice task only and, as will be shown below, they are not fully adequate. In this paper I will therefore try to provide a more comprehensive picture of the inferences involved in the two-way choice task and the inferences drawn in the oneobject situation will be discussed as well. 2.1. The inferences in the two-way choice task Diesendruck and Markson (2001: 631) describe the inferences in the twoway choice task in the following way: (1) i. The objects in front of us are part of our common ground and thus relevant for our interaction.4 ii. If she had wanted me to pick the car, she would have asked for the car. iii. She said blicket. Conclusion: She is not referring to the car (but to the other object). In this sequence of inferential steps, premise (ii) corresponds to Clark’s Principle of Conventionality and Contrast (Clark 1988, 1990). While providing an insightful characterization of the inferences drawn in the twoway choice task, a problem with Diesendruck and Markson’s (2001) description of the inference is that the conclusion actually consists of two parts and that only the first part logically follows from the premises. However, it is the second part (in brackets) that is required in order to resolve the reference between the two choices. Thus, on logical grounds, another premise is required: Namely, that the speaker is referring to one of the two available objects. Since this premise is missing in Diesendruck and 4 I omit this assumption in what follows since it forms the basis of making any inference at all. This assumption is essentially one of cooperativeness in communication (and interaction) (cf., Grice 1975; Clark 1996) or the subtly different assumption of relevance (Sperber and Wilson 1995).

Word learning by exclusion – pragmatics, logic and processing

71

Markson’s schema, strictly speaking, the second part of the conclusion is not supported, the inference thus invalid. Consequently, conventionality and contrast not sufficient in order to identify the novel object as the referent of the novel word in the two-way choice task. But, conventionality and contrast lead an addressee to (initially) reject the familiar object as a likely referent of the speaker’s communicative act (the first part of the conclusion in Diesendruck and Markson’s account). In Gathercole’s (1989) description of the inference, the overall disjunctive structure of the two-way choice task is more apparent (cf. pp. 693694): (2) i. The adult is referring to one of the two objects on the table by the novel word blicket. ii. There is a car and an object I don’t know. iii. The adult must not have meant the car or she would have asked for the car. Conclusion: She is referring to the other object. Markman (1994), and Halberda (2003, 2006) were even more specific than Gathercole about the disjunctive inference in word learning by exclusion. Markman (1994) states, “>t@he logic of this problem is of the form of a disjunctive syllogism: ‘The novel word must refer to either object A or to object B. It is not A (by mutual exclusivity), therefore, it must be B’” (p. 221). Put in a schema the disjunctive syllogism reads: (3) i. She is referring5 to EITHER the car OR the novel object. ii. She is NOT referring to the car. Conclusion: She is referring to the novel object. Halberda (2003; 2006) however highlights, that the disjunction in the two-way choice task is not as straight forward as it might seem, since the second premise (‘NOT a’ or ‘she is NOT referring to the car’) is usually not made explicit, but only implicit in the use of a novel word. Where does 5 Reference is a communicative act by which the speaker (intends to) direct the addressee's attention to a certain entity. Whenever I use the word refer it can be understood as shortcut for 'intends to direct the addressee's attention'.

72

Susanne Grassmann

that second premise come from? According to constrains accounts, Mutual Exclusivity provides the second premise for the disjunction (eg. Markman 1994). According to pragmatic accounts however this premise is what follows from conventionality and contrast (see (1)). Thus, the second premise would be the conclusion of an embedded inference that has the form of a Modus Tollens, or denial of the consequent. A Modus Tollens is another kind of exclusion inference (as compared to disjunction) and it takes the following schematic form: (4) i. IF a THEN y ii. NOT y Conclusion: NOT a Applied to the example: (4a) IF she wants to refer to the car, THEN she would say ‘car’. She did NOT say car. Conclusion: She is NOT referring to the car. Thus, overall the inference in the two-way choice task can be described as an embedded exclusion inference: A Modus Tollens embedded in a disjunctive syllogism. Before spelling out that overall inference in schema (5), I want to highlight that a closer look at Gathercole’s (1989) analysis of the inference in word learning by exclusion reveals a similar general structure to the one suggested here: Her premise (2ii) has two parts and the first part (NOT the car) could be seen as the conclusion drawn from the second part of that very premise and the observation that the speaker said blicket (entailed in premise (2i)). Putting each premise in the inference in a single line the overall inference in the two-way choice task thus reads: (5) i. She is referring to EITHER the car OR the novel object. ii. IF she wants to refer to the car, THEN she would say car. iii. She said blicket iv. ‘Blicket’ ≠ ‘car’ Conclusion i . Thus, she did NOT say car. Conclusion ii. She is NOT referring to the car. Overall Conclusion: She is referring to the novel object.

Word learning by exclusion – pragmatics, logic and processing

73

The Modus Tollens is the core of the exclusion effect in word learning. It is descriptively equivalent to Clark’s Principle of Conventionality and Contrast. But the basis for the Modus Tollens is the expectation of a certain verbal behavior as formulated in a conditional (‘If this is about the car then she would say car’). This is what the general expectation of conventional language use comes down to in a concrete situation. To summarize, the inference in word learning by exclusion in the twoway choice task is based on a specific naming expectation for the familiar object. An apparent mismatch between the expected name and the heard novel word implies that the speaker is unlikely to refer to the familiar object. In order to resolve and interpret the speaker’s actual communicative intention a further inference is necessary, which in the two-way choice task is framed by the structure of the situation: identifying which one of several objects the speaker wants. 2.2. The inferences in the one-object situation The situation in which children exclude familiar objects as referents of novel words in the absence of any alternative object is quite different. There are no descriptions of this inference in the literature. However, based on the previous schemas the inference seems quite straightforward: The inference should be the same as in (5), but without the overall disjunction: (6) i. IF she wants to refer to the car, THEN she would say car. ii. She said blicket iii. ‘Blicket’ ≠ ‘car’ Conclusion i. Thus, she did NOT say car. Conclusion ii. She is NOT referring to the car. According to this inference, the familiar object should always be excluded from being the target of the speaker’s communicative act in the oneobject situation. The problem is that (to my knowledge) there is not much support for such strict exclusion. Indeed many studies found that children readily accept novel words for familiar objects (Markman and Liittschwager 1991; Saylor et al. 2002; 2004). Thus, in the one-object situation children do not seem to arrive at the conclusion suggested by the inference as described in (6).

74

Susanne Grassmann

So what is the inference then? The key to understanding the inference is that the principle of contrast does not work at the level of reference but on the level or meaning and contrastive forms might as well share a referent, as for example terms of different taxonomic levels (Clark,1987; 1990; 2007; 2009).6 Accordingly, the conditional in premise (6i) would have to be reformulated. Note, that therefore the conclusion of the Modus Tollens also changes in order to remain valid: (7) i. IF this communicative act is about the car and about the car in the standard way, THEN she would say ‘car’. ii. She did NOT say car. Conclusion: This is NOT about the car OR it is not in the standard way about the car.7 Unfortunately, the conclusion in that schema is not particularly helpful when one seeks to interpret what a speaker intended to convey when using a novel word. However, just as the structure of the situation provides information to the interpretation of the speaker’s communicative intention in the two-way choice task (object choice and thus an overall disjunctive structure), similarly the structure of the one-object situation might provide such clues as well. Specifically, the pragmatics of the situation do not allow doubts about the relation of the speaker’s utterance to the familiar object since the speaker might be looking at or pointing to the familiar object or even manipulating it. This might enable the interpreter to perform a disjunctive syllogism on the conclusion of the inference in (7): (8) i. IF this communicative act is about the car and about the car in the normal way, THEN she would say ‘car’. ii. She did NOT say car. Conclusion I: This is NOT about the car OR it is not in the normal way about the car. iii. This IS about the car. 6 Applying Contrast at the level of reference as I did so far in the analysis of the inference in the two-way choice task may be allowed by the pragmatics of the situation in this task. In the two-way choice task children have to determine the reference of a novel word. 7 Actually one should add here 'or she is talking about a part or a property or etc.'

Word learning by exclusion – pragmatics, logic and processing

75

Overall Conclusion: this is about the car but in a special way. The inferences in (5) and (8) seem to be adequate descriptions of the exclusion inferences in word learning in the two-way choice task and the one-object situation. However, it is peculiar that the expectation (the conditional involving what linguistic behavior to expect), which is fundamental to both inferences (and supposedly ‘provided’ by the Principle of Conventionality and Contrast), differs between one inference to the other. The conditional in (8i) more adequately relates to Clark’s Principle of Conventionality and Contrast than the one in (5ii). Therefore, the naming expectation in the two-way choice task should maybe also be: ‘IF this is about the car and about the car in the normal way, THEN she would say ‘car’.’ However, formulated in this way the inference in the two-way choice task would not work and would not lead to the intended overall conclusion. This leads to interesting predictions of how a situation would be resolved in which a speaker uses a term that might be applied to the familiar object (but in an non-standard way). For an empirical investigation of this see Grassmann, Schreiner, and Tomasello (in prep). Thus, strictly speaking, Clark’s Principle of Conventionality and Contrast is not applied in the twoway choice task. Therefore, rather than pulling on these general principle as the origin of expecting certain expressions for particular entities, I suggest that the expectations about the speaker’s language use is by default (and thus in the one-object task as well as in the two-object choice task): ‘IF she wants to refer to the car, THEN she would say car.’ Nevertheless, the inference employing this conditional as a premise (6) does not provide a comprehensive characterization for the inference in the one-object situation. How then is the inference resolved? I suggest that the initial exclusion inference does not resolve the speaker’s intended meaning. Rather, the conclusion of the inference (as in (6)) is in conflict with what the speaker otherwise indicates he intends talking about (e.g., by pointing). I suggest that this conflict leads to a re-analysis of the whole situation and eventually a (new) interpretation for the speaker’s utterance – maybe involving an inference as described in (8). To summarize, I suggest that Conventionality and Contrast are not the basis of word learning by exclusion. First, because the expectation of conventional language use is not sufficient for drawing any inference. And second, because the actual form of the principle of Contrast is not helpful in resolving reference in the two-way choice task. Instead I suggest, that the addressee has certain specific naming expectations for the familiar object(s) available in a situation. When the speaker then uses a novel (un-

76

Susanne Grassmann

expected) word the addressee’s expectation is violated and that conflict between the expected and the heard expression triggers a re-analysis process which eventually leads to the appropriate interpretation of the novel word – either as referring to an available novel object or a particular perspective to the familiar object. Importantly, this re-interpretation process is not constrained (as it would be under the assumption of the Principle of Conventionality and Contrast in (8)). That means, it is in principle possible for the addressee to be unable to generate an alternative conceptualization for the speaker’s unexpected utterance of a novel word, which would lead to the interpretation that the novel word serves the same communicative function as the initially expected familiar one. 3. Hypotheses about mental processing exclusion In this section I will briefly describe how this expectation and the subsequent exclusion inference in the two-way choice task and the one-object situation potentially can be translated into cognitive processes of spreading activation, competition and inhibition. The general idea is that for object selection and meaning-ascription activation of conceptual information is decisive. That is, in the one-object situation, the most activated conceptual representation is interpreted as the meaning of the novel word and in the two-way choice task the object corresponding to the concept holding highest activation levels will be chosen. When interpreting an utterance, the addressee’s task is to come up with a conceptualization that fits the speaker’s communicative intention. Interpretation is easy when the speaker uses an expression that is known to the addressee and thus directly interpretable: The intended conceptualization is automatically activated in the addressee’s mind by spread of activation from the perceived phonological word-form to the corresponding concept. However, interpreting the speaker’s utterance is somewhat more difficult when a novel expression is used. 3.1. Processing in the two-way choice task Several sources of information and their parallel processing play a role when interpreting a novel word when a novel and a familiar object are available. First, perception of objects activates corresponding conceptual representations and subsequently corresponding phonological forms (e.g., Dell 1986; Friedrich and Friederici 2004; 2005; 2006; Mani and Plunkett

Word learning by exclusion – pragmatics, logic and processing

77

2010; Levelt 1999; Meyer and Damian 2007; Morsella and Miozzo 2002). The degree of activation of a given concept depends on a number of factors such as familiarity or discourse salience (e.g., Chafe 1994; Gundel, Hedberg, and Zacharsky 1993; Metzing and Brennan 2003; Smith et al. 1999)8. In this sense, there may be less activation of conceptual representations related to novel objects. The phonological word form the speaker is using also leads to activation in the mental lexicon (e.g., Norris, Cutler, and McQueen 2006). When seeing a familiar object and hearing a novel word, thus, the addressee is in a situation where phonological word forms are activated based on different sources (the speaker’s utterance versus spreading activation from object perception) and in conflict with each other. Presumably, this conflict triggers competition, which in turn results in (relative) inhibition of the familiar object label and the corresponding concept. Such an inhibition process requires something like a pragmatic monitor that boosts the activation stemming from another person’s utterance as to indicate that this information/activation is more important than activation created internally (bottom up). Such a pragmatic monitor could be switched off if a speaker behaves in unpredictable and unreliable ways leaving the mental lexicon with competition and inhibition processes that are not weighted. Thus, random object choice is expected (and found by Diesendruck et al. 2010). However, for object selection, supposedly not activation level of phonological word forms but activation levels of conceptual representations are decisive. Then, how might differential phonological activation lead to differential conceptual representation? If one assumes that not only activation can spread in the mental lexicon but also inhibition, then inhibition of the familiar word form could spread to the conceptual level and result in a situation where the (ad hoc) concept related to the novel object remains as the most activated one and thus the novel object is chosen.9 3.2. Processing in the one-object situation Processing in the one-object situation is identical with the processing in the two-way choice task to the point that a concept and a representation of a 8 Activation of related concepts will be ignored for the moment. 9 Of course, the mere visual perception of the two objects always re-activates the corresponding concepts – so that the concept corresponding to the familiar object is always a bit activated (and thus always a side-option to be chosen as the referent of the novel word).

78

Susanne Grassmann

verbal expression for the familiar object at hand are activated. Further, a representation of the speaker’s utterance is active and again, provides most relevant information as to the interpretation of the speaker’s utterance. As in processing of the two-way choice task, competition between the representations corresponding to the heard novel word and the seen familiar object leads to inhibition of the representations corresponding to the familiar object. Additionally, the speaker’s non-verbal indication of his communicative intention provides information just as relevant as his utterance. Specifically, it indicates that the conceptual representation corresponding to the familiar object is highly relevant as well (potentially again due to the work of the pragmatic monitor). In the one-object situation, the apparent conflict between the activation and the inhibition of the representation corresponding to the familiar object cannot be resolved by canceling out one source of incoming information/activation, since both are provided by the speaker. Therefore, the addressee needs to start out anew and re-analyze and re-conceptualize the whole situation including the object at hand. Potentially, such re-analysis leads to a new conceptualization of the familiar object or it may draw the addressee’s attention to a so far unnoticed aspect of the object – which then may serve as an interpretation for the novel word. Importantly, pragmatic directions provided along with a novel expression facilitate the reinterpretation process (see Clark and Wong 2002; Saylor et al. 2002, 2004). How do the proposed mental processes fit empirics? The hypothetical cognitive processes underlying word learning by exclusion as outlined above draw pretty much on automatic activation of concepts and spreading activation to corresponding word-forms.10 The outlined processes adequately describe various findings as to under when, under which conditions, and to which extend children exclude familiar objects as referents of novel words. For example, the proposed possesses can explain why in the two-way choice task atypical familiar objects and novel words that are phonologically similar to the familiar object’s label lead to reduced exclusion effects (Merriman and Schuster 1991): Atypical objects are likely to lead to problems with how to categorize them. Hence a corresponding linguistic form may not be, or not sufficiently, activated to elicit further processing. This in turn would lead to no mismatch between word forms (the automatically activated label for the familiar object and the heard novel 10 This spread of activation can be seen as the translation of "expecting" a certain word into a cognitive process.

Word learning by exclusion – pragmatics, logic and processing

79

word) and as a result, competition and inhibition of the representations corresponding to the familiar object would not occur. Similarly, the tiny differences between the automatically activated familiar object label and a similar-sounding novel word may be not sufficient for triggering competition and inhibition (unless some training sharpens attention, cf. Merriman and Marazita 1995). The processes described predict that automatic activation of a verbal expression for a familiar object is required to trigger the whole process. A certain level of word knowledge might be required for that – presumably productive mastery of a certain object label, since when one wants to name an object, then the cognitive processes involved seem pretty similar to expecting another person to use a certain word: Based on visual perception, the phonological form gets activated and further processed for articulation. In a recent study, we found that indeed the ability to name a familiar object is a prerequisite to excluding it as the referent of a novel word in the twoway choice task (Grassmann, Schulze, and Tomasello, in prep). Thus, for words that are not yet in a child’s productive vocabulary the link between the concept representation and the corresponding word form is not strong enough to lead to sufficient activation of the word – neither to produce it nor to elicit the processes necessary for word learning by exclusion. An interesting question might be whether and how related exclusion phenomena would be explainable? An interesting case are exclusion inferences about facts (Diesendruck and Markson 2001; Scofield and Behrend 2007). Note, that facts actually are expressed as complex linguistic expressions (e.g., ‘this is the one I got for my birthday’). Thus, assigning a fact to an object is highly similar, if not identical, to establishing a conceptual pact about a referring expression to be used for a certain object (cf., Brennan and Clark 1996). The mental processes as described above are not specific for words but in line exclusion when a speaker would use any expression other than the one the addressee would expect. Just the basis of the naming expectation might be different in word learning and in conceptual pact/ fact exclusion situations. Thus, interpreting a communicative act in which a conceptual pact is broken (or a novel fact voiced) might turn out an instance of exclusion inference. Interestingly, in such cases a speaker uses an expression that the addressee did not expect but that is perfectly applicable to the object in question. That is, within a particular situation two people might implicitly agree on clipper (or ‘the one I got for birthday’) to refer to a stapler and adhere to this agreement, with all the discussed cognitive consequences when that agreement is broken. Thus, when a speaker suddenly says stapler to the

80

Susanne Grassmann

stapler, according to the processing outlined about, this should lead to competition, inhibition and subsequent exclusion of the stapler previously called clipper and then, just as in the one-object exclusion situation a reinterpretation of the situation is required which might be exactly why adults and children take a while to resolve the reference of terms that break conceptual pacts (Matthews, Lieven, and Tomasello 2010; Metzing and Brennan 2003). Stretch the argument even further, even non-literal language such as metaphors might be processed in a way similar to exclusion. Of course the cognitive processes in this case might differ from those in word learning by exclusion, as not only the non-linguistic context (a certain potential referent) but also the expression itself leads to certain expectations that are violated in non-literal language use. A very interesting and surprising example for exclusion inferences based on unexpected expressions is children’s exclusion of familiar objects as the referents of totally appropriate super-ordinate terms. In a recent study (Grassmann, Schreiner, and Tomasello, in prep) we presented children with one novel and one familiar exemplar of a familiar super-ordinate category (e.g., an apple and a star-fruit). The children’s task was to hand over the fruit. Results show that children chose the novel exemplar above chance level. That is, they excluded the familiar exemplar. This is in line with what would be predicted be the processes described above in the following way: The perception of a familiar object leads to high activation of its basic-level term. Upon hearing the super-ordinate term, a conflict with the term that the speaker uttered leads to competition and inhibition of the concept corresponding to the apple and eventually to the familiar object’s exclusion. Finally, does the outlined processing fit with what we know about the neuro-cognitive processes related to word learning by exclusion? Friedrich and Friederici (2004; 2005; 2006) demonstrated that word learning by exclusion is related to at least two EEG components: the (early) MMN, indicating a (negative) phonological priming and the N400, indicating increased processing in order to integrate certain information semantically. The cognitive processes as outlined above pull on conflicts and reinterpretation that are commonly linked to these EEG components. One could hypothesize that the MMN would relate to the mismatch between an expected verbal expression and the speaker’s actual utterance. And the N400 would be related to the process of re-interpretation and actual reference resolution that is computed at the conceptual level. Further EEGstudies seem to be well suited to address the question whether and when

Word learning by exclusion – pragmatics, logic and processing

81

inappropriate expectations are canceled out. Specifically, it might be that pragmatically inappropriate naming expectations do not cause negative phonological priming (thus no MMN) – thus are canceled out early. Alternatively, later EEG components that are related to semantic and pragmatic integration such as N400 and P600 may differ for violations of pragmatically appropriate versus inappropriate expectations. In sum, the general structure of the mental processing that translates the logical form of word learning by exclusion fits well with empirical data. However, in its current state, no full account for person- and language specificity of exclusion can be made (cf., Au and Glusman 1990; Diesendruck 2005). Future specifications of the cognitive processes also need to comprise an in-depth discussion of current knowledge on linguistic processing and conceptualization. 4. Building expectations in word learning and beyond In the previous section I already mentioned quite a few phenomena beyond word learning by exclusion, which could be described as exclusion inferences. All these examples share the characteristic that the inferences are based on expectations of certain verbal expressions which conflict with a speaker’s actual utterance. In this section I will give some more examples for exclusion inferences – moving beyond the interpretation of others’ verbal behavior. Expectations comprise a special kind of knowledge. Namely, it is knowledge that people use in order to predict the near future. Regularities in others’ behavior play an important role in creating such expectations. That is, after an event has occurred sufficiently often under similar conditions, the relation between the event and the behavior becomes predictive. Conventions are an important and reliable kind of behavioral regularities (Clark 1996; Lewis 1969): One can predict an actor’s goal under the assumption that he or she acts in a conventional way. But one can also predict what someone would do when one knows what the agent intends to achieve by a certain action. This bi-directionality is especially clear in linguistic conventions. A linguistic convention is a pair of a linguistic form and a meaning (communicative intention). In word learning by exclusion, linguistic conventions are used to predict a form, given a meaning: ‘If she intends to refer to the car, then she would say car.’ However, exclusion inferences would also work based on expectations that use conventions the other way around: Predicting an interpretation of an utterance based on its

82

Susanne Grassmann

form. Processing of garden-path sentences (e.g., ‘The horse raced past the barn fell.’) might be an example for such an exclusion inference. However, as noted above, in this section I want to go beyond linguistic examples for exclusion inferences. Furthermore, the following examples for exclusion inferences are not limited to cases in which conventions are violated. Expectations about other peoples’ behavior can have different origins. Non-conventional (behavioral) regularities also provide a basis for expectations and therefore for exclusion inferences. For example: Children know, and thus expect, that people get excited and talk about things that are ‘new’. Therefore, when there is nothing new in a situation and yet a speaker appears excited, children tend to search for ‘an explanation’. That is, children exclude the things established in common ground as the cause for the speaker’s excitement (Moll, Koring, Carpenter, and Tomasello 2006; Grassmann et al. 2009). Similarly, children may learn (and thus expect) that people look at the things they are naming. And so, in Baldwin’s (1991) seminal bucket-study, children might actually have drawn an exclusion inference: Since the speaker is not looking at the object in front of me, I infer that she is not naming it. Indeed, Baldwin’s findings are consistent with this suggestion: One of her findings is that 16-month-olds excluded the object at which the speaker did not look while using a novel word. But only by 18-months children resolved the actual reference of the novel word. Further, children’s selective imitation of a novel action (Gergely et al. 2002) might also be due to exclusion: Children may have learnt that (at least when sitting at the table) people tend to manipulate objects by using their hands. When this expectation is violated, children seek to interpret what is special about the actor’s goal to behave in such a peculiar way and imitate the peculiar action. When however the expectation that the interlocutor would use her hands is not felicitous (since her hands are occupied) there appears no conflict between an expected and an observed action and thus children are not tuned into a re-interpretation and learning process. That is, when something unexpected happens, the organism starts analyzing reasons for the mismatch and may learn something new from it, for example, a new word, or a new way of manipulating objects. To stretch the argument even further, experimental paradigms such as the violation of expectancy paradigm, and various habituation-dishabituation paradigms may actually measure processes related to exclusion inferences. These paradigms make use of the fact that after a certain number of repetitions of a particular something children notice and are surprised when something else happpens (e.g., Baillargeon, Spelke, and Wasserman

Word learning by exclusion – pragmatics, logic and processing

83

1985; Spelke et al. 1992; Werker et al. 1998). The children’s surprise about the change usually causes enhanced attention. It seems likely that during exposure to a certain stimulus of interest during an experiment (or even preceding to it) children build ‘expectations’, which are violated later in the experiments. I suggest that the children’s enhanced attention is related to engaging in re-interpretation as described in the model.11 5. Summary In this paper I spelled out a detailed logical description of the inference that appears to underlie children’s interpretations of novel words by exclusion. It was shown that earlier descriptions of the inference in word learning by exclusion were not adequate. Overall the inference required for word learning by exclusion can be described as a Modus Tollens embedded in a Disjunctive Syllogism. However, I do not claim that children consciously draw such a complex inference.12 I outlined some ideas how competition, inhibition and spreading activation might account for the complex inference. Based on these analyses, I suggest that the general Principle of Conventionality and Contrast is insufficient for explaining exclusion inferences in word learning. Instead, I suggest that mismatches between expected and observed utterances trigger re-interpretation processes that (usually) lead to exclusion effects. A variety of linguistic and non-linguistic examples for how people interpret others’ (verbal) behavior were put forward in a claim that they are structurally analogous to inferences in word learning by exclusion. The core of the approach to exclusion inferences as it is presented in the current paper is that the relevant inferences rely on specific expectations and their violation.

11 Bernstein (1969; 1979) found that detecting mismatches between what is expected vs. observed increases learning success. And to use Piaget's terms (Piaget 1985; Piaget and Inhelder 1969): Assimilation would be the process of building expectations. Accommodation occurs based on exclusion. 12 Even adults have problems with explicit Modus Tollens inferences (cf. Byrne & Tasso 1999).

84

Susanne Grassmann

Acknowledgements I thank Cornelia Schulze and Mike Tomasello for a great time discussing, designing and running experiments that are the empirical basis for the thoughts outlined in this paper. Many thanks also to Eve Clark, Dorothé Salomo, Eileen Graf, and Daniel Schmerse for their challenging but helpful comments on earlier drafts that shaped and sharpened my thinking. References Akhtar, Nameera, and Michael Tomasello 2000 The social nature of words and word learning. In Becoming a word learner, Roberta M. Golinkoff, K. Hirsh-Pasek, L. Bloom, L. Smith, A. Woodward, and N. Akhtar (eds.), 115–135. New York: Oxford University Press Au, Terry K., and Mariana Glusman 1990 The principle of mutual exclusivity in word learning: to honor or not to honor. Child Development 61: 1474–1490. Baldwin, Dare A. 1991 Infants’ contribution to the achievement of joint reference. Child Development 62: 875–890. Baldwin, Dare A., and Michael Tomasello 1998 Word learning: A window on early pragmatic understanding. In The Proceedings of the 29th annual Child Language Research Forum, Eve V. Clark (eds.), 3–23. Stanford, CA: CSLI Publications. Baillargeon Renee, Elizabeth S. Spelke, and Stanley Wasserman 1985 Object permanence in five-month-old infants. Cognition 20: 191208. Bernstein, Alvin 1969 To what does the orienting response respond? Psychophysiology 6: 338–350. 1979 The orienting response as novelty and significance detector: Reply to O’Gorman. Psychophysiology 16: 263–273. Brennan, Susan E., and Herbert H. Clark 1996 Conceptual pacts and lexical choice in conversation. Journal of Experimental Psychology: Learning, Memory, and Cognition 22: 1482–1493. Byrne, Ruth, and Alessandra Tasso 1999 Deductive reasoning with factual, possible, and counterfactual conditionals. Memory and Cognition 27: 726–740.

Word learning by exclusion – pragmatics, logic and processing

85

Bloom, Paul 2000 How children learn the meanings of words. Cambridge: The MIT Press. Carey, Susan, and Elsa Bartlett 1978 Acquiring a single word. Papers and Reports on Child Language Development [Stanford University] 15: 17–29. Chafe, Wallace 1994 Discourse, Consciousness, and Time. Chicago, IL: Chicago University Press. Clark, Eve V. 1987 The principle of contrast: A constraint on language acquisition. In Mechanisms of language acquisition, Brian MacWhinney (ed.), 133. Hillsdale, NJ: Lawrence Erlbaum Associates. 1988 On the logic of contrast. Journal of Child Language 15: 317–335. 1990 On the pragmatics of contrast. Journal of Child Language 17: 417– 431. 2007 Young children’s uptake of new words in conversation. Language in Society 36: 157–182. 2009 Lexical meaning. In Cambridge Handbook of Child Language, Edith Bavin (ed.), 283-299. Cambridge: Cambridge University Press. Clark, Eve V., and Herbert H. Clark 1979 When nouns surface as verbs. Language 55: 767–811. Clark, Eve V., and James B. Grossman 1998 Pragmatic directions and children’s word learning. Journal of Child Language 25: 1–18. Clark, Eve V., and Andrew D.-W. Wong 2002 Pragmatic directions about language use: words and word meanings, Language in Society 31: 181–212. Clark, Herbert H. 1996 Using language. Cambridge: Cambridge University Press. Dell, Gary S. 1986 A spreading-activation theory of retrieval in sentence production. Psychological Review 93: 283–321. Diesendruck, Gil 2005 The principles of conventionality and contrast in word learning: an empirical examination. Developmental Psychology 41: 451–463. Diesendruck, Gil, and Lori Markson 2001 Children’s avoidance of lexical overlap: a pragmatic account. Developmental Psychology 37: 630–641. Fennell, Christopher T., and Janet Werker 2003 Early word learners’ ability to access phonetic detail in well-known words. Language & Speech 46: 245-264.

86

Susanne Grassmann

Gathercole, Virginia C. 1989 Contrast: A semantic constraint? Journal of Child Language 16: 685–702. Gergely, György, Harold Bekkering, and Ildikó Király 2002 Rational imitation in preverbal infants. Nature 415: 755. Graham, Susan A., Diane Poulin-Dubois, and Rachel K. Baker 1998 Infants’ disambiguation of novel object words. First Language 18: 149–164. Grassmann, Susanne, Cornelia Schulze, and Michael Tomasello (in prep) Young children exclude familiar objects as referents of novel words only if they can produce the familiar label. Grassmann, Susanne, Marén Stracke, and Michael Tomasello 2009 Two-year-olds exclude novel objects as potential referents of novel words based on pragmatics. Cognition 112: 488–493. Grassmann, Susanne, and Michael Tomasello 2010 Young children follow pointing over words in interpreting acts of reference. Developmental Science 13: 252–263. Grice, Herbert Paul 1957 Meaning. The Philosophical Review 66: 377–388. Gundel, J. K., Nancy Hedberg, and Ron Zacharaski 1993 Cognitive Status and the form of referring expressions. Language 692: 274–307. Halberda, Justin 2003 The development of a word-learning strategy. Cognition 87: B23– B34. 2006 Is this a dax which I see before me? Use of the logical argument disjunctive syllogism supports word-learning in children and adults. Cognitive Psychology 53: 310–344. Levelt, Willem J. M. 1999 Models of word production. Trends in Cognitive Sciences 3: 223– 232. Lewis, David K. 1969 Convention: A philosophical study. Cambridge, MA: Harvard University Press. Markman, Ellen M. 1989 Categorization and naming in children: Problems of induction. Cambridge: The MIT Press. 1990 Constraints children place on word meanings. Cognitive Science 14: 57–77. 1992 Constraints on word learning: Speculations about their nature, origins and domain specificity. In Modularity and Constraints in language and cognition. The Minnesota Symposia on child psychology, Megan

Word learning by exclusion – pragmatics, logic and processing

87

R. Gunnar, and Michael P. Maratsos (eds.), 59–101. Hillsdale, NJ: Lawrence Erlbaum. 1994 Constraints on word meaning in early language acquisition. Lingua 92: 199–227. Markman, Ellen M., and Gwyn F. Wachtel 1988 Children’s use of mutual exclusivity to constrain the meanings of words. Cognitive Psychology 20: 121–157. Markman, Ellen M., Judith L. Wasow, and Mikkel B. Hansen 2003 Use of the mutual exclusivity assumption by young word learners. Cognitive Psychology 47: 241–275. Matthews, Danielle, Elena Lieven, and Michael Tomasello 2010 What’s in a manner of speaking? Children’s sensitivity to partnerspecific referential precedents, Developmental Psychology 46: 749– 760. Meltzoff, Andrew N. 1995 Understanding the intentions of others: Re-enactment of intended acts by 18-month-old children. Developmental Psychology 31: 838– 850. Merriman, William E., and Laura L. Bowman 1989 The mutual exclusivity bias in children’s word learning. Monographs of the Society of Research in Child Development 54: 1–55. Merriman, William E., and John M. Marazita 1995 The effect of hearing similar-sounding words on young 2-year-olds’ disambiguation of novel noun reference. Developmental Psychology 31: 973–984. Merriman, William E., and Joneen M. Schuster 1991 Young children’s disambiguation of object name reference. Child Development 62: 1288–1301. Merriman, William E., Julia A. Evey-Burkey, John M. Marazita, and Lorna H. Jarvis 1996 Young two-year-olds’ tendency to map novel verbs onto novel actions. Journal of Experimental Child Psychology 63: 466–498. Mervis, Carolyn B., and Jacquelyn Bertrand 1994 Acquisition of the novel name-nameless category (N3C) principle. Child Development 65: 1646–1662. Metzing, Charles, and Susan E. Brennan 2003 When conceptual pacts are broken: Partnerspecific effects on the comprehension of referring expressions. Journal of Memory and Language 49: 201–213. Moll, Henrike, Cornelia Koring, Malinda Carpenter, and Michael Tomasello 2006 Infants determine other’s focus of attention by pragmatics and exclusion. Journal of Cognition and Development 7: 411–430.

88

Susanne Grassmann

Morsella,, Ezequiel, and Michele Miozzo 2002 Evidence for a Cascade Model of Lexical Access in Speech Production. Journal of Experimental Psychology: Learning, Memory, and Cognition 28, 555-563. Norris Dennis, Anne Cutler, and James McQueen 2006 Phonological and conceptual activation in speech comprehension. Cognitive Psychology 53: 146–193. Piaget, Jean 1985 Equilibration of cognitive structures. University of Chicago Press. Piaget, Jean, and Bärbel Inhelder 1969 The psychology of the child. New York: Basic Books. Quine, Willard V. O. 1960 Word and Object. Cambridge, MA: MIT Press. Saylor, Megan M., and Mark A. Sabbagh 2004 Different kinds of information affect word learning in the preschool years: The case of part-term learning. Child Development 75: 395– 408. Saylor, Megan M., Mark A. Sabbagh, and Dare A. Baldwin 2002 Children use whole-part juxtaposition as a pragmatic cue to word meaning. Developmental Psychology 38: 993–1003. Scofield, Jason, and Douglas A. Behrend 2007 Two-year-olds differentially disambiguate novel words and facts. Journal of Child Language 34: 875–889. Smith, Peg H., Jeanette Whitmore, Wendelyn Shore, Christopher Robinson, and Wallace E. Dixon 1999 Infants’ responses to objects representing levels of word knowledge and categorical information. Infant Behavior and Development 22: 511–526. Sperber, Dan, and Deirdre Wilson 1995 Relevance: Communication and cognition (2nd edn). Oxford: Blackwell. Spelke Elizabeth S., Karen Breinlinger, Janet Macomber, and Kristin Jacobson 1992 Origins of knowledge’. Psychological Review 99: 605-632. Swingley, Daniel, and Richard N. Aslin 2002 Lexical neighborhoods and word-form representations of 14-montholds. Psychological Science 13: 480-484. Tomasello, Michael 2000 The social-pragmatic theory of word learning. Pragmatics 10: 40114. 2003 Constructing a language: a usage-based theory of language acquisition. Harvard, MA: Harvard University Press.

Word learning by exclusion – pragmatics, logic and processing

89

Vincent-Smith, Lisbeth, Diane Bricker, and William Bricker 1974 Acquisition of receptive vocabulary in the toddler-age child. Child Development 45: 189–193. Werker, Janet F., Leslie B. Cohen, Valerie L. Lloyd, Marianella Casasola, and Christine L. Stager 1998 Acqusition of word-object associations by 14-month-old infants. Developmental Psychology 34: 1289–1309.

Children’s knowledge of scales in the acquisition of almost Patrícia Amaral

1. Introduction When children produce sentences like I’m almost 5 to indicate their age, one may wonder whether they fully understand the implications1 conveyed by this expression. The meaning of almost is quite complex: being ‘almost 5 years old’ means not being five (yet) and being an age close to five (e.g. 4 years and ten months). In order to compute these implications, it is necessary to be familiar with the numeric order and to identify an element in that order that is less than five and close to five. Does a 4 year-old understand the components of the meaning of almost? And does the child grasp the meaning of almost when it occurs with expressions other than number words? In fact, almost may modify expressions of different syntactic categories: verbs, adjectives, prepositions, other adverbs. Do children understand its meaning in all cases and if yes, do they rely on the same abstract semantic representation? From a broader perspective, this question matters for word learning and the acquisition of cross-categorial modifiers; how do children learn the semantic value of these modifiers from occurrences across different categories? Previous studies in the field of acquisition claim that children aged 4–6 have difficulties with interpreting almost, and only reach adult-like interpretation of the adverb by age 7 or 8. According to these studies, young children understand the ‘proximity’ component of the meaning of the adverb (in the example above, being ‘close to five’) but not its ‘scalar’ component (i.e. if you are ‘almost 5 years old’ you are ‘less than’ 5). This study focuses on children’s interpretation of almost as a modifier of number words (e.g. almost five), directional prepositional phrases (e.g. almost to the lilypad) and equative comparative phrases (e.g. almost as far 1 I use the term implication in order to remain neutral about the classification of the components of the meaning of almost (i.e. whether they are logical entailments, implicatures, etc.). For a thorough discussion of this issue, see Horn (2002) and Amaral (2010).

92

Patrícia Amaral

as). It addresses the following research questions: (i) Do children interpret almost differently from adults across the board? (ii) How do children acquire the different components of the meaning of almost? (iii) Are there any differences in the acquisition of the cross-categorial instances of modification by almost? This paper is organized as follows. In section 2 I describe the meaning of almost. In section 3, I summarize the findings of previous studies. Section 4 describes the aims and methods of this experimental study. Section 5 presents the results of the experiment and section 6 provides a discussion of the results. Section 7 provides concluding remarks. 2. The meaning of almost 2.1. Scalarity of almost Building on scalar accounts of the meaning of almost (Hitzeman 1992, Morzycki 2001, Amaral and Del Prete 2010, among others) I assume that the interpretation of almost differs from the interpretation of other terms denoting approximation (e.g. about, approximately) in that it requires that the expression modified by the adverb, interpreted in context, provides a scale. A scale is a set of alternatives to the value of the modified expression with a linear order on it. A linearly ordered set is defined by a relation that is transitive, antisymmetric, reflexive, and connected (Landman 1991), for which I use the notation ‘here (pause) \Sure. What is it?

Here B’s level tone seems to indicate that B does not want A to speak until B indicates the end of her utterance with a falling tone accompanying the word sure. However, a stereotypical exchange involving a level tone is the one involving a police officer interviewing a citizen represented in (18): (18) Officer: Mr. Smith: Officer: Mr. Smith: Officer:

>Name John Smith >Occupation Plumber >Address

160

Billy Clark

Here, the level tone indicates a lack of engagement but clearly not the end of an utterance. It seems, then, that the analysis of level tones also needs to be modified. There are at least two possibilities which we might consider. One is that the level tone indicates something about a unit which may be larger than an individual utterance or turn, along the lines of (19): (19) LEVEL:

the speaker has not completed the contribution she is currently making to the interaction

This is problematic for at least two reasons. First, it raises the question of what a ‘contribution’ might be. We know it is not a turn in the exchange or an individual utterance but would need to find another way to define it or to explain how a hearer manages to resolve this in context. Another is that this is not in line with the nature of the analyses proposed for other tones, particularly given the thoughts behind the modification of the analysis of rises suggested above. This might suggest an alternative along the lines of (20): (20) LEVEL:

no indication as to whether the speaker intends the hearer to recover a representation of a thought entertained as a description of a state of affairs or an interpretation of someone else’s thought.

This could be simplified even more drastically as in (21) or even, most drastically of all, as in (22): (21) LEVEL:

no indication of the nature of any thoughts the speaker is entertaining.

(22) LEVEL:

0

Of course, (22) could be understood as a way of saying that level tones do not encode anything. If we take this route, we have in effect removed level tones from the set of prosodic forms with an encoded meaning. To produce an utterance with a level tone is to encourage the hearer to infer mental states such as lack of engagement because of an absence of encoded meaning. The claim then would be that to produce a level tone is similar to speaking with no eye contact or with a marked lack of nonverbal behaviour. It is a methodological challenge for future research to consider how we might be able to determine whether this tone encodes something or is

Procedures and prosody: understanding weak communication

161

simply a form used to avoid producing a tone which does have an encoded meaning. There is, of course, a big difference between not encoding anything and encoding the absence of something. Whichever of the analyses we choose, a level tone is taken here to amount simply to a way of avoiding any indication along the lines usually suggested by rises or falls. Clearly, we have moved again to a much weaker analysis than the one originally proposed. 4. Weak communication and weak encoded meanings This section explores some of the questions raised by this approach, in particular those associated with the notion that both the proposed encoded meanings for the tones and what is communicated by utterances associated with them are very weak. It has long been assumed within relevance theory that any communicated assumption can be communicated more or less strongly as well as that what is semantically encoded can be very weak, amounting in some cases to what might be thought of as a subtle hint or a gentle push in a particular direction. Nevertheless, the analyses considered here are particularly weak, raising questions about the role they might play in an explanatory account of the interpretation of particular utterances and how to investigate the claims empirically. This section begins by considering some of the ways in which what is communicated can vary in strength; within relevance theory, this depends on how much evidence a communicator provides for a particular assumption. It then considers some of the ways in which encoded meanings vary in strength; this, by contrast, depends on how specific and precise an encoded meaning is. Within relevance theory, all encoded meanings can be understood as relatively weak, a view which has been described as the ‘underdeterminacy thesis’ (see, for example, discussion by Carston 2002: 19-30). Finally, it considers the ways in which the current proposal sees the encoded meanings of tones as particularly weak and reasons for thinking that the weakness of these analyses appropriately reflects how they contribute to utterance interpretation. 4.1. Strong and weak communication One development in pragmatics since the work of Grice (1975; 1989) is that we have become much more aware of how vague, open-ended and indeterminate what is linguistically or non-linguistically communicated can

162

Billy Clark

be. Grice did make reference to indeterminacy in discussing implicatures, saying that: Since to calculate a conversational implicature is to calculate what has to be supposed in order to preserve the supposition that the Cooperative Principle is being observed, and since there may be various possible specific explanations, a list of which may be open, the conversational implicatum in such cases will be disjunction of such specific explanations; and if the list of these is open, the implicatum will have just the kind of indeterminacy that many actual implicata do in fact seem to possess. (Grice 1989: 39-40)

However, his account of individual examples tended to assume far less indeterminacy than would be recognised by most current pragmatic theories with regard to both what is said and what is implicated. To take just one example, here is the first example Grice used to illustrate cases where ‘no maxim is violated’: (23) A: B:

I am out of petrol There is a garage round the corner.

Grice’s summary of what B communicated here is as follows: (24) What is said: What is implicated:

There is a garage round the corner. ‘The garage is, or at least may be, open, etc.’ (based on Grice 1975: 51)

Of course, Grice was aware that his suggestions were programmatic and anticipated that others would develop them more fully. He might not, however, have anticipated how much more indeterminacy would be assumed in subsequent work, both on the explicit and on the implicit side. One of the key assumptions of relevance theory, and of other approaches, is that pragmatic processes are involved in recovering what Grice would have termed ‘what is said’ as well in recovering ‘what is implicated’. This was first discussed from a relevance-theoretic point of view by Wilson and Sperber (1981), before many of the details of a relevance-theoretic approach had been developed. More recent work has expanded understanding of a wide range of ways in which both what is explicit and what is implicit need to be inferentially developed. Given this, it follows that there is scope for vagueness and indeterminacy with regard to both what is implicated and what is explicated. However, there has been greater focus on the nature of indeterminacy on the implicit side.

Procedures and prosody: understanding weak communication

163

4.2. Weak implicit communication Discussions of indeterminacy on the implicit side have focused largely on the notion of ‘weak implicature’ (see, for example, discussion in Wilson and Sperber 2004) and the relevance of this notion in explaining literary or ‘poetic’ effects (see Pilkington 2000 for discussion). An assumption made here is that a communicator can make it more or less manifest that she intends to communicate certain assumptions. This can be illustrated by almost any exchange which gives rise to implicatures. (25) A: B:

Have you seen Wuthering Heights? I hate period dramas.

Clearly, B here is implicating strongly that she has not seen Wuthering Heights (A would not have understood the utterance without understanding this) and, almost if not quite as strongly, that she will not be going to see it. She gives considerable evidence that she will not be going to see other period dramas such as Jane Eyre (at the time of writing, movie adaptations of both books are showing in cinemas in the UK). She has given less strong evidence that she will not be going to see other films which are less definitely categorisable as period dramas, e.g. films set in more recent times such as Tinker Tailor Soldier Spy or The Iron Lady. There are also assumptions A might make based on what he thinks is likely to be true of people who don’t like period dramas, e.g. he might decide that they won’t like to read historical fiction. Finally, he might derive some conclusions which B is not responsible for at all, e.g. he might decide that she is a snob. This last conclusion is not an implicature at all since A will not think B intended to communicate it. We might summarise what we have said so far in the following list:

164

Billy Clark

Explicature: B hates period dramas Strong implicature: B hates or would hate Wuthering Heights Other implicatures arranged in descending order of approximate strength: B has not seen Wuthering Heights B will not be going to see Wuthering Heights B will not be going to see Jane Eyre B will not be going to see Tinker Tailor Soldier Spy B will not be going to see The Iron Lady B does not like to read historical fiction Implication (inferred conclusion not intentionally communicated by B): B is a snob Relevance theory also assumes that some utterances give rise not to a small set of relatively strong implicatures but to a much wider range of relatively weak ones. The most well-known example is Flaubert’s comment on the writer Leconte de Lisle: (26) Son encre est pale [‘His ink is pale’] Sperber and Wilson (1995: 237) point out that there is no one clear candidate implicature communicated by this utterance. Rather, this utterance provides some evidence for a wide range of relatively weak conclusions such as the following: (27) Possible implicated conclusions derived from Flaubert’s utterance: a. b.

Leconte de Lisle’s writing will not last. Leconte de Lisle’s writing is passionless.

Procedures and prosody: understanding weak communication

c. d. e.

165

Leconte de Lisle’s writing is emotionless. Leconte de Lisle’s writing is underwhelming. Leconte de Lisle does not put serious energy into his writing.

Utterances which achieve relevance in this way are perceived to be highly creative and, in some cases, literary. Effects such as these are described as ‘poetic effects’, most extensively discussed within a relevancetheoretic approach by Pilkington (2000). While this way of thinking about the varying strength of implicatures shares something with Grice’s notion that indeterminacy might be accounted for by reference to a ‘disjunction’ of possible implicatures, the notion that an utterance could convey such a wide range of potential implicatures goes far beyond the indeterminacy envisaged by Grice. The notion that implicatures come with varying amounts of strength is not present in Grice’s approach. Relevance theory also diverges from Grice in not assuming that communicators necessarily have a clear idea in mind of the set of assumptions which they intend to communicate. The following three examples illustrate this: (28) A: B:

What do you think of Wendy? She’s an iphone

(29) A: B:

What did you think of Avatar? Meh

(30) (A and B are standing on a hilltop looking at the view) A: What are you thinking? B: Breathes in ostentatiously, spreads her arms and sighs (29) resembles Flaubert’s comment in (27), providing evidence for a range of assumptions about Wendy, e.g. that she is very popular, that she is different from other acquaintances, that she is popular for fashionable reasons rather than for more serious reasons, and so on. It is quite possible not only that A is not sure what exactly B intended but also that B might not have a clear representation of the precise set of assumptions she intends to communicate. B’s utterance in (30) uses an expression which encodes a kind of indeterminacy and whose usage has increased in British English in recent years. Someone who uses this form represents their opinion as not clearcut, neither wholly positive nor wholly negative. Some users will de-

166

Billy Clark

scribe its meaning as something fairly definite such as ‘indifference’ while others would suggest that it does not even commit the speaker to this. In (31) it is not clear what exactly B’s nonverbal behaviour is intended to communicate. It conveys an impression rather than a set of clear propositions. 4.3. Weak explicit communication So the relevance-theoretic understanding of weak communication is different from Grice’s approach in general and with regard to implicature. It is also different with regard to explicature. It follows from the assumption that pragmatic inference is involved in recovering explicatures that it is also possible for these to be indeterminate. Carston (2002) explores the pragmatic inferences involved in recovering explicatures in detail. Here are three examples she presents early in the book (Carston 2002: 22) with bracketed questions indicating some of the inferential processes which need to be followed in order to understand what has been explicated: (31) Paracetamol is better (32) It’s the same (33) She’s leaving (Carston 2002: 22)

[than what?] [as what?] [from where?]

As Carston makes clear in subsequent discussion, the required inferences go further than this. Other required inferences include an assumption about what paracetamol is better for in (32), what kind of sameness is involved in (33), what kind of leaving she is undertaking in (34). We will also need to decide on referents for pronouns, assumptions about when or under what circumstances the propositions are held to be true, and so on. In some cases, the hearer will be able to recover one or more clear explicatures after inferentially developing the linguistically encoded meanings. In other cases, it will not be possible to arrive at clear conclusions. There are several explicatures of B’s utterance in (30) which the speaker gives some evidence for. This would also be true of (29) on some accounts (see discussion in 4.2 below). It is also possible for poetic effects to be derived because an utterance gives rise to a wide range of possible explicatures leaving the addressee unsure about which ones are intended. Doğan (1992) discusses a range of examples in short poems and epigrams where it is clear that the author intended the referents of pronouns and other aspects

Procedures and prosody: understanding weak communication

167

of explicatures to be indeterminate. Of course, it is quite common in poetry, song and some prose fiction for there to be uncertainty about who or what is being referred to. Here is an example from a famous haiku: (34) Sick and feverish Glimpse of cherry blossoms Still shivering. (Ryunosuke Akutagawa 1892-1927) While we might focus here on the range of weak implicatures which this gives rise to (as Sperber and Wilson 2008 do in discussing a haiku by Basho), we might also point out that there is uncertainty about aspects of the explicit content here. We do not know, and have no way of knowing, who is sick and feverish, glimpsing the cherry blossoms, and so on. We can, however, follow different inferential paths based on different assumptions about this. More recently, Carston (see, for example, discussion in Carston 1997, 2002) has explored ways in which explicatures can be indeterminate, including in her work on ‘ad hoc concept construction’. On this view, the hearer of an utterance such as (36) has to construct an ad hoc concept {BULLDOZER} which he then uses in understanding this utterance: (35) A: B:

What’s your new teacher like? He’s a bulldozer.

On this view, part of the hearer’s task in understanding B’s utterance is to construct a BULLDOZER concept which he can then use in deriving conclusions based on B’s utterance. This may be something which is stubborn, does not respond to input from others, moves ahead relentlessly, and so on. The new concept will not have non-human qualities associated with bulldozers and some of the qualities it does have are, of course, not usually associated with bulldozers. These have been termed ‘emergent features’in the literature (for discussion within a relevance-theoretic framework, see Vega Moreno 2007; Wilson and Carston 2006, 2008). A key thing to notice is that hearers might not identify one definite ad hoc BULLDOZER concept here. As Carston (2002: 358) puts it, in discussing a slightly different example: ‘There may be quite a range of subtly different concepts licensed by an utterance of, for instance, “Robert is a bulldozer” . . . No specific one is

168

Billy Clark

strongly communicated and the hearer’s construction of any one of them is good enough for the communication to have succeeded’.

From this discussion, then, it is clear that relevance theory, and other pragmatic theories, envisage indeterminacy as a property of some explicatures as well as of some implicatures. The proposal for the meanings of tones developed here assumes that they often give rise to indeterminacy about explicatures and implicatures. Following the modification to the analysis of rising tones suggested above, there is greater indeterminacy than previously assumed since it may be unclear which particular explicatures or implicatures are the ones which (a rise indicates) are not being entertained by the speaker. While it may be clear what a tone encodes about one or more thoughts entertained by the speaker, it may not be clear which of a wide range of explicatures and implicatures are the ones entertained in this way. 4.4. Strong and weak encoded meanings Given the underdeterminacy thesis, it is clear, of course, that what is encoded by the linguistic expressions represented in any utterance overall must be weak in a slightly different sense. The relative strength of a communicated assumption follows from how manifest it is that the communicator intends to communicate that assumption. The relative strength of an encoded meaning follows from how specific the encoded meaning is. There is, of course, considerable inferential work to be done in fleshing out what is encoded by expressions such as (32)-(34) above, and the same is true of almost any utterance. It does not necessarily follow from this, however, that what is encoded by each linguistic component is weak. It is possible that each lexical, syntactic and prosodic unit contributes something fairly definite but that what they amount to overall leaves considerable space for inference. This is not assumed by relevance theory. In earlier work, the assumption seemed to be that the encoded meanings of expressions with conceptual meanings, contribute conceptual material to mental representations, were relatively precise while encoded procedural meanings, which guide inferential processes, were less so. The classic source for this distinction is Blakemore’s (1987) book in which she considered expressions which encoded ‘constraints’ on the search for relevance (for more recent discussion, see Wilson and Sperber 1993; Blakemore 2002; Escandell-Vidal, Leonetti and Ahern 2011). These constraints were later

Procedures and prosody: understanding weak communication

169

described in terms of ‘procedural meaning’ and the range of such expressions and the kinds of procedures they encode have been extended considerably since. Blakemore’s initial work looked at expressions such as but, so and after all: (36) a. b. c. d.

Jess has read Pride and Prejudice. She’s a teenager. Jess has read Pride and Prejudice. [But] she’s a teenager. Jess has read Pride and Prejudice. [So] she’s a teenager Jess has read Pride and Prejudice. [After all] he’s a teenager.

The hearer of (37a) could infer any of a wide range of connections between the two clauses and the main thing which will influence the particular direction chosen will be assumptions about teenagers (female teenagers in particular) and how likely they are to have read Pride and Prejudice. The expressions but, so and after all encourage the hearer to go down a particular inferential path. As Blakemore (1987) put it, they ‘constrain’ the inferences made by the hearer, reducing the number of inferences he might make. Recent work has extended both the range of expressions taken to encode procedural meanings and the range of things which they can contribute to. Wilson and Sperber (1993) survey the possibilities, showing that procedural expressions can contribute to the proposition expressed (the lowest-level explicature of an utterance) as it does in (38), higher-level explicatures as interrogative syntax does in (39) and implicatures as illustrated by the examples in (37b-d) above, repeated here as (40a-c): (37) A: B:

Why don’t you heat that up in the microwave? It’s broken.

Proposition expressed: [The microwave] is broken (38) A: B:

I wish I could heat this up in the microwave. Why can’t you? Is it broken?

170

Billy Clark

Higher-level explicature: [B would find it relevant to know that] the microwave is broken. (39) a.

Jess has read Pride and Prejudice. But she’s a teenager.

Implicatures: You would not expect a teenager to have read Pride and Prejudice. Jess’s behaviour contradicts a common assumption about teenagers. b.

Jess has read Pride and Prejudice. So she’s a teenager.

Implicature: The fact that Jess has read Pride and Prejudice provides evidence that she is a teenager. c. Jess has read Pride and Prejudice. After all, she’s a teenager. Implicature: The fact that Jess is a teenager provides evidence that she has read Pride and Prejudice. As mentioned above, one of the key assumptions about procedural meanings is that the procedures they activate are ‘sub-personal’ (in the sense of Dennett 1969) and not directly accessible to introspection. They are by definition hard to pin down and can not be characterised straightforwardly in non-procedural terms. They are, then, perhaps the most obvious example of linguistically encoded meanings which can be understood as ‘weak’. Sperber and Wilson (1998) suggest another kind of encoded meaning. They propose that some words encode ‘pro-concepts’. The examples they give at first are my, have, near and long. They suggest that these share with pronouns the property that they require inference to arrive at the content

Procedures and prosody: understanding weak communication

171

they contribute to propositions expressed, but that they differ in having some conceptual content. They point out, for example, that there are a number of ways of understanding my book, e.g. as ‘the book I am thinking of’ or ‘the book I wrote’, and that we need to make an inference about what counts as ‘near’ when understanding an utterance such as (41): (40) The school is near my house. So far, then, we have established three kinds of meaning: conceptual meaning where lexical items encode conceptual representations; procedural meaning where expressions activate procedures; and ‘pro-conceptual’ meanings, where expressions both encode conceptual representations and activate procedures. It might seem, then, that there is a contrast in strength of encoding between expressions which encode concepts and those which encode procedures or pro-concepts. However, Sperber and Wilson (1998: 185) suggest that ‘quite commonly, all words behave as if they encoded pro-concepts: that is, whether or not a word encodes a full concept, the concept it is used to convey in a given utterance has to be contextually worked out’. This insight lies behind the notion of ‘ad hoc concept construction’ developed by Carston in a number of publications (including Carston 1997, 2002) and mentioned in section 4.1 above. On this view, then, all varieties of linguistically encoded meaning can be understood as weak, since there is no simple correspondence between the expression and how it is understood in a given context. Arguably, though, we can suggest that words which encode concepts encode something stronger than those which encode pro-concepts and that procedural meanings are the weakest of all. 4.5. Critiques of the conceptual-procedural distinction This approach has been critically discussed from a number of points of view. Here, I mention just two alternative viewpoints which have implications in different ways for the distinction between conceptual and procedural encoding and which might have an impact on the approach to the meaning of the tones developed here. A number of theorists have suggested that we should see all meanings as procedural. Espinal (1996), for example, has suggested that a language can be understood as a set of procedures and that the meanings of lexical items can be understood as ‘instructions for content building’ (Espinal

172

Billy Clark

1996: 109). To some extent, this view can be seen as going along similar lines to the ‘ad hoc concept’ approach mentioned above. It can also be seen as relating to work in ‘dynamic syntax’ ((Cann, Kempson and Marten 2005; Kempson, Meyer-Viol and Gabbay 2001; Marten and Kempson 2006) which sees language as a process of constructing representations rather than a static knowledge system. However, even if we accept the view that all language is procedural, we would still need to acknowledge (as Espinal does) a difference between expressions which encode a procedure based on conceptual material, however schematic, and expressions which do not. Another challenge to this approach has been developed by Bezuidenhout (2004). She argues against the possibility that procedural meaning can be semantic. She presents a number of arguments to support her view that procedures are aspects of ‘performance’ (how we use language in communication) rather than of ‘competence’ (the linguistic system itself). She also mentions the risk of ‘rule regress’ if expressions can include not just conceptual content but procedures which are followed to generate that content. Her argument is that we should consider procedures as ‘dispositional’, i.e. rather than claiming that expressions ‘encode’ procedures, we should think of these procedures as occurring whenever we encounter an instance of a particular expression. Wilson (2011) responds to these points with several observations, including the view stated in earlier work (Sperber and Wilson 1995: 172-173) that the relationship between lexical items and concepts is the same as that suggested between linguistic expressions and procedures. Neither conceptual representations nor procedures are part of the language or of linguistic meaning. In both cases, the claim is that something which activates or ‘points to’ a concept or procedure is part of the linguistic system and of linguistic meaning. On this view, neither concepts nor procedures are part of something akin to Chomskyan competence but both can be ‘activated’ by linguistic expressions. It is clear from this discussion that there are a number of questions still to be addressed about the nature of procedural meaning. However, both Bezuidenhout and Wilson assume a distinction between the kinds of meanings encoded, or dispositions activated, by expressions such as bulldozer and by expressions such as but. Whatever the details of this, the proposal about the tones developed here locates them in the same category as other expressions taken to encode procedural meaning within relevance theory.

Procedures and prosody: understanding weak communication

173

4.6. Weak meanings of tones The proposal here is that the tones have procedural meanings, i.e. that they are within the group of expressions with the weakest encoded meanings. In fact, the meanings suggested are weak even compared to those of other procedural expressions. The claim is that falls, rises, rise-falls and fall-rises encode procedures which affect either the explicatures or implicatures of utterances and often lead to interpretations where it is not clear exactly which explicatures or implicatures are the relevant ones. This might raise concerns about how weak suggested meanings can be and still play a role in explaining utterance interpretation. This section considers how this proposal fits into a pattern in post-Gricean work on semantics and pragmatics and suggests that the weakness of these proposals is not automatically a cause for concern. The next stage in the research project will be to test and refine them with reference to a wider range of data. One of Grice’s aims in developing his ‘theory of conversation’ (1975, 1989) was to suggest that natural languages were less ambiguous than had previously been assumed. Assuming pragmatic principles such as the maxims of conversation meant that some of the supposed ambiguities could be understood in terms of different inferential processes made on the basis of relatively stable encoded meanings. Grice went on to suggest as a methodological principle his ‘modified Occam’s razor’, namely that ‘senses are not to be multiplied beyond necessity’ (Grice 1989: 47). In this spirit, semantic analyses have become simpler and weaker as theorists have left more of the work to be done by pragmatics. There are many examples within the work of relevance theorists. Carston (1988, 2002) developed a relevance-theoretic account of the interpretation of utterances containing and which was very much in the spirit of Grice’s work in assuming that the word and has one encoded meaning, roughly equivalent to logical &. While this approach is Gricean in that and is taken to encode one meaning, the details of the account are quite different. Sperber and Wilson (1995 (originally 1986); Wilson and Sperber 1988) argued against both speech act and mood-based analyses of sentence meanings, replacing them with weaker analyses linking them with expressions of propositional attitude. Clark (1991, 1993) and Jary (2010) have developed and adjusted these analyses, in each case suggesting very weak semantic analyses which are then developed by the powerful pragmatic processes assumed by relevance theory. Perhaps one of the most extreme examples of a semantic analysis which seems very weak is Blakemore’s (2002: 128-148) suggestion for the en-

174

Billy Clark

coded meaning of one sense of well. She suggests that this might encode that the utterance following the occurrence of well is optimally relevant. On this view, B’s utterance in (42) encodes something we can roughly characterise as (43): (41) A: B:

Do you know Wendy? Well, I’ve met her.

(42) [This utterance is optimally relevant] B has met Wendy. This will strike some readers as odd, since relevance theory assumes that every utterance conveys a presumption that it is optimally relevant. This may also seem to have something in common with the semantic analysis of falls proposed here, since the proposition expressed by every utterance will be understood as either a description of a state of affairs or an interpretation of a thought. However, Blakemore points out that there is a difference between an utterance which encodes a guarantee of optimal relevance and one which generates it by virtue of being an act of ostensive communication: As we have seen, this guarantee is communicated by any speaker who is manifestly engaged in ostensive inferential communication. However, a speaker may recognize that there are circumstances in which his utterance will be recognized as being consistent with the principle of relevance only if certain assumptions which are manifestly not manifest to the hearer are made manifest. These may be assumptions which the hearer uses in the derivation of cognitive effects, or they may be assumptions about the speaker’s interests and preferences. Since it is in the speaker’s interests that the hearer take up the guarantee of relevance he is communicating and invest effort in the derivation of cognitive effects, it will be in his interests in such circumstances for him to provide a linguistically encoded signal that there are cognitive effects to be derived, or, in other words, that all is well. (Blakemore 2002: 147)

The hearer of B’s utterance in (42) has an extra guarantee that the following utterance is optimally relevant and so can be sure that B really thinks that this is relevant enough and as relevant as she can be given her abilities and preferences. This is, of course, consistent with an interpretation where B is not sure whether the limited acquaintance she has with Wendy will be enough to justify an affirmative response to the question. This is, of course, a good example to demonstrate the lexical pragmatic

Procedures and prosody: understanding weak communication

175

processes involved in working out the intended sense of a particular word, in this case know. B might not be sure how well B needs to know Wendy for that to count as the kind of ‘knowing’ that A has in mind. Similarly, there is a difference between an utterance which encodes that it represents either a description or an interpretation and one that does not. In fact, what is important here is the contrast between an utterance with a fall and one of the other tones. We might illustrate this by considering interpretations of B’s utterance in (42) if she had not used well but had used one of the tones considered here. The various possibilities are illustrated in (44): (43) a. b. c. d. e.

I’ve \ met her I’ve / met her I’ve /\ met her I’ve \/ met her I’ve > met her

On this view, (44a) indicates that the speaker is either describing the fact that she has met Wendy or attributing this thought to someone else. If the former, one interpretation is that she is expressing her belief that she has met Wendy. This would leave the hearer to infer whether or not this is enough to justify the inference that B knows Wendy. If B is taken to be attributing the thought that she has met Wendy to someone else, one possibility is that this is being attributed to A. B could in effect be asking A whether B has met Wendy. This is not very likely, of course, since we assume that B knows the answer and it is not clear how this would help to answer the question. A more plausible possibility is that B is reminding A that she has met Wendy. On this interpretation, perhaps A will decide that B thinks that any kind of encounter will be enough to provide evidence of ‘knowing’ Wendy. This brief discussion is perhaps enough to show how this approach accounts for the intuition that this utterance with a falling tone and no expression such as well will be hard to interpret. What about (44b)? Here the rise encodes that there is a proposition which B is not entertaining herself. One possibility is (45): (44) [B is wondering whether A knows that] B has met Wendy This interpretation would suggest that B thinks A can answer the question if he realises that B has met Wendy. This is not very plausible unless

176

Billy Clark

A and B think that one encounter will be enough. An alternative assumption is (46): (45) [B is wondering whether A thinks that it is enough in this context to know that] B has met Wendy. This is a more plausible interpretation since it would mean that B is unsure not about whether A knows she has met Wendy but whether A thinks that this is enough to provide a relevant answer. The rise-fall in (44c) encodes the same as the fall in (35a) but it adds the assumption that an explicature is surprising. This could be fleshed out to the higher-level explicature in (38): (46) [B is surprised that A needs to be reminded that] B has met Wendy On this interpretation, B clearly thinks that one encounter is enough to justify the assumption that she has met Wendy. The fall-rise in (44d) encodes the same as a fall but also indicates that an explicature is an interpretation of a thought. This might lead A to recover the interpretation in (48): (47) [B wonders whether A considers it relevant to know that] B has met Wendy To some extent, the effects of this interpretation are similar to those A might derive from the utterance with well in (42). This follows since both provide evidence that B is not sure whether her utterance is relevant enough. Finally, what about the level tone in (44e)? The claim here is that the speaker is not describing or interpreting (yet). The effect of this might be to lead A to conclude something along the lines of (49): (48) B has not yet responded to my question. This will lead A to expect B to say something further which will answer the question. I hope that this discussion shows both that these relatively weak analyses do provide the basis for accounts of how utterances with the various tones are understood and that they can form the basis of an account of how

Procedures and prosody: understanding weak communication

177

they interact with evidence from other sources including the use of expressions such as well. Clearly, there is much work still to be done here. This will include a more thorough exploration of similarities and differences with other accounts as well as more thorough empirical testing in the light of a variety of data. What I hope this has shown is that the line taken here, while tentative and proposing very weak analyses, is consistent with a broader relevance-theoretic framework and that it has the potential to play a role in an explanation of how a range of utterances are understood. 5. Concluding remarks The specific aims of this paper have been to develop Clark’s (2007) proposal about the meanings of Southern British English tones within a relevance-theoretic framework and to consider both the weakness of these semantic analyses and how they relate to broader relevance-theoretic assumptions, particularly about weakness in communication in general and in the relevance-theoretic account of linguistically encoded meanings. The proposal is that the meanings of tones fall within the domain of procedural meaning, the weakest of three kinds of linguistically encoded meaning assumed within relevance theory. Even within this domain, the analyses proposed here are very weak. The next step is to explore and develop the approach by considering a wider range of data. The hope is that analyses along the lines sketched here will reflect the intuitions of speakers, i.e. that they will be weak in just the right way. Acknowledgements I am very grateful to Frank Liedtke, Cornelia Schulze and their team for organising and hosting the Beyond the Words conference at Universität Leipzig, and to members of the audience there for very helpful comments and discussion.

178

Billy Clark

References Bezuidenhout, Anne 2004 Procedural meaning and the semantics/pragmatics interface. In The Semantics-Pragmatics Distinction, Claudia Bianchi (ed.), 101–133. Stanford CA: CSLI. Blakemore, Diane 1987 Semantic Constraints on Relevance. Oxford: Wiley-Blackwell. 2002 Relevance and Linguistic Meaning: The semantics and pragmatics of discourse markers. Cambridge: Cambridge University Press. Blass, Regina 1989 Grammaticalisation of interpretive use: the case of re in Sissala. Lingua 79: 299–326. 1990 Relevance Relations in Discourse: A study with reference to Sissala. Cambridge: Cambridge University Press. Brazil, David 1975 Discourse Intonation I. Birmingham: English Language Research, Birmingham University. Cann, Ronnie, Ruth Kempson, and Lutz Marten 2005 The Dynamics of Language. Oxford: Elsevier. Carston, Robyn 1997 Enrichment and loosening: complementary processes in deriving the proposition expressed? Linguistische Berichte 8, Special Issue on Pragmatics: 103–127. 2002 Thoughts and Utterances. Oxford: Wiley-Blackwell. Clark, Billy 1991 Relevance theory and the semantics of non-declaratives. PhD thesis, University College London. 1993a Relevance and pseudo-imperatives. Linguistics and Philosophy 16 (1): 79–121. 1993b Let and let’s: procedural encoding and explicature. Lingua 90: 173– 200. 2007 ‘Blazing a trail’: moving from natural to linguistic meaning in accounting for the tones of English. In Interpreting Utterances; Pragmatics and its Interfaces. Essays in Honour of Thorstein Fretheim, Randi Alice Nilsen, Nana Aba Appiah Amfo and Kaja Borthen (eds.), 69–81. Oslo: Novus. Cruttenden, Alan 1995 Rises in English. In Studies in General and English Phonetics, J. Windsor Lewis (ed.), 155–173. London: Routledge. Dennett, Daniel 1969 Content and Consciousness. London: Routledge.

Procedures and prosody: understanding weak communication

179

Doğan, Gürkan 1992 The pragmatics of indeterminacy and indirectness of meaning: a relevance-theoretic approach to epigrams and graffiti in Turkish. Ph.D. diss., University of Manchester. Espinal, Teresa 1996 On the semantic content of lexical items within linguistic theory. Linguistics 34: 109–131. Grice, H. Paul 1975 Logic and conversation. In Syntax and Semantic 3: Speech Acts, Peter Cole and Jerry L. Morgan (eds.), 41–58. New York: Academic Press. Reprinted in Grice 1989. 1989 Studies in the Way of Words. Cambridge MA: Harvard University Press. Gussenhoven, Carlos 1983 Stress shift and the nucleus. Linguistics 21: 303–339. Reprinted in Gussenhoven 1984. Gussenhoven, Carlos 1984 On The Grammar and Semantics of Sentence Accents. Dordrecht: Foris. House, Jill 2006 Constructing a context with intonation. Journal of Pragmatics 38 (10): 1542–1558. 2009 Prosody and context selection: A procedural approach. In Prosody Meets Pragmatics, Dagmar Barth-Weingarten, Nicole Dehé and Anne Wichmann (eds.), 129–142. (Studies in Pragmatics 8) Bingley: Emerald Group Publishing. Imai, Kunihiko 1998 Intonation and relevance. In Relevance Theory: Applications and implications, Robyn Carston and Seiji Uchida (eds.), 69–86. Amsterdam: John Benjamins. Jary, Mark 2010 Assertion. Basingstoke: Palgrave Macmillan. Kempson, Ruth, Meyer-Viol, Wilfred, and Dov Gabbay 2001 Dynamic Syntax: The flow of language understanding. Oxford: Wiley-Blackwell. Marten, Lutz, and Ruth Kempson 2006 Dynamic syntax. In Encyclopedia of Language and Linguistics, E. Keith Brown (ed.), Volume 4: 33–37. Oxford: Elsevier. Nadeu, Marianna, and Pilar Prieto 2011 Pitch range, gestural information, and perceived politeness in Catalan. Journal of Pragmatics 43: 841–854.

180

Billy Clark

Nicolle, Steve, and Billy Clark 1999 Experimental pragmatics and what is said: a response to Gibbs and Moise. Cognition 69: 337–354. Pilkington, Adrian 2000 Poetic Effects: A relevance theory perspective. Amsterdam: John Sperber, Dan, and Deirdre Wilson 1986 Relevance: Communication and cognition. Oxford: WileyBlackwell. 1995 Relevance: Communication and cognition. 2d. ed. Oxford: WileyBlackwell. 1998 The mapping between the mental and the public lexicon. In Language and Thought: Interdisciplinary themes, Peter Carruthers and Jill Boucher (eds.), 184–200. Cambridge: Cambridge University Press. 2008 A deflationary account of metaphor. In Handbook of Metaphor and Thought, Ray Gibbs (ed.), 84–105. Cambridge: Cambridge University Press. Steedman, Mark 1991 Structure and intonation. Language 67: 260–296. 2001 The Syntactic Process. Cambridge MA: MIT Press. Vega Moreno, Rosa 2007 Creativity and Convention: The pragmatics of everyday figurative speech. Amsterdam: John Benjamins. Wharton, Tim 2009 Pragmatics and Non-Verbal Communication. Cambridge: Cambridge University Press. Wichmann, Anne 2000 Intonation in Text and Discourse. Harlow: Pearson. Wilson, Deirdre 2011 The conceptual-procedural distinction: past, present and future. In Procedural Meaning: Problems and perspectives, Victoria Escandell-Vidal, Manuel Leonetti and Aoife Ahern (eds.), 3–31. Bingley: Emerald Group Publishing. Wilson, Deirdre, and Robyn Carston 2006 Metaphor, relevance and the ‘emergent property’ issue. Mind and Language 21 (3): 404–433. 2008 Metaphor, relevance and the ‘emergent property’ problem. The Baltic International Yearbook of Cognition, Logic and Communication 3: 1–40. Wilson, Deirdre, and Dan Sperber 1993 Linguistic form and relevance. Lingua 90: 1–25.

Procedures and prosody: understanding weak communication 1998

181

Mood and the analysis of non-declarative sentences. In Human Agency: Language, duty and value, J. Dancy, J. Moravczik and C. Taylor (eds.), 77–101. Stanford, CA: Stanford University Press. 2004 Relevance theory. In The Handbook of Pragmatics, Laurence R. Horn and Gregory Ward (eds.), 607–632. Oxford: Wiley-Blackwell. Wilson, Deirdre, and Tim Wharton 2006 Relevance and prosody. Journal of Pragmatics 38 (10): 1559–1579.

Pragmatic templates and free enrichment Frank Liedtke

1. Introduction Many, maybe the most, expressions are used in a way that makes it impossible to interpret them without assumptions about the utterance context. In this paper, I want to discuss a type of utterances which has been treated frequently in pragmatic literature. Examples are: (1) “I’ve had breakfast.”, (2) “It is raining.” It is usually assumed that in the mentioned examples mere propositional radicals are articulated which, on their own, cannot be used for further pragmatic processes (for example, conversational implicatures). This can only be done after adding so-called unarticulated constituents to the propositional radical; this way, a complete proposition can be expressed. “It is raining.” is complemented into “It is raining here.”, “I’ve had breakfast.” into “I’ve had breakfast today.” So, unarticulated constituents close the gap between what is articulated and what is expressed, the complete proposition. Pragmatic enrichments of that kind are assumed by a number of authors (including Perry 1998; Bach 1994; Carston 2002; Récanati 2002; Elugardo and Stainton 2004; Garrett and Harnish 2007). The assumption of unarticulated constituents was criticised by other authors, mainly with the argument that there is no criterion limiting the number and types of unarticulated constituents by means of pragmatics. An element such as “here”, which completes a propositional radical into a complete proposition, cannot simply be added. Instead, only those complements could be allowed that leave a trace in the logical form of the uttered sentence by appearing there as variables, even though they do not become visible on the “surface”. Another argument against unarticulated constituents is that some sentences contain operators which bind variables. The sentence (1) Wherever I am, it rains. contains an operator (Wherever I am... ) which binds, among other things, the variable here in the logical form of the sentence, because the value of the variable changes depending on the value of the operator: Here is exact-

184

Frank Liedtke

ly where I am. If here was not represented as variable in the logical form, the expression could not be bound; consequently, terms such as here are not unarticulated but they are indeed articulated as variable in the logical form. These critical arguments were presented in particular by Stanley (2000). Borg (2005) and Martí (2006) take a similar point of view. The idea that seemingly unarticulated constituents appear as bound variables in the logical form of the sentence caused a controversy about the question if added information (such as here) should be represented semantically as bound variable or pragmatically as unarticulated constituent (see Stanley and Szabó (2000a); Bach (2000); Neale (2000); Stanley and Szabó (2000b). Bach (2000) defends the pragmatic approach by permitting pragmatic enrichment processes – he calls them implicitures. These are elements located between what is said and what is implicated. Hall (2008) argues that unarticulated constituents can be limited by means of pragmatics, by using Fodor’s concept of language of thought. Therefore, they could be furthermore considered as a case of pragmatic enrichment. The debate about unarticulated constituents is very complex, and the individual standpoints of pro or contra cannot be represented extensively in this paper. Therefore, I will limit myself to a presentation of Perry’s and Récanati’s approach, because Perry is the inventor of unarticulated constituents, and Récanati offers the most extensive and most differentiated discussion of the elements of free enrichments (Récanati 2002, 2010). Like most of the pragmatically arguing authors, I am not of the opinion that the enrichment of a propositional radical is only possible when it is represented as a variable in the logical form of the sentence. But if unarticulated constituents are interpreted as free enrichments in Récanati’s sense, a basic problem occurs which is linked to the rational use of linguistic means. In my critique of Récanati’s approach, I refer in particular to the grammatical restriction that constituents are only unarticulated when they are not required by the valency of the verb, that is, they are not linguistically mandatory. I believe that this restriction contravenes the principle of the rational usage of linguistic means and therefore cannot be maintained. I want to propose an alternative procedure which is, at least for the mentioned examples of utterances, more suitable to represent the relevant processes of pragmatic interpretation than the concept of the unarticulated constituent is. I am following the maxim that what is articulated and what is expressed should be identified with each other within the model of description whenever possible, and I am of the opinion that such identification is possible in the mentioned examples.

Pragmatic templates and free enrichment

185

As an alternative to the concept of unarticulated constituents, I propose two pragmatic categories to describe the context-dependence of the above mentioned expressions. These categories are the central use of an expression – this category was derived from Levinson’s (2000) principle of informativeness – and that of the pragmatic template of its use. These proposals result from the idea that not only utterances can be assigned to particular types containing a morphosyntactic and a semantic representation. I rather suppose that also the connections between utterances and the situations they are mainly uttered in can be assigned to certain types. This results in privileged relations of utterances-with-a-meaning with stereotypical situations. These fixed, privileged relations between utterance and situation are a major part of our competence of verbal communication. The notion of a pragmatic template has some ancestors, eg. the notion of a “Sprachspiel” (Wittgenstein PI: § 23), or that of a “script” as a device for handling “stylized everyday situations” (Schank and Abelson 1975: 151). The term pragmatic template is a pragmatic correspondence to that of a semantic template or semplate as it is used by Levinson and Burenhult (2008), which is understood as a type of patterning in the lexicon, “which organizes distinct sets of words in terms of the same recurring oppositions” (154 f.). In syntax, especially in Role-and-Reference-Grammar and in Construction Grammar, grammatical structures are thought of as being stored as constructional templates, “each with a specific set of morphosyntactic, semantic and pragmatic properties, which may be combined with other templates to form more complex structures” (Van Valin and LaPolla 1997: 73; cf. Goldberg 1995). The common trait of the notion of a pragmatic template with its ancestors is the fact that a set of properties or entities forms a structured whole with characteristic features which is acquired and used as a linguistic or semiotic unit together with pragmatic factors of its use. The notion of a central use, which is roughly speaking a use the type of utterance is made for, has some analogies to the term “proper function”, which Millikan used on a different theoretical background (see Millikan 2008). Before I will say something about these notions, I will discuss the concept of unarticulated constituents as formulated by Perry and made productive for modern pragmatics by Récanati. After this, I try to clarify the concept of pragmatic templates and give further explanations of the related term of the ‘central use’. Lastly, I present arguments which demonstrate that pragmatic templates can explain at least some central cases of language use better than unarticulated constituents.

186

Frank Liedtke

2. John Perry on unarticulated constituents In his standard paper, Perry (1998) draws distinctions between the different ways the context of an utterance contributes to its meaning. The presemantic use helps us to figure out with which meaning a word is being used – this is about resolving ambiguity. When I ask for a “bank” I have to clarify whether I am looking for a financial institution or a fishing ground. The semantic use of context allows us to determine the reference of indexicals and anaphora such as he or she. Depending on linguistic or extralinguistic context, we use he or she to refer to other people. Therefore, it can be said that the meanings of anaphora and indexicals exploit the context (Perry 1998: 4) by relying on its contribution to the representation of meaning. The decisive use of context is the postsemantic one. As the name says, the meaning – or at least an essential part of the overall meaning – is already clarified, but the addressee of the utterances needs further information to figure out what the speaker has meant. According to Perry, there is a lack of “material” necessary to identify the proposition expressed by a statement. This lack of material remains even after resolving ambiguities and assigning reference to indexicals. Perry calls this missing material “unarticulated constituents”. The special thing about unarticulated constituents is that context does not inform us about the meaning of a used expression because no expression has been used. Therefore, this is not a case of clarifying an expression or making it more precise, but a case of adding or replacing. Some examples of utterances containing unarticulated constituents were already given: (2) It is raining. (3) They are serving drinks at the local bar. To understand (2) or (3) completely, we must know where the speaker is and which place is meant. So these sentences could be paraphrased (4) It is raining here. (5) They are serving drinks at the bar near here. (see Perry 1998:9) The expressions here and near here are unarticulated constituents of (2) and (3). However, these paraphrases are not always valid because it is possible that (2) or (3) refer to places different from where the speaker is – then it has to be there or near you. An appropriate localisation depends on the intention of the speaker. In any case, a constituent must be added to

Pragmatic templates and free enrichment

187

what is said to assign a truth value to the proposition. This constituent must be supplied by the context because the sentence does not contain a morpheme carrying the necessary information (Perry 1998: 9). Perry writes: “... we don’t articulate the objects we are talking about, when it is obvious what they are from the context” (Perry 1998: 11). He distinguishes between three types of cases in which constituents of what is said are left unarticulated because of contextual information. The first type of case is that, within a certain n-ary relationship, an argument is always occupied by the same object. Therefore it is irrelevant for the speaker to articulate that object. For example, time is usually not reported in relation to a particular time zone because some speakers (children, say) do not know it, and others (adults) are not always aware of it. In some cases we also speak “egocentrically”, when we do not mention our own position, for example in (6) The house is on the left. vs. (7) The house is on the left of me. The second type of case is that the position of the speaker is constantly changing – for example, when he is moving through different time zones – therefore, the occupant of the unarticulated argument varies. Perry characterises cases of the third type by mentioning Lewis’ concept of possible worlds – we are always including them when we are talking about the necessity and possibility of things. After all, Perry’s concept of unarticulated constituents is shaped by semantics – even though it is about the postsemantic usage of context. The argument of an n-ary predicate is left empty in the expressed proposition; an n-1ary predicate emerges. Assumptions about the context inform us how to fill this role. Nevertheless, it seems indefinite how many places the predicate has, that is, how big n is. Is the time zone always an unoccupied argument when we are talking about time? And is it a 1-place predicate when the time zone is not mentioned? Taking modern physics into consideration, statements about time are even more relative, but should that be considered an argument in utterances about time in every case? Perry’s approach raises several problems of that kind; before I discuss these problems, I want to present the opinion of another author who has an opposite standpoint to Perry. He defines the concept of unarticulated constituents pragmatically, namely as a type of free enrichment.

188

Frank Liedtke

3. François Récanati: Free enrichment F. Récanati (2002) restricts the concept of unarticulated constituents in a decisive way, considering only those phenomena which – in contrast to Perry’s concept – do not occur as a (missing) argument of a predicate. This way, the first step towards a pragmaticalisation of the concept is taken. Récanati’s concept of unarticulated constituents differs in some respects from Perry’s concept and his approach is outlined in the following section. Récanati distinguishes between metaphysically and communicationally unarticulated constituents. Actions – for example the fact that Mary is dancing – do always take place somewhere at some time, however, knowledge about the place and time is not necessary to understand the utterance that Mary is dancing. Also relating to a time zone is usually not necessary to identify a time of day. This is the metaphysical sense of an unarticulated constituent. On the other hand, the communicational sense is given when it is obligatory to know the unarticulated information in order to understand the utterance. That is the case in example (2). If the addressee has no information about the place where the event occurs, he or she cannot fully represent the proposition that it is raining – he or she needs the corresponding unarticulated constituent (see Récanati 2002: 305 f.). In the mentioned paper, only the communicational interpretation of unarticulated constituents is relevant, because it is about a theory of understanding utterances, and for this purpose, the assumption of unarticulated constituents in a metaphysical sense is neither necessary nor useful. The separation of metaphysical and communicational unarticulatedness can also reduce decisively the number of unoccupied arguments of the predicate – only those are relevant which are responsible for communicational unarticulatedness. Consequently, a part of the problems related to Perry’s concept of unarticulated constituents is solved. Within the communicationally unarticulated constituents, another distinction is made which is about the consequences of unarticulatedness for the understanding of an utterance. Either the utterance is vague, so that information has to be added to determine the fact the speaker is referring to, or the utterance is incomplete, so that information has to be added to identify the fact at all (see Récanati 2002: 307 f.) The first type (the Atype) occurs in those cases Récanati gives as example of unarticulated constituents. It contains the types of usage which are given as examples in pragmatic literature since Sperber and Wilson:

Pragmatic templates and free enrichment

189

(8) A: Would you like to eat something? – B: I’ve had a large breakfast. or (9) Mary took out her key and opened the door. (8) is usually understood the way that B has had breakfast on the day of the utterance. His answer is extended by A in a way that it contains the adverbial this morning. If A would not make this enrichment, the relevance implicature (that B does not want to eat because he has just eaten something) was not possible. But this is the relevant information B wants to convey with his answer. Utterance (9) has to be interpreted the way that there is a relation between the clauses that contains an instrumental component. The key is the instrument the door is opened with, so the enrichment is: ‘... and opened the door with the key.’ The second type of unarticulated constituents, the B-type, contains cases in which facts cannot be identified without the unarticulated constituent, which means that no proposition had been expressed. This is given in the case of the rain-example: No truth value can be assigned to the proposition without occupying the argument of the place. It is decisive for Récanati’s further argumentation to ignore unarticulated constituents of the B-type – they are irrelevant for the question how much pragmatic information a proposition must contain to be assigned with a truth value. (see Récanati 2002: 308). Following Récanati, the reason for limiting unarticulated constituents to those of the A-type is that only these are cases of unarticulatedness in a strong sense; in the B-cases the constituents in question are not uttered but articulated in a weak sense. Récanati uses the expression too small to clarify what matters. The comparison class in relation to which a man is described as too small is a necessary element of the interpretation of this expression. If somebody is too small for a basketball player, that does not mean he is too small compared to non-basketball players. This kind of additional information, which is necessary to understand the utterance, is articulated by uttering the words too small. They trigger the search for a relevant comparison class. Therefore, the sentence contains, in a very abstract way, something which can be regarded as articulated (see Récanati 2002: 311 f.). The restriction of unarticulated constituents to those of the A-type is explained with the distinction between saturation and free enrichment,

190

Frank Liedtke

which is, for Récanati, a basic distinction between the different pragmatic amounts of what is said. Saturation is defined as the contextual assignment of semantic values for context-sensitive expressions. It is a bottom-up pragmatic process, and it is triggered by the linguistic expression of the uttered sentence. Free enrichment is not triggered by a linguistic item of the sentence but by contextual conditions – thus it is a top-down pragmatic process. B’s answer in (8) is an example of free enrichment in which the information this morning is added (= I’ve had breakfast this morning) without being syntactically required – the sentence is correct without the enrichment. The already mentioned case (10) is an example of saturation: (10) He is too small. To evaluate if the utterance made in (10) is true, we must know in which respect he is too small. Especially the adverb too establishes a relation between the property of being small and another property, for example of being a basketball player. Because of this property, too is an indexical adverb (here, Récanati is referring to Heal 1997). If unarticulated constituents are only those which are not triggered by an expression within the sentence, pragmatic saturations cannot be (or at least only in the weak sense) cases of unarticulated constituents. Thus, the only possible candidates for unarticulated constituents are free enrichments. Récanati introduces an argument for the distinction of unarticulated constituents in the weak and the strong sense by examining the intransitive use of two-place predicates. He makes an essential distinction between the non-realization of an existing argument and the absence of an argument (see Récanati 2002: 314 f.). We can use the verb eat in a way that we express the person’s property of ‘eating’, for example, if this person has not eaten for a long time: (11) She eats. We cannot use other verbs this way, for example: (12) She finished. For using (12) appropriately, the context of the utterance must give information about what the person finished, otherwise (12) is not possible. This is because the verb denotes, in the semantic sense, a relation (between

Pragmatic templates and free enrichment

191

the person and what she finished), but not a property. This condition is not valid for (11); it can be absolutely irrelevant what the person is eating, thus, a property is denoted. This is not omitting one argument containing a relatum of a two-place relation – there is no second argument. According to that, Récanati uses a very restricted concept of unarticulated constituents. If an argument within the proposition’s structure is not filled, this signals that the constituent is merely unarticulated in the weak sense. Only in the second case, when the proposition does not contain an element realizing the argument we can call it unarticulated. Récanati sums up this position in a subheading: “True Unarticulated Constituents are Never Mandatory” (2002: 313). Récanati’s analysis, as far as I have outlined it here, has the advantage that it is based on a clear linguistic distinction between two cases: the case in which, for example, the argument of a transitive verb is not carried out (unarticulatedness in the weak sense), and the case in which a transitive verb is used intransitively, and no second argument requires realization besides the agent position (unarticulatedness in the strong sense). Récanati offers the following rule of thumb to distinguish between both cases: “Can we imagine a context in which the same words are used normally, and a truth-evaluable statement is made, yet no such constituent is provided? If we can imagine such a context, then the relevant constituent is indeed unarticulated. ” (2002: 316). Based on this criterion, Récanati formulates the Optionality Criterion: “Whenever a contextually provided constituent is (truly) unarticulated, we can imagine another possible context of utterance in which the contextual provision of such a constituent would not be necessary for the utterance to express a complete proposition.” (2002: 323)

Verbs which can be used transitively as well as intransitively do fulfill this criterion. If I say (intransitively) about somebody, that she is eating, the additional information that she is eating mushrooms is a free enrichment, because I can imagine a context in which I can express a complete proposition by saying that she is eating – without further information. This does not hold true for the verb finish, because there is no imaginable context in which I can express a complete proposition by saying that she finishes – without further information. An argument place remains unoccupied, therefore it is the bottom-up semantic process of saturation, not the top-down pragmatic process of free enrichment. This might be an overgeneralization, because I can imagine a context in which I say about a person that she finishes (quickly) whatever – for example because she is irresolute

192

Frank Liedtke

or timid. In these cases, the verb finish is just used as if it were intransitive – under the condition that a stable characteristic of a person is denoted, and not a one-time behaviour. So we should not formulate the difference between eat and finish as an absolute one, but as one of tendency. Speakers show the (strong) tendency to use finish transitively, but an intransitive use is not always ungrammatical – even without context. Until now I have only described a part of Récanati’s complex analysis of unarticulated constituents, but in my opinion it is a central part, containing their definition and the Optionality Criterion. I want to discuss this restrictive version critically, and I will do so by using the fundamental rationality of speech acts as an argument. There is, I think, a very basic property of using speech acts rationally which can be formulated as follows: Whenever we communicate, we choose (linguistic) means which are most appropriate for the realization of our intentions. This has been already expressed in Asa Kasher’s rationality principle, which he puts in the place of Grice’s cooperation principle: “Given a desired end, one is to choose that action which most effectively, and at least cost, attains that end, ceteris paribus.“ (Kasher 1976: 205; see also Keller 1995). From the addressee’s perspective, Sperber and Wilson are relating to the rationality of speech act use, which is the basis and prerequisite of the relevance principle: “In trying to identify [the] informative intention, the addressee must assume the communicator is communicating rationally […]. Intentions are identified by assuming that the agent is rational, and by trying to find a rational interpretation of her actions.” (Sperber and Wilson 1986: 165). In the context of the present case, the desired end of the speaker’s speech act is to be understood by his addressee(s). That means that the choice of a certain expression for a communicative goal is rational, if the expression helps at lowest cost to attain this end, i.e. helps to identify the speaker’s intention. If we consider the use of (transitive or intransitive varieties of) verbs, we will see: speakers will choose an intransitive verb as an appropriate means if they intend to express, for example, that a person’s ‘property’ is to eat (because this is important), without mentioning what the person is eating (because this is irrelevant). Maybe a child was ill and now has started eating again. Even though the child is eating things we consider as unhealthy, we are happy that the child is healthy again and eats. On the other hand, if it is important for us or our interlocutor what the person eats, we choose the transitive variety of the verb as appropriate means. Maybe the child did not like to eat vegetables but know starts to eat vegetables. In this case, the information that the child eats vegetables (and not French fries) is extremely relevant.

Pragmatic templates and free enrichment

193

Let’s get back to Récanati’s concept of unarticulated constituents, which he identifies with free enrichments: How do we use them in communication? In situations where context or knowledge of the world provides relevant information, it is rational for a speaker – on the basis of the least-cost-strategy following from Horn’s R-principle (cf., Horn 1984) – to utter only that information that does not result from context knowledge – one reason might be our limited processing capacity (see Levinson’s bottleneck-argument; Levinson 2000). It must be clear which information from the context is relevant and how it should be integrated into the utterance, so that the utterance succeeds. That means for the addressee (following from Horn’s Q-Principle, see Horn 1984) that the effort to interpret this utterance increases correspondingly, and this must also be anticipated by the speaker. Generally speaking, free enrichments will only be added when they are communicatively relevant; if this relevance is lacking, only the articulated utterance is used without the enrichment. In our example, utterance (11) is pragmatically enriched when it is important what the child eats – she eats things containing vitamins, what she has not done before. The utterance will not be enriched when it is relevant that the child eats – no matter what. From the Optionality Criterion results that free enrichments are only allowed in the first, intransitive case, when it is about that the child eats, no matter what – because otherwise we could not imagine a situation in which a complete proposition is expressed with that expression. But these are exactly those cases in which no enrichment is being made, because it is just not relevant. We, the rational speakers, have the (strong) tendency to choose the intransitive verb, because the second argument (the direct object) cannot or does not have to be named. There is no reason why we should expect that the utterance should be enriched and by doing so bother the addressee with additional cognitive effort (in the sense of Sperber and Wilson 1986). If one, like Récanati, follows the principle that unarticulated constituents (in the strong sense) can only be assumed when there is a possible context in which they are not triggered by the utterance (that is, when there is an intransitive use), a conflict of relevance criteria arises. Relevance of free enrichment is not given when we choose an intransitive verb to express a property (and not a relation) that is not a common characteristic of the described person. If the fact what somebody is eating was relevant, we would, as mentioned, choose a transitive verb, because that seems reasonable due to the choice of expressions (as a form-meaning-pair). In brief: If it is relevant to describe a situation with an intransitive verb that has a transi-

194

Frank Liedtke

tive alternative, it is irrelevant to enrich the utterance pragmatically, and vice versa. If this observation is correct, it follows that the most important condition for unarticulated constituents, which is that they are not linguistically triggered, gets into conflict with the basic rationality principle which controls the choice of linguistic means for particular communicative purposes. Choosing a linguistic means with certain grammatical properties is a pragmatic fact like the free enrichment of a semantically underdetermined utterance. That means for the present case that unarticulated constituents can exist even if a second relatum is linguistically triggered. In other words: The restriction of unarticulated constituents to expression usages that do not demand any further complement in at least one context is wrong. If the situation is as it is described in the forgoing paragraph, one can draw two consequences. Either one may give up the concept of unarticulated constituents because it leads to a conflict within the relevance criteria; or one uses another concept of unarticulated constituents which is based on another defining criterion, so that the criteria for rational communication are fulfilled. I do not rule out the possibility that such a defining criterion exists, but in the following I will propose an alternative concept which replaces, at least partially, the concept of unarticulated constituents. I believe we can achieve theoretically consistent analyses of speech acts if we avoid two things as far as possible: i) allowing a too big gap between what is articulated and what is expressed, for example between the uttered sentence and the performed speech act, ii) assuming constituents which are not present in the uttered sentence in any way. In this spirit, I will propose an alternative strategy how to explain the communicative function of unarticulated information. 4. Pragmatic templates and utterance interpretation In this section, I will give some reasons in favour of the position that the information which speakers do (sometimes) not articulate in their utterances, although they want the addressee to get that information, should not be represented regularly as part of the utterance (of its communicative content). It is nothing the speaker says. It is rather something the addressee (hypothetically) already knows, which thus does not need to be articulated because it can be derived from the context. Nevertheless, the speaker is calculating with this knowledge, it is part of the “nonlinguistic infrastructure” (see Tomasello 2009: 69) to which speaker and addressee are refer-

Pragmatic templates and free enrichment

195

ring. Now we have to explain what that something is that can be derived from context. The rational usage of linguistic means in utterances is subject to certain conditions, and these conditions restrict the usage of expression types. They are not an unsorted collection of constraining propositions, but they form structured clusters of conditions of usage. They contain the obligatory context elements in which a certain expression can be used appropriately. These clusters are a part of what I call pragmatic templates. If we want to identify such a template, we have to assign the utterance to a specific type (form type). We also have to be able to give criteria of what is part of a template and what is not. I propose to consider the content of pragmatic enrichments or unarticulated constituents as a part of such a pragmatic template – it is part of the knowledge of language users which is activated when they hear or read an utterance in which not everything is articulated that could be relevant in that situation. In this section, I would like to confine myself to the classical type of utterance (2) “It is raining”. The question is how we can include the place of the rain into the analysis without assuming unarticulated constituents. How do we obtain here (or there ...)? To answer this question, I would first of all like to explain the concept of pragmatic templates, conceived as types of environments into which utterances are embedded. Such an environment contains the speaker, the addressee(s), the common background, a relevant section of the ongoing dialogue, and a relevant section of encyclopaedic knowledge. A pragmatic template represents a structured section of this entirety, and it adopts a specific form for the different classes of utterances. J.L. Austin presented several examples of those clusters in his lectures on performative utterances and formulated the felicity conditions for those utterances. The conditions A and B, which have to be fulfilled to prevent the respective utterance from being a misfire (misinvocation or misexecution), can be seen under the aspect that they form conditions for a template which is valid for speech acts such as: “I name this ship the Queen Elisabeth” (cf. Austin 1976: 17). There are (at a christening) certain actions that have to be performed, the christening person must be authorised, the time must be right, and the utterance must correspond to strict formal criteria. Searle’s taxonomy of speech acts contains the classes of expressives and declarations which are also ritualised and determined to specific situations. As Searle himself says, declarations form that class of speech acts that resembles the class of performative utterances (see Searle 1976). We have the competence to perform and understand declarations because we dispose of relatively fixed connections between utterances of a certain

196

Frank Liedtke

form and the conditions of usage specific to these forms of utterances. If speakers can perform declarative speech acts, this means they have learned these connections and can apply them now – one might say that they have a pragmatic template at their disposal. Expressive speech acts on the other hand like ‘greeting’ or ‘condoling’ are also determined to specific situations which are characterised by a structured cluster of context features. We can only greet someone we see for the first time of that day, and the way of greeting depends on the relationship S and A have, and / or on the situation in which S and A meet. Language usage does not only consist of speech acts uttered by authorised people in specific institutions or as part of a ritual with determined roles. Besides these cases, there are of course cases in which we spontaneously utter things because we intend to convey something specific in a specific situation. But even here at least parts of the utterance have standardized features and are therefore connected with certain characteristics of the context which are known by the addressee. Until now, we achieve the following interim result: I define templates as holistic clusters of relations between utterances on the one side and elements of an utterance situation on the other. These relations can be singular, they can be created for a certain utterance, if this utterance, or the combination of utterance and context, does not have a model. But, on the other side of the spectrum, it might also be the performance of a declarative speech act or an expressive ritual; thus, the utterance has a determined pattern, or the combination of utterance and context follows a model. In this case, we can consider it a relation between an utterance and a stereotypical utterance situation. In any case, the possession of such a holistic cluster is in different degrees a prerequisite for understanding an utterance. Speakers know this and they construct their utterances in a way that they, combined with the assumed context elements, fit into such a template so that they form an integral part of it. After these general remarks on pragmatic templates, we have to answer the following question: Which role do templates perform in non-explicit communication, and to what extent is this concept more useful to explain linguistic behaviour than the concept of unarticulated constituents? I want to clarify this point by applying the outlined idea on the example discussed in the beginning: (2) “It is raining”. What is the classical or canonical situation we are thinking of when we hear a sentence like that? A possible situation is: A is looking out of the window. B is in the same room, reading. Both want to go for a walk. A says to B: “It’s raining”. If we want to describe this situation a little more abstract we could say:

Pragmatic templates and free enrichment

197

Speaker and addressee are in the same room, A perceives an actual event which has immediate relevance for A and B respectively which they are a part of, and B does not perceive this event. In brief: A calls B’s attention to an event within the immediate circuit of their interaction. When a person is calling another person’s attention to something, it is presupposed that the speaker, the addressee, and the event the speaker is referring to are co-present. As we will see later, the concept of co-presence is very vague. For the present purpose it is sufficient that these three “entities” are part of a situation, spatial as well as temporal. On condition that such a description is true for the use of (2), we have described a possible template for this type. Of course other specimens can be assigned to that type too, for example (2i) “It’s snowing”, (2ii) “It’s hailing”, (2iii) “It stinks”. We might call it the involvement-template: speaker / addressee are involved in the described event. What matters in this short description is that the co-presence of A, B, and the event is not part of the utterance, but it is conceived as part of the “nonlinguistic infrastructure”. How do we know that speakers think it is raining here when they say that it is raining? Even more difficult to answer is the question: How do we know that speakers mean it is raining here when they say that it is raining? We know that (2) was uttered, and we assume that it was used rationally, because we want to interpret it. We assume that (2) was used rationally without a modifying adverb when the environment of the utterance contains the event of ‘raining’. We also assume that (2) was used less rationally when the environment of the utterance does not contain this event – unless the context contains another feature which suggests another localisation. It is possible that A and B were talking about a daytrip to the mountains, and in that context A utters (2). Under these circumstances, the co-presence of A and B on the one hand and of the event on the other hand are not part of the template, but the localisation of the event in the place they were talking about – and nowhere else. If what has been said is part of the conditions for the rational usage of forms of expression, it must not be mistaken for what speakers think, mean, or even say. We do not transmit the presumptions of rationality that we, the communicators, mutually assume, but we take them for granted for communication. That also holds true for an event such as rain. If we do not make this restriction about the nature of discourse knowledge, unarticulated constituents will most likely proliferate, because, in principle, any element of the template can be represented as an unarticulated constituent. So it would be impossible to avoid the absurd conse-

198

Frank Liedtke

quence that the utterance is: (2abs i) “Hello B vocative, it’s raining here.” – because the addressee is part of the utterance situation. Or you get: (2abs ii) “A is speaking: Hello B vocative, it’s raining here.” – because speaker and addressee are part of the utterance situation, and so on, ad lib. No one would ever accept this absurd consequence; but it would be the result when we do not make that essential distinction between what speakers mean (or even say), and the conditions for the successful performance of a speech act. One might go so far as to assume that the co-presence of speaker, addressee, and event make up the central use of an utterance type such as (2), that means that co-presence is assumed when there was no other place mentioned before – as a default-value (see Jaszczolt 2002). I consider this assumption plausible because it corresponds with the intuition linked to utterances of type (2). But this default-value is not valid for similar-looking but different types of utterances. One could assume that the event is necessarily not co-present but takes place in a certain distance – as in (2iv) “There is lightning”. Here, a co-presence is not possible for physical reasons. In this case, this would be the central use, and the template would contain the event in another modus – it does not have to be present but perceptible. There are also cases in which the event (or the state) does not even have to be perceived (as in 2v “There are rough seas”). In this case, the template contains a general epistemic condition that the speaker has his knowledge from a source of information, either from own consideration or from the report of another person. However, it seems to be difficult to explain the differences in default-values on a communicative basis. The way they are formulated here, one goes back to what Récanati calls a metaphysical reason, which is problematic. I don’t want to rule out the possibility that default-values for templates can be formulated nevertheless. But that would be only possible under the prerequisite that metaphysical conditions could be translated into communicative conditions of rational choice of expressions. For the moment, I’m not going to continue that search. Let’s get back to (2): In that case, it is important for the concept of pragmatic templates that a localisation of the event is a part of this template – and we master it because we have learned it during language acquisition within our culture. It is also important to recognise that type and composition of templates are subject to cultural differences – this applies a fortiori to the exploration of concepts of space. It is imaginable that values are allocated differently in different cultures, so that the co-presence condition of weather verbs is not given in that rigour, because different concepts of involvement prevail. This means that the involvement-template would

Pragmatic templates and free enrichment

199

be realized differently depending on the culture-specific “readings” of the involvement concept. 5. The architecture of pragmatic templates I am of the opinion that utterances of sentences such as (2) and related ones can be interpreted correctly because of their assignment to a relevant template. When we interpret an utterance in using a pragmatic template, we trace it back to an acquired holistic structure of characteristics of the environment it is used in. So we do not have to add an unarticulated constituent in every situation, but we can use our standardised conversational knowledge. That knowledge provides us with prototypical applications (the central use) of specific types of utterances. Thus, it also supplies the conditions of the relevant template. It is self evident that this tracing back does not complete the interpretation. There are always parts of the meaning related to the situation that have to be represented ad hoc. But I claim that a major part of the interpretation can be accomplished by identifying a type of utterance as part of a specific pragmatic template. Even if we assume the template has the hypothetically assumed structure, it is still described in a very general way, and many questions remain open. Not only the template’s architecture can vary, but also the filling of the positions of the cluster. Considering the rain-example, there is the question what can be accepted as ‘co-presence in a shared place’ – and what not, even against one single cultural background. Of course we can call the most different things ‘a shared place’: a room, a house, a plot of land, or even a region or a continent. Furthermore, the concept of place can be used figuratively for an interaction, an institution, a culture, a religion. I’m not sure whether the outlined template can include also these places and represent nevertheless an instantiation of the central use. In example (2) we would perhaps assume a region – not for linguistic but ontological reasons – which is the maximum expansion of an area of rain. Of course the here-place can be smaller than the area of rain, and it can also be larger in a specific way, so that the coordinates of the area of rain are a subset of the coordinates of the entire place. Related to that case, there is the question if A and B can be at the margin, or even outside the area of rain, without invalidating the template. If A and B are inside a house, they are not in the rain, but the house they are in is in the rain. Furthermore, we can surely accept the interpretation, if the rain is within sight but the house is not getting wet. Such a concept of co-presence is probably

200

Frank Liedtke

even a part of the central use, because (2) would not be very relevant if A and B were in the rain. But we are pushing against the bounds of acceptability when the distance between the rain and the interlocutors increases. Let’s assume A says (2): “It is raining.” and means the next mountains without mentioning this before. In this case, it is unquestionably difficult for B to find a relevant interpretation, because he does not know spontaneously what A is talking about – it cannot be the shared (dry) place, because A is apparently not lying in such a naïve manner. For a successful interpretation, at least a past history is needed in which the planned trip was mentioned. A related case is: A and B want to go on a city trip together, and A is watching the central place of the city, where it is raining, in a web cam. In this case, the only important fact seems to be that A is looking at the picture in that moment, like he was looking out of the window in the first example and B observes or knows that, so that it is obvious that A refers to what he is perceiving at that moment. Thus, we can extend the template (and our notion of involvement) and say that there must be an event or the pictorial representation of an event A and B are exposed to. John Perry (1998: 9) discusses in his paper a situation in which the speaker is talking on the phone to a relative C and then interrupts the conversation to utter (2) to the present family members. In this case, one could say that the template is modified, because speaker A and the addressees B1 ... Bn are in one place, and the event (as well as the relative C who also uttered (2)) in another. In this case, the unarticulated constituent is not here but there. But I don’t think that we are forced to take that consequence, because the decisive fact in this example is that relative C uttered that sentence before – what the addressees B1 ... Bn conclude because of the mentioned epistemic condition. The utterance (2) by speaker A is not the original utterance but a verbatim quotation, and in case of verbatim quotations, deictic expressions are not modified. To be precise, the utterance of the speaker on the phone must be written like this: (2vi) “‘It is raining’.” It is evident that in this case, here remains the unarticulated constituent in Perry’s sense. Accordingly we can assume that speaker A and addressees B1 ... Bn activate the original involvement-template in which the whole situation is shifted to the place where relative C stays. It is also possible that speaker A does not perceive the rain immediately. It is sufficient that A perceives a reliable indicator for the rain. For example, meteorological devices are indicating the rain, or the cat is not leaving the house, or a scar is aching. But this case is not central in the full sense, because an error would more likely be put down to the mistake of inferring

Pragmatic templates and free enrichment

201

rain from the indicator – or to a malfunction of the indicator – but not to the speaker’s lacking perspective faculty. The immediate perception can be replaced by other aids, but A and B know that these aids become less reliable with increasing distances. So we have to classify the description of the template more clearly by saying that the event E, or a pictorial representation of E, or an indicator of the presence of E must be present. A last case has to be considered, concerning the presence of the addressee. Is it really necessary? One might imagine that B (staying in Germany) is phoning with A (staying in Sweden), who says (2), whereas B is answering (13) “The sun is shining.” This would not work as a rational verbal interchange, because neither B nor A did make clear in their contribution on which site they refer respectively. So this case seems to be analoguous to the “mountains”-example mentioned above: One has to make explicit the relevant place, under penalty of conversational breakdown. Can the proposed sketch of pragmatic templates be applied to other examples used for unarticulated constituents? Let’s take the following, well established example: (8) A: Would you like to eat something? – B: I’ve had a large breakfast. This example differs from (2) in the point that a pragmatically appropriate interpretation enables us to specify the time quite exactly, without heuristic assumptions about the context: Breakfast is the meal taken after getting up, and in the case of (8) it is at the day of the utterance. Because breakfast and all other regular meals are taken once a day in the western culture, the strategy of the unarticulated constituent is implausible. Why should a speaker mean by his answer in (8) that he’s had breakfast today, when the addressee is aware of it? It would be absolutely irrelevant for A to say this to B, just as irrelevant as saying: “I ate a steak (with fork and knife)”. We all know that steaks (in our culture, in which steaks are an ordinary meal) are eaten with fork and knife, and we all know that we have breakfast once a day (not three times a day or once a week). Therefore there is no need to communicate it. I am of the opinion that unarticulated constituents as well as articulated constituents are subject to the Gricean cooperation principle and maxims, because they are conceived of as being part of the utterance, of what was said, and they enter into conversational inferences like particularized or generalized implicatures. From this follows, that something like (8i) “I’ve had breakfast today.”, if trated as something what-is-said, contravenes the

202

Frank Liedtke

maxim of relevance. It triggers the implicature (based on this maxim) that B does not have breakfast every day, which means that the breakfast is something extraordinary. The total signification, that is, the mentioned relevance implicature of the utterance, is therefore contrary to the pragmatically appropriate reading of B’s answer in (8), which is precisely that B has had breakfast on that very day – which is only possible if one assumes that B is having breakfast every day after getting up. The contradiction between the relevance implicature of (8) plus unarticulated constituent and the pragmatically appropriate meaning of (8) is a strong argument for dropping the assumption of unarticulated constituents in this case, too. Also here, the alternative is to represent the culturally specific knowledge about time and frequency of meals (which are known to be divergent from culture to culture too) as part of the interlocutors’ prior knowledge. This knowledge is part of their cultural competence. In contrast to the more variable weather-example, the breakfast-example is a case of encyclopaedic knowledge, which we acquire in our socialisation. Consequently, there is no need to generate it from scratch every time (8) or a related sentence is uttered, but it is at our disposal all the time. Let me call it the periodic template. Other examples are (8i) “I had lunch”, (8ii) “I took my sabbatical”, (8iii) “We did our spring-cleaning.” Here, the cognitive economy of the template-model is rather evident. I have used these two examples to demonstrate the idea and functioning of pragmatic templates. Now it must be illustrated what such a template may look like. To conclude, I try to present a scheme on which such a template could be based on for the weather-example and the breakfastexample: PRAGMATIC TEMPLATE FOR “IT IS RAINING” (INVOLVEMENT TEMPLATE) Form of utterance: Morpho-syntactic form of S {Expletivum, V3rd pers.sg.pres.} Sub-categorisation of V {weather verb} Encyclopaedic knowledge: Rain is a phenomenon that cannot be manipulated, and that is spatially and temporally limited. Lexical meaning of to rain Stereotypical properties of the utterance situation: Speaker – spatially-temporally present Addressee – spatially-temporally present Event – spatially-temporally present

Pragmatic templates and free enrichment

203

PRAGMATIC TEMPLATE FOR “I’VE HAD A LARGE BREAKFAST” (PERIODICAL TEMPLATE) Form of utterance: Morpho-syntactic form of S {Pro, V 1st pers.sg.past tense} Sub-categorisation of V {meal taken after getting up} Encyclopaedic knowledge: One has breakfast once every day. Lexical meaning of ‘ I’, of ‘large’ and of ‘to have breakfast’ Stereotypical properties of the utterance situation: Speaker – spatially-temporally present Event – taking place at the day of the utterance The description of these two (types of) templates might be incomplete, and maybe here and there a better way of describing the constitutive parts is possible, but I am convinced that something of that kind must take the place of unarticulated constituents. When relevant information is conceived as part of the pragmatic template which we acquire in learning how to communicate, a concise and economic model of utterance production and interpretation should be possible. The “nonlinguistic infrastructure” of the utterance situation must not be projected into the utterance itself, because the conversational balance between utterance and inference is disturbed – with the consequence of unwanted implicatures. Only when this is avoided, it is possible to create an adequate pragmatic theory of utterance understanding. References Austin, John L. 1976 How to do things with Words. Oxford: UP. Bach, Kent 1994 Conversational Impliciture. Mind and Language 9: 124–162. 2000 Quantification, Qualification and Context. A Reply to Stanley and Szabó. Mind and Language 15: 262–283. Borg, Emma 2005 Saying what you mean: Unarticulated constituents and communication, In Ellipsis and nonsentential speech, Reinaldo Elugardo, and Robert J. Stainton (eds.), 237–262. Dordrecht: Springer.

204

Frank Liedtke

Carston, Robyn 2002 Thoughts and Utterances. The Pragmatics of Explicit Communication. Oxford: Blackwell. Elugardo, Reinaldo, and Robert J. Stainton 2004 Shorthand, Syntactic Ellipsis, and the Pragmatic Determinants of What is Said. Mind and Language 19 (4): 442–471. Garrett, Merill, and Robert M. Harnish 2007 Experimental pragmatics. Testing for implicitures, Pragmatics and Cognition 15 (1): 65–90. Goldberg, Adele 1995 Constructions. A Construction Grammar approach to argument structure. Chicago: University of Chicago Press. Hall, Alison 2008 Free enrichment or hidden indexicals? Mind and Language 23 (4): 426–456. Heal, Jane 1997 Indexical Predicates and Their Uses. Mind 106: 619–640. Horn, Laurence R. 1984 Toward a new taxonomy for pragmatic inference: Q-based and Rbased implicature, In Meaning, form, and use in context: Linguistic applications, Deborah Schiffrin (ed.), 11–42. (GURT ‘84). Georgetown University Press, Washington. Jaszczolt, Katarzyna 2002 Default semantics: foundations of a compositional theory of acts of communication. Oxford: OUP. Kasher, Asa 1976 Conversational Maxims and Rationality, In Language in Focus, Asa Kasher (ed.), 197–216. Dordrecht: Reidel, Keller, Rudi 1995 Rationalität, Relevanz und Kooperation. In Implikaturen, Frank Liedtke (ed.), 5–18. Tübingen: Niemeyer, Levinson, Stephen C. 2000 Presumptive meanings. The Theory of Generalized Conversational Implicatures. Cambridge: MIT Press. Levinson, Stephen C., and Niclas Burenhult 2009 Semplates: A new concept in lexical semantics? Language 58 (1): 153–172. Martí, Luisa 2006 Unarticulated Constituents Revisited, Linguistics and Philosophy 29 (2): 135–166. Millikan, Ruth Garrett 2004 Varieties of Meaning. The 2002 Jean Nicod Lectures. Cambridge: MIT Press.

Pragmatic templates and free enrichment

205

Neale, Stephen 2000 On being explicit. Comments on Stanley and Szabó, and on Bach, Mind and Language 15: 284–294. Perry, John 1998 Indexicals, Contexts, and Unarticulated Constituents, In Computing Natural Language, Dag Westerstahl, Atocha Aliseda, and Rob van Glabbeek (eds.), 1–11. (CSLI Publications). Stanford. Récanati, François 2002 Unarticulated Constituents. Linguistics and Philosophy 25: 299–2 45. Schank, Roger C., and Robert P. Abelson 1975 Scripts, Plans, and Knowledge. In Proceedings of the 4th Joint Conference on Artificial Intelligence, Vol.1: 151–157. Searle, John Rogers 1976 A classification of illocutionary acts. Language in Society 5/1, 1-23. Sperber, Dan, and Deidre Wilson 1995 Relevance. Communication and Cognition. Oxford: Wiley - Blackwell. Stanley, Jason 2000 Context and Logical Form. Linguistics and Philosophy 23: 391–434. Stanley, Jason, and Szoltan G. Szabó 2000a On Quantifier Domain Restriction. Mind and Language 15: 219–2 61. 2000b Reply to Bach and Neale. Mind and Language 15: 295–298. Tomasello, Michael 2008 The Origins of Human Communication. Cambridge/Mass.: MIT Press. Van Vanlin, Robert D., and Randy LaPolla 1997 Syntax. Structure, meaning and function. Cambridge: CUP. Wittgenstein, Ludwig 1958 Philosophical Investigations. Oxford: Blackwell.

Pragmatic enrichment in adjectival passives: the case of the post state reading Helga Gese

1. Introduction Until recently, the discussion of so-called “adjectival” or “stative” passives such as (1) was centred on the morphosyntactical status of the participle (Rapp 1996; Maienborn 2007). The discussion was particularly vivid in the linguistic literature on German as here, unlike in English, adjectival passives can be clearly differentiated from verbal ones: The two types of passives are built using two different verbs. Whereas the verbal passive in (3) combines a form of the auxiliary werden (‘to become’) with a past participle, the adjectival passive takes a form of sein (‘to be’) instead, cf. (2). In German, sein occurs as a copula and as an auxiliary. In recent years a consensus has been reached on analyzing the verb sein in adjectival passives as a copula combining with an adjectivized participle, cf. e.g. Rapp (1996); Maienborn (2007); Stolterfoht et al. (2009), Gese et al. (2011) for arguments in favor of this adjectival analysis. (1) The door was closed for years. (2) Die Tür ist The door is.COP ‘The door is closed.’

geschlossen. closed. (adjectival passive)

(3) Die Tür wird The door becomes.AUX ‘The door is closed.’

geschlossen. closed. (verbal passive)

Having clarified the morphosyntactic status of the participle the discussion moved on to semantic issues. Departing from the observation that adjectival passives have two interpretations, a more qualitative one and a more temporal one exemplified by the two continuations in (4a) and (b), the question was how to account for these two different readings.

208

Helga Gese

(4) Der Schatz ist vergraben... The treasure is buried ... ‘The treasure is buried...’ a. … und nicht versenkt. ... and not sunk ‘…and not hidden on the ocean floor.’ (target state reading) b. …, jetzt können wir fliehen. ... now can we escape ‘...now we can make our escape.’ (post state reading)

There are basically two competing accounts which are characterized by the different role they assign to the two readings in (4a) and (4b). Whereas the lexicalist account defended by Kratzer (2000) treats the difference between the two readings as a case of homonomy, Maienborn’s (2009) underspecification account of adjectival passives describes the readings as contextually driven and locates the difference in pragmatics. The main aim of this article is to argue for a pragmatic account of the post state / target state distinction. I will present the results of a self-paced reading study that falsify the lexicalist predictions concerning post state readings and speak in favor of a semantic underspecification analysis. In the last part of the paper I will embed the analysis of the two readings in the recent discussion concerning event kind reference in adjectival passives. I will argue for the claim that, while adjectival passives semantically do not refer to event particulars but to event kinds, post state readings serve to pragmatically introduce an event particular. 2. Two accounts for the two readings One of the most prominent lexicalist approaches dealing with the two readings is Kratzer’s (2000) analysis. Kratzer treats the difference between the qualitative, so-called target state reading in (4a) and the temporal, post state reading in (4b) as a case of homonymy. The two readings in (4a) and (4b) are built by applying one of the two different Ø-affixes in (5) and (6) to the verbal participle yielding one of the two different meanings for the adjectival participle, (5b) and (6b) respectively. (5) a. Ø-affix (target state): λR λs ‫ ׌‬e [R(s, e)] b. λs ‫ ׌‬e [bury(def-treasure, e) & below-ground(def-treasure, s) & cause(s, e)]

Pragmatic enrichment in adjectival passives

209

(6) a. Ø-affix (post state): λP λt ‫ ׌‬e [P(e) & τ(e) < t] b. λt ‫ ׌‬e [bury(def-treasure, e) & τ(e) < t]

To build an adjectival passive with target state reading a participle with a lexically given result state is converted into an adjective by applying the Ø-affix in (5a), i.e. by existentially quantifying over its event argument. After application of the affix only the lexically provided target state argument of the verbal base remains available for composition. Target state adjectival passives formed by means of the affix under (5) imply that the state s resulting from the occurrence of the verbal event e holds at the reference time. They can only be built with verbal bases that lexically provide such a result state s, i.e. with a subgroup of accomplishment verbs. According to Kratzer these target state verbs can be identified by adding immer noch (‘still’) to the resulting adjectival passives. If immer noch modification is possible we are dealing with a target state reading permitting verbal base, cf. (7). The unavailability of immer noch modification on the other hand shows that the verbal base is restricted to forming adjectival passives with post state reading such as (8). (7) Die Reifen sind immer noch aufgepumpt. The tires are still uppumped ‘The tires are still inflated.’ (8) Das Theorem ist (*immer noch) The theorem is (*still) ‘The theorem is proven.’

bewiesen. proven

Verbs without a lexically given and compositionally available result state cannot be stativized by the affix under (5). These verbs only build socalled resultant or post state adjectival passives. For a post state adjectival passive formed on the basis of the Ø-affix in (6a) to be true it suffices that the event denoted by the verbal base of the participle took place before the reference time t – regardless of whether the subject referent is still in the result state of this event (note that the result state s does not occur in the post state affix at all). As the affix in (6) involves an aspectual operator, post state adjectival passives formed by means of this affix have a perfective semantics. They are “less adjective-like“ (Kratzer 2000:14). For years, the lexicalist account that directly links the semantics of adjectival passives to the result state of the verbal base was undisputed. It was shared by different authors such as Rapp (1996); Kratzer (2000); Gehrke (2011). Recently, however, Maienborn (2009) questioned the direct

210

Helga Gese

link between the lexical entry of the verbal base and the state expressed by an adjectival passive and argued in favor of a semantic underspecification account of adjectival passives. In her analysis adjectival passives ascribe an event-based but pragmatically inferred ad hoc property to their subject referent. This means that the state referred to by an adjectival passive sentence results from the event denoted by its verbal base but, at the same time, is not fully specified by it. Full specification is done by pragmatics, on the basis of context and world knowledge. Consider (9) and (10) as an example for such an ad hoc property ascription that varies with context. Without context or in a context like (9), the property denoted by the participle geöffnet (‘opened’) doesn’t differ a lot from the lexically given property OPEN plus the information that this property is caused by an openingevent. But in a context like (10) geöffnet has a certain ad hoc flavor: it means something like WITHOUT-ITS-ORIGINAL-TAGS-AND-PACKAGING. (example due to Maienborn 2011:325). (9)

Das Fenster ist nicht zu es ist geöffnet. The window is not closed it is opened ‘The window is open.’

(10)

[In a classified ad:] Der Karton ist zwar geöffnet, aber die The box is indeed opened but the Legosteine sind unbespielt und absolut neuwertig. Lego brick s are unplayed and absolutely like new ‘Even though the box is without its original tags and packaging, the Lego pieces have not been played with and are as good as new.’

What is special in Maienborn’s account is that contextual information is included in the semantic meaning constitution of adjectival passives. In formal semantics this is modeled by means of a free context variable, Q, in the adjectival Ø-affix in (11), where Q stands for the property that holds of the subject referent x in state s (this state being the result of the event e denoted by the verbal base of the participle). An adjectival passive sentence such as (10) is thus interpreted as (12), with Q needing further specification via contextual input. As an example the specifications provided by the contexts in (9) and (10) are given in (13) and (14). (11)

Ø-affix: λP λx λs ∃e [Q(x, s) & result(e, s) & P(e)] (Maienborn 2009: 44)

Pragmatic enrichment in adjectival passives (12) (13) (14)

211

‫ ׌‬s ‫ ׌‬e [Q(def-box, s) & result(e, s) & open(def-box, e)] Q = OPEN Q = WITHOUT-ITS-ORIGINAL-TAGS-AND-PACKAGING

In Maienborn’s analysis, the value assigned to the free variable Q depends on the context – and here the different readings come into play: If the context introduces an alternative, contrasting property Q’ which by contrast helps to further specify Q, this yields the qualitative, target state reading, cf. (15). If on the other hand the context introduces an alternative, temporally preceding state s’ in which the property Q did not yet hold for the subject referent, cf. (16), this gives rise to a post state reading. Q is then specified as a new property or, in a job-is-done context, as a new property resulting from a job completion. (15) (16)

...contrast (s, s’) & Q’(x, s’) ...contrast(s, s’) & ¬Q(x, s’) & s’ < s

(Maienborn 2009:46)

Let’s look at (17) uttered in the contexts (19) and (22), to spell out how the contextual specification works. The meaning of (17) without contextual specification, i.e. with free context variable is given in (18): The sentence is interpreted as referring to a state s caused by a shaking event in which the vodka martini has a certain property Q, with Q needing further specification. The contexts (19) and (22) offer such specifications for Q. (17)

(18)

Der Wodka-Martini ist geschüttelt. The vodka martini is shaken ‘The vodka martini is shaken.’ ‫ ׌‬s ‫ ׌‬e [Q(def-vodka martini, s) & result(e, s) & shake(def-vodka martini, e)]

The first sort of context, the target state context in (19), introduces the properties GERÜHRT (‘stirred’) and KLAR (‘clear’) as being the properties of an ideal vodka martini. The customer complains that his vodka martini does not have these properties but that it is shaken. In this context, geschüttelt (‘shaken’) is interpreted as CLOUDY or FROTHY, cf. the contextually introduced properties in (20) and the value assignment in (21). The sentence receives a target state reading in which the subject referent, the vodka martini, is evaluated as being cloudy and frothy, thus as having the wrong property.

212

Helga Gese

(19)

[Customer complaint in a cocktail bar:] Das weiß doch jeder: Nur gerührte This knows but everyone Only stirred Cocktails bleiben schön klar. Der Wodkacocktails remain beautifully clear The vodkaMartini ist aber geschüttelt! Martini is but shaken ‘Everybody knows that: Only stirred cocktails remain clear. But the vodka martini is shaken.’ (20) ...contrast (s, s’) & clear(def-vodka martini, s’) (21) Q = CLOUDY-FROTHY

The same sentence uttered in a typical post state or job-is-done setting receives a different interpretation: If, e.g. a waitress asks a barkeeper whether he has already prepared the vodka martini and the barkeeper answers (20), he thereby conveys that the vodka martini has a new property: that it is ready to be served. Here, the contrast is not between alternative properties but between alternative states which are temporally ordered, cf. (23). The adjectival passive sentence receives a post state interpretation and the free variable is interpreted as in (24). (22) [uttered by a barkeeper:] Der Wodka-Martini ist geschüttelt. Er kann The vodka martini is shaken. It can jetzt serviert werden. now served become. ‘The vodka martini is shaken. It can be served now.’ (23) ...contrast(s, s’) & ¬ ready-to-be-served(x, s’) & s’ < s (24) Q = NOW-READY-TO-BE-SERVED

To sum up: Kratzer’s lexicalist account links the interpretation of an adjectival passive sentence directly to the result state lexically provided by its verbal base. If a verb does not lexically characterize a result state it can only be stativized by the post state affix. The resulting post state adjectival passive sentence receives a purely temporal, perfective interpretation – it is “less adjective-like” (Kratzer 2000:14). Rejecting this homonomy account, Maienborn (2009) proposes a uniform semantic analysis for the two readings of adjectival passives. In her theory, an adjectival passive sentence ascribes a semantically underspecified property to its subject referent. The

Pragmatic enrichment in adjectival passives

213

specification of the property is done by pragmatics and only at this level do the two readings come into play. 3. Post state readings and the status of the result state: The current study We now turn to a prediction of the two analyses that allows us to experimentally decide between them. Both accounts of the semantics of adjectival passives predict that it should be possible to use an adjectival passive without referring to the result state lexically provided by its verbal base. But they do so for different reasons and with different restrictions: Kratzer’s formalization of the post state affix does not make reference to result states at all; it only contains a predication over times. Therefore it should be possible to utter a post state adjectival passive, e.g. as a report of a job completion, even if the result state of the completed event no longer holds. To take our example from above: (25) should be true if a successful hiding of the treasure took place regardless of whether the treasure has since been digged out again, cf. (26). In these cases what the speaker reports “is merely that the job assigend to him is done” (Kratzer 2000:11). Adjectival passives with verbal bases that do not lexically provide a result state should always receive such a purely temporal interpretation. (25)

Ich habe meine Aufgabe erledigt: Der I have my job fulfilled The ist vergraben! is buried ‘I have done my job: The treasure is buried.’

Schatz treasure

(26)

Der Schatz ist vergraben! Mir doch egal The treasure is buried Me but equal dass ihn jemand schon wieder ausgegraben hat! that it someone already again digged.up has ‘I don’t care that someone has dug it out again.’

In Maienborn’s account, a property ascription contradicting the verbal result state is, in principle, also possible. But it has to be strongly supported by context. Without contextual support the property contained in the verbal result state is the most economic and therefore optimal candidate for the value of the free context variable Q, cf. Maienborn (2009:45f.). An

214

Helga Gese

example for a legitimizing context was given in (10) above. In such a context, which does not necessarily have to be a post state context, it should be possible to explicitly contradict the lexically given result state, cf. (27) adapted from Maienborn (2011:325) in which the lexically provided property OPEN is contradicted by the contextually given property SEALED: (27)

[In a classified ad:] Der Karton ist zwar geöffnet, aber sorgfältig The box is indeed opened but carefully wieder zugeklebt. Die Legosteine sind unbespielt again sealed The Lego bricks are unplayed und absolut neuwertig. and absolutely like new ‘The box is without its original tags and packaging, but it has been carefully sealed again. The Lego pieces have not been played with and are as good as new.’

To test whether adjectival passives indeed allow for an interpretation without reference to the verbal base’s result state I conducted a self-paced reading study with sensicality judgment task. The study tested how the processing and acceptability of adjectival passives can be affected by a context contradicting the result state provided by their verbal bases. It compared adjectival passives, verbal passives and copula sentences with genuine adjectives in result state-contradicting contexts such as (28). A second set of items presented adjectival passives and verbal passives formed from verbs of creation with the same sort of result statecontradicting continuation, cf. (29). (28)

Causative verbs (with lexically given result state):

(AdjP)

Der Brief ist geöffnet,... The letter is opened (VerbP) Der Brief ist geöffnet worden,... The letter is opened become (Adj) Der Brief ist auf,... The letter is open ...aber er ist schon wieder zugeklebt. but it is already again sealed ‘The letter is open/opened but it is already sealed again.’

Pragmatic enrichment in adjectival passives (29)

215

Verbs of creation (without result state): (AdjP) Das Feuer ist entfacht, The fire is lit (VerbP) Das Feuer wurde entfacht, The fire became lit ... aber es brennt schon nicht mehr. but it burns already not more. ‘The fire was lit but it does not burn any more.’

Verbs of creation were chosen for two reasons. First, without lexically including a result state they nevertheless have a conceptually salient, typical result: the coming into existence of the verbal theme referent. Second, as in most cases it is not very informative to assert the mere existence of an object, adjectival passives with verbs of creation standardly receive a temporally enriched (and therefore more informative) job-is-done interpretation. The hypotheses were formulated separately for each set of items, causative verbs and verbs of creation. 3.1. Hypotheses 1 and 2: Causative verbs (with lexically given result state) As both the underspecification account and the lexicalist account permit an interpretation of adjectival passives without reference to the lexically given result state the first hypothesis is formulated as follows: (H1)

Adjectival passives with result state-contradicting continuations should cause less processing difficulty and receive more sensible responses in the sensicality judgment task than the clearly contradictory items with genuine adjectives.

In contrast to verbal passives that do not refer to result states at all, adjectival passives with causative verbal bases might preferably receive an interpretation in which the property characterized by the verbal result state is ascribed to their subject referent. In Maienborn’s underspecification account such a preference for the assignment of a lexically provided property to the free variable Q is explicitly built in: In searching for a value for the free variable Q, the best/most economic instatiation for it is the one that gets by with the fewest contextually not li-

216

Helga Gese

censed additional assumptions. If the conceptual knowledge associated with the verb’s event referent happens to already specify a resulting property, this will of course be the best choice for Q. In this case, there is no need to draw further inferences and derive more remote ad hoc property candidates – unless the context explicitly forces us to. (Maienborn 2009:45f.)

In the experimental sentences, the lexically provided property is not compatible with the property ascription in the second conjunct of the sentence. This forces the readers to look for an alternative ad hoc property. In our example (10) from above, the context facilitates the search for such an alternative value for the underspecified variable Q. In such a context, a contradiction of the lexically given result state is possible, cf. (27). However in the experiment, the sentences offer no context that would facilitate the search for an alternative ad hoc property. Therefore, the search for a result state-contradiction permitting value for Q should be difficult. This difficulty should be reflected in the reading times and the sensicality judgments, cf. (H2). (H2)

Adjectival passives with result state-contradicting continuations should cause more processing difficulty and receive fewer yes responses in the sensicality judgment task than the clearly acceptable items with verbal passives

3.2. Hypothesis 3: Verbs of creation (without result state) Adjectival passives formed on the basis of verbs of creation have a clear post state flavour. This is shown by the fact that clear post state contexts enhance the acceptability of adjectival passives with such verbs, cf. the contextless sentence (30) vs. its version (31) in which the post state interpretation is made overt by adding the adverbial endlich (‘finally’) and by a typical post state continuation. (30)

?Der Film ist gedreht. The movie is shot ‘The movie is shot.’

(31)

Der Film ist (endlich) gedreht, jetzt kann ich The movie is (finally) shot, now can I mich dem Marketing widmen. me the marketing dedicate ‘The movie is (finally) shot, now I can attend to the marketing.’

Pragmatic enrichment in adjectival passives

217

As verbs of creation do not lexically characterize a result state they clearly fail Kratzer’s criterion for target state permitting verbal bases: Adjectival passives with verbs of creation verbal bases do not allow modification by immer noch (‘still’). The lexicalist account therefore predicts that target state passive formation is impossible for this group of verbs. The adjectival passives formed on the basis of verbs of creation only permit for a purely temporal post state interpretation that does not refer to result states at all. The result state-contradicting continuation in the experimental items should therefore cause no problems. By contrast, the underspecification account analyzes all adjectival passives as property ascriptions i.e. as referring to states. The event denoted by a verb of creation has no lexically given but a conceptually salient result state: the existence of the verbal theme referent. The underspecification account therefore predicts that EXISTENCE is standardly chosen as the result state referred to by an adjectival passive formed on the basis of a verb of creation. If the underspecification account is right, the result state-contradicting continuation in the experimental items should cause processing difficulties and low sensicality ratings in the adjectival passive conditions, cf. (H3): (H3)

We should find longer reading times and fewer sensible responses for ‘verbs of creation’ adjectival passive items than for the corresponding verbal passive items.

3.3. Method Thirty undergraduate students from Tübingen University participated and received a monetary reimbursement. All were native speakers of German. Materials consisted of thirty experimental sentences in three conditions (APASS vs. VPASS vs. ADJ) and twelve experimental sentences in two conditions (APASS vs. VPASS). The forty-two experimental sentences were combined with sixty filler sentences covering a range of sentences with a form of sein (‘to be’), werden (‘to become’) and haben (‘to have’) plus a participle or a genuine adjective. For the thirty ‘causative verbs’ items in three versions, adjectival and verbal passives were formed from causative accomplishments such as zähmen (‘to tame’), füllen (‘to fill’), lockern (‘to loosen’), cf. (28) above. In condition ADJ the genuine adjective referred to the lexical result state of the verbal bases used in condition VPASS and APASS (e.g. gezähmt – zahm (‘tamed’ – ‘tame’). For the twelve ‘verbs of creation’ items in two versions, adjectival and verbal passives formed on the basis of verbs of

218

Helga Gese

creation were chosen, e.g. bauen (‘to build’), gründen (‘to found’), herstellen (‘to fabricate’), cf. (29). Experimental sentences were of type X, aber Y (‘X but Y’) where X was in condition APASS an adjectival passive sentence, in condition VPASS a verbal passive and in condition Adj a copula sentence with a genuine adjective. In conditions VPASS and APASS the second conjunct Y contradicted the result state of the verbal base from which the verbal or adjectival passive was formed (e.g. ...gezähmt, aber schon wieder scheu (‘...tamed, but already shy again’). In the filler sentences the conjunct chosen was in twenty-for cases aber (‘but’) and in thirty-six cases und (‘and’). To ensure that participants did not anticipate a contradiction in the second conjunct, filler items did not contain contradictions. 3.4. Design and Procedure Six presentation lists were constructed in which the forty-two experimental items were randomly mixed with the sixty filler items. The lists were counterbalanced across items and conditions: Each participant saw only one version of each experimental sentence. Sentences were presented in a selfpaced fashion on a PC using the E-Prime software (Psychology Software Tools, Inc.). After each sentence the participants were asked to judge the sensicality of the sentence by answering the question “War das eine sinnvolle Aussage?” (‘Did this utterance make sense?’) with yes or no. 3.5. Results: Reading times Reading times were analyzed seperately for the two sets of items. For the analysis, the reading times of the second conjunct starting with aber (‘but’) were merged and a two-step procedure was employed to eliminate outliers: The analysis excluded reading times that were shorter than 800 ms or longer than 8000 ms as well as reading times that were more than 2.5 standard deviations from the participant’s mean in that condition. This led to less than 1 % data loss. The remaining reading times were submitted to two separate repeated measures ANOVAs – one with an error term that was based on participant variability (F1) and one with an error term that was based on item variability (F2). The results are presented in table 1 for the two sets of items.

Pragmatic enrichment in adjectival passives

219

Table 1. Mean reading times and standard deviations for the second conjunct in ms.

type of verbal base

sentence type

causative verbs

APASS VPASS ADJ 2977.50 (790.32) 2606.80 (593.03) 2633.63 (590.47)

verbs of creation

2871.57(757.41) 2637.60(817.21)

3.5.1. Causative verbs (with lexically given result state) The statistical analysis revealed a highly significant main effect of sentence type [F1(2,58) = 11.143, p1 ≤ .001; F2(2,58) = 12.584, p2 ≤ .001] and the planned comparisons found significantly longer reading times for the second conjunct in adjectival passives than in verbal passives [F1(1,29) = 12.582, p1 ≤ .001; F2(1,29) =16.996, p2 ≤ .001]. Contrary to our hypothesis (H1) reading times for the second conjunct weren’t significantly longer but significantly shorter in condition ADJ than in condition APASS [F1(1,29) = 13.418, p1 ≤ .001; F2(1,29) = 15.231, p2 ≤ .001]. 3.5.2. Verbs of creation (without result state) Similarly, for the set of verbs of creation items, reading times were longer in the adjectival passive condition than in the verbal passive condition. The difference was marginally significant in the subject analysis but reached full significance in the item analysis [F1(1,29) = 3.604, p1 = .068; F2(1,11) = 6.225, p2 ≤ .05]. 3.6. Results: Sensicality judments Table 2. Mean percent of yes responses in the sensicality judgment task (with standard deviations)

type of verbal base

sentence type

causative verbs

APASS 39.33 (18.18)

VPASS 89.33 (10.81)

verbs of creation

51.11 (30.93)

92.78 (10.43)

ADJ 12.33 (13.31)

220

Helga Gese

3.6.1. Causative verbs (with lexically given result state) The analysis of the answers in the sensicality judgment task revealed a highly significant main effect of sentence type [F1(2,58) = 238.119, p1 ≤ .001; F2(2,58) = 154.229, p2 ≤ .001]. Adjectival passives with continuations contradicting the result states of their verbal bases received fewer yes responses to the sensicality question than the corresponding verbal passives. But compared to the clearly contradictory items in condition ADJ adjectival passives were judged sensible more often. Both differences were highly significant [sensicality APASS < VPASS: F1(1,29) = 137.658, p1 ≤ .001; F2(1,29) = 107.673, p2 ≤ .001; sensicality APASS > ADJ: F1(1,29) = 71.827, p1 ≤ .001; F2(1,29) = 29.874, p2 ≤ .001]. 3.6.2. Verbs of creation (without lexically given result state) The set of verbs of creation items displayed the same pattern of answers in the sensicality judgment task as the set of causative verbs items (in condition APASS and VPASS): Adjectival passives with result statecontradicting continuations were judged sensible significantly less often than the corresponding verbal passives [F1(1,29) = 45.502, p1 = ≤ .001; F2(1,11) = 77.830, p2 ≤ .001]. 3.7. Discussion The results of the experiment partly confirm (H1): The significant differences in the sensicality judgments between the APASS-items and the clearly contradictory ADJ-items show that adjectival passives, in principle, allow for an interpretation without reference to the lexically given result state: Adjectival passives with result state-contradicting continuation in the second conjunct are not clearly nonsensical (39% sensible judgments for adjectival passives vs. 12% for the corresponding genuine adjective sentences). Interestingly, we found opposite differences in the reading times between condition ADJ and condition APASS: Readings times were significantly longer in APASS than in ADJ and there was no difference in the reading times between the clearly nonsensical ADJ-items and the clearly sensible VPASS-items. This result may be explained by Bott’s (2010) observation that clear semantic mismatches only lead to very local processing difficulties and to a speed-up in reading times at the end of the sentence. As the critical region in the experiment comprised the entire second part of

Pragmatic enrichment in adjectival passives

221

the sentence starting with the conjunct aber (‘but’) it is possible that a very local effect had no significant influence on the reading times of the entire region. The difference in sensicality between the APASS-items and the ADJitems was predicted by both the underspecification and by the lexicalist account. By contrast, the results concerning (H2) clearly speak in favor of the underspecification account: They show that adjectival passives and verbal passives differ with respect to the semantic relevance of the result state of their verbal bases. Whereas contextual negation of the result state does not affect the sensicality of a verbal passive, adjectival passives show a preference for referring to the lexically given result state: The contextual negation of the lexically provided result state causes processing difficulties and leads to fewer sensible judgments in the APASS-items compared to the VPASS ones (89% sensible judgments for verbal passives vs. only 39% for adjectival passives, significantly longer reading times for adjectival passives than for verbal ones.). This preference for the choice of a lexically provided property over an alternative ad hoc property was predicted by the underspecification account. It is explained by the optimality principle guiding the search for a value for the free variable: “In searching for a value for the free variable Q, the best/most economic instatiation for it is the one that gets by with the fewest contextually not licensed additional assumptions“ (Maienborn 2009:45). Kratzer’s lexicalist account, in principle, is compatible with a preference for one of the two readings, the purely temporal, result statecontradiction permitting post state reading or the qualitative, result-state denoting target state one – but in her theory, such a preference needs to be argued for. In the underspecification account, the preference for the reading in which the lexically provided property is ascribed to the subject referent is already builtin. The results of the study confirm the assumption of such a preference. While the results for the causative verbs items (with lexically given result state) are, in principle, compatible with both the underspecification as well as the lexicalist account, this is not the case for the verbs of creation items (without lexically given result state). The results for these items clearly contradict the predictions of the lexicalist account and support an underspecification analysis: We found longer reading times and fewer sensible responses for the APASS-items formed from verbs of creation than for the corresponding VPASS-items. Verbs of creation clearly fail Kratzer’s criterion for target state permitting verbal bases. The lexicalist account therefore predicts that they only form post state adjectival passives

222

Helga Gese

with purely temporal interpretation. These should be compatible with the context presented in the second conjunct of the experimental items in the same way as verbal passives are compatible with it. The fewer sensible responses and the processing difficulties observed for the verbs of creation items in condition APASS compared to the corresponding VPASS-items show that adjectival passives formed from verbs of creation do not receive a purely temporal interpretation. On the other hand, the contextually forced interpretation without reference to the typical result state of a creation event, namely the state in which the referent of the verbal theme argument exists, is not totally odd: The results showed 51% sensible responses in condition APASS. This is compatible with Maienborn’s (2009) underspecification account. This account predicts that the conceptually salient result property EXISTENCE is standardly chosen as value for the underspecified property. But, context permitting, an alternative property ascription may be chosen. In the self-paced reading experiment, the second conjunct of the experimental sentences is incompatible with the conceptually salient result state. It therefore forces the participants to look for an alternative value assignment for the underspecified property. But the experimental items did not offer any extra context facilitating this search. This could explain that the search for a suitable ad hoc property was successful only in 51% of the cases. What is important for us is the difference between adjectival passives and verbal ones: The results show that, unlike verbal passives formed on the basis of verbs of creation, the corresponding adjectival passives are not fully compatible with the result state-contradicting context in the second conjunct of the experimental sentences. To sum up, while the confirmation of (H1) supports both the underspecification and the lexicalist account of adjectival passives, two results of the study speak in favor of an analysis relying on semantic underspecification: First, the lexicalist account would have to resort to additional assumptions concerning the preferences between target state and post state interpretations to explain the results concerning (H2). Second and most importantly, the predictions of the lexicalist account concerning adjectival passives formed on the basis of verbs of creation without a lexically given result state were falsified: Verbs of creation do not form post state adjectival passives in Kratzer’s sense, i.e. they do not receive a purely temporal perfect-like interpretation, cf. (H3). A last question to be answered concerns the post state flavor of adjectival passives formed from verbs of creation: At first glance, the observation that, even without further context, adjectival passives with a verb of creation basis standardly receive a post state or job-is-done interpretation

Pragmatic enrichment in adjectival passives

223

seems to contradict the underspecification account. This account predicts that the (underspecified) property ascribed to the subject referent has to result from the occurrence of the event denoted by the verbal base. Yet, the state typically resulting from a creation event is the state in which the theme of the creation event exists. So why not just choose EXISTENT as the property ascribed to the subject referent? Well, this property might simply not be informative enough. In fact, existence is already given by the existence presupposition of the definite article. To find the optimal candidate for the value assignment, the property thus has to be further specified. As the most obvious specification of existence is done along the temporal dimension, a temporally enriched ad hoc property, something like EXISTENT-AS-OF-LATE, is chosen. The preference of verb of creation adjectival passives for post state readings can thus be explained without resorting to the lexicalist account and without assuming a perfect-like reading of adjectival passives. In the remaining part of the paper I want to briefly explore the consequences of such an analysis of post state readings and embed it in a new version of the underspecification account supplemented with event kind reference. 4. Pragmatic enrichment in post state readings Recent semantic investigations of adjectival passives (Gehrke 2011; Gese 2011) have shown that the event underlying adjectival passive formation cannot be a concretely instantiated event particular but that it has to be a generically interpreted event, i.e. an event kind. Whereas Gehrke’s analysis is based on a lexicalist account of adjectival passives supplemented with event kind reference I adapt Maienborn’s underspecification account of adjectival passives to event kind reference. Besides data on event related modification and experimental data that call into question the presence of an event particular in the semantics of adjectival passives the discourse structural properties of adjectival passives fit better with reference to event kinds: In fact, in adjectival passives neither an event particular nor its agent participant is available in discourse, as the following three observations show. First, bridging to agents is easy with eventive passives but difficult with adjectival ones: Mauner (1996:248f.) presented evidence from a self-paced reading study for processing difficulties on the indefinite bridging-pronoun they following adjectival passive sentences such as (32) compared to eventive passives like (33). As event kinds, contrary to event particulars, do not

224

Helga Gese

include agent arguments, the absence of a discourse structural accessible implicit agent speaks in favor of the event kind reference of adjectival passives. (32) The rebel priest was profoundly tormented for days. ???They wanted him to reveal where the insurgents were hiding out. (33) The rebel priest was being tormented for days. They wanted him to reveal where the insurgents were hiding out.

Second, bridging to events is difficult with adjectival passives but bridging to event kinds does not cause problems. In (34) the weak bridging-pronoun das (‘that’) anaphorically refers to an event particular in the preceding sentence (cf. Asher and Lascarides 1998 on weak bridgingpronouns). The deviance of (34) proves the absence of an event particular in adjectival passives. In (35) das is disambiguated by the kind selecting predicate Art des Haarefärbens (‘kind of hair-dyeing’) to referring to an event kind (cf. Chierchia 1995 for kind-selecting predicates). Bridging to this event kind is possible: (34)

Die Haare sind blondiert. ??? Das hat The hair are bleached-blond That has gestunken. smelled ‘The hair is bleached. ???That smelled terrible.’

fürchterlich terribly

(35)

Die Haare sind blondiert. Das ist die The hair are bleached-blond That is the beliebteste Art des Haarefärbens. popular.COMP kind of hair-dyeing ‘The hair is bleached. This is the most popular way of dyeing one’s hair.

Third, unlike verbal passives adjectival ones cannot control into purpose clauses, cf. (36), i.e. they do not contain an implicit agent to control the PRO-subject of the purpose clause. In this respect, adjectival passives pattern like copula sentences with genuine adjectives that clearly do not contain an implicit agent argument, cf. Gese (in preparation) for experimental evidence for this claim. (36)

???Der Briefumschlag war geöffnet, um den The envelope was opened to the Inhalt kontrollieren zu können. content control to can ???’The envelope was opened to allow an inspection of the content.’

Pragmatic enrichment in adjectival passives

225

In the light of these discourse structural data it is plausible to assume that the property denoted by the participle in an adjectival passive does not depend on an event particular but on an event kind (note that event kinds, unlike event particulars, do not have to specify their participants, e.g., their agent arguments). The affix proposed in (38) therefore adapts Maienborn’s adjectival affix repeated here as (37) to event kind reference. In (38) the event predicate P provided by the verbal base is type-shifted into the event kind nP by Link’s (1995) type-shifting operator n, which transforms predicates into kinds. The property Q then has to be specified as belonging to a certain kind of property, nQ, that can be conceptualized as resulting from the event kind nP. (37) (38)

Ø-affix (Maienborn 2009: 44): λP λx λs ∃e [Q(x, s) & result(e, s) & P(e)] Ø-affix (with kind reference): λP λx λs [Q(x, s) & result(nP, nQ]

For an example consider sentence (17) from above, repeated here as (39). Applying the affix (38) yields the interpretation (40): The sentence refers to a state s in which the underspecified property Q holds of the vodka martini where Q has to be characterizable as belonging to a certain kind of properties, ↑Q, resulting from a certain kind of events, namely shakingevents. Dependent on context, Q then is specified as a target state or a post state property, cf. (20)-(21) and (23)-(24), repeated here as (41)-(44). (39)

Der Wodka-Martini ist geschüttelt. The vodka martini is shaken ‘The vodka martini is shaken.’

(40) ‫ ׌‬s [Q(def-vodka martini, s) & result(↑shake, ↑Q) (41) ...contrast (s, s’) & clear(def-vodka martini, s’) (42) Q = CLOUDY-FROTHY (43) ...contrast(s, s’) & ¬ ready-to-be-served(x, s’) & s’ < s (44) Q = NOW-READY-TO-BE-SERVED

What are the consequences of assuming event kind reference in adjectival passives for the difference between post state and target state readings? As we have seen above, post state adjectival passives ascribe a property to their subject referents that is interpreted as being a “new property”. This in turn entails a change of state: A new property, per definitionem, is a property that did not hold before the actual state. Post state readings thus imply a change-of-state causing event particular temporally situated before the actual state. As such an event particular is not present in the semantics

226

Helga Gese

of adjectival passives it has to be pragmatically inferred. By requiring the instantiation of the event kind, post state readings thus lead to a pragmatic enrichment of adjectival passive sentences. Let’s quickly run through our discourse structural diagnostics from above to substantiate this claim. First, bridging to agents, cf. (45), or to event particulars, cf. (46), is easy for adjectival passives in post state reading. This was not the case with the adjectival passives without post state context, cf. the difference in acceptability between (46) and (34) from above. To trigger a specification along the temporal dimension yielding a post state reading, a verb of creation or a temporal modification with endlich (‘finally’) is used; see the explanation at the end of section 3. (45) Der Film ist endlich gedreht. Der Regisseur The film is finally shot. The director ist zufrieden mi t seinem Werk. is satisfied with his work. ‘The film is finally shot. The director is satisfied with his work.’ (46)

Die Haare sind endlich (fertig) blondiert. The hair are finally (completely) bleached-blond Das hat fürchterlich gestunken. That has terribly smelled ‘The hair is finally bleached. That smelled terrible.’

Second, and most obviously, adjectival passives in post state reading are able to control into purpose clauses. The deviance of adjectival passives with purpose clauses disappears if we add a post state inducing context in the preceding sentence, cf. the difference between (36) from above and (47). In (48) the same effect is due to the use of a post state inducing verb of creation as the base of the adjectival passive. (47) Nun aber raus aus dem Now but out from the Die Haare sind schließlich The hair are after all zu werden! to become ‘But now leave the hairdresser’s bleached to be admired.’ (48) Der Eintrittspreis ist The entrance fee is schließlich gedreht, after all shot

Frisiersalon und ab in die Disco: hairdresser’s and of in the disco: blondiert, um bewundert bleached-blond to admired and go to the disco. After all, the hair is

angemessen. Ein Film fair. A film um damit Geld to with it money

ist is zu verdienen! to earn

Pragmatic enrichment in adjectival passives

227

‘The entrance fee is fair. After all, the film was shot to make money with it.’

The increase in acceptability from (36) to (47) and (48) was empirically corroborated by the results of two self-paced reading studies: The first experiment showed longer reading times for the purpose clause introducing conjunction um (‘to) in adjectival passives than in verbal ones and no significant difference compared to copula sentences with genuine adjectives (which clearly do not contain an implicit agent). The effect disappeared in a second experiment that tested verbal passives and adjectival passives with post state inducing verbs of creation as their bases. The results of the studies show that in post state readings of adjectival passives the implicit agent argument contained in the pragmatically inferred event particular is able to pragmatically establish PRO-control. For space limitations, the details of the two studies will have to be reported elsewhere (Gese in preparation). Still, the data concerning the discourse structural status of the event and its agent participant in (45) – (48) allow for the following conlusion: While adjectival passives semantically do not refer to event particulars but to event kinds, post state readings serve to pragmatically enrich the interpretation by introducing an event particular. 5. Conclusion The paper presented a self-paced reading study that allowed us to experimentally test the claims made by two competing analyses of the post state / target state distinction in adjectival passives. The results of the study disconfirmed the lexicalist assumption: The existence of a reading in which adjectival passives receive a purely temporal interpretation without reference to a lexically or conceptually provided result state could be experimentally ruled out. Given these results, I suggested an underspecification analysis of adjectival passives, where adjectival passives, in both readings, are property ascriptions, i.e. they refer to states. The two readings then turn out to be strategies in the search for an optimal specification of the property denoted by the participle of an adjectival passive: Dependent on contextual information, the semantically underspecified property can be specified either along the temporal dimension or along the qualitative dimension. In the first case, the resulting adjectival passive has a post state interpretation, in the second case it is interpreted as a target state adjectival passive (cf. Maienborn 2009). Moreover, the present paper shed some light on a recent

228

Helga Gese

semantic discussion concerning event kind reference in adjectival passives: Data on the discourse status of the event in adjectival passives showed that post state readings serve to pragmatically enrich the interpretation by introducing an event particular. It is this pragmatically inferred event particular which makes the post state interpretation of adjectival passives a ‘more temporal one’. Acknowledgements Work on this paper was supported by the German Science Foundation (DFG) as part of the project A1 “The semantics and pragmatics of combinatory meaning variation” within the collaborative research center SFB 833, Universität Tübingen. I would like to thank my supervisor, Claudia Maienborn, the members of A1, and the audience of Beyond the Words for helpful discussion and comments. References Asher, Nicholas, and Alex Lascarides 1998 Bridging. Journal of Semantics 15: 83–113. Bott, Oliver 2010 The processing of events. Amsterdam: John Benjamins Publishing Company. Chierchia, Gennaro 1995 Individual-level predicates as inherent generics. In The Generic Book, Gregory N. Carlson and Francis Jeffry Pelletier (eds.), 176– 223. Chicago: University of Chicago Press. Gehrke, Berit 2011 Stative passives and event kinds. In Proceedings from Sinn und Bedeutung 15, Ingo Reich, Eva Horch, and Dennis Pauly (eds.), 241–2 57. Saarbrücken: Saarland University Press. Gese, Helga in prep. Empirische Untersuchungen zum Zustandspassiv Ph. D. diss., Deutsches Seminar, Universität Tübingen. 2011 Events in adjectival passives. In Proceedings from Sinn und Bedeutung 15, Ingo Reich, Eva Horch, and Dennis Pauly (eds.), 259–273. Saarbrücken: Saarland University Press.

Pragmatic enrichment in adjectival passives

229

Gese, Helga, Claudia Maienborn, and Britta Stolterfoht 2011 Adjectival conversion of unaccusatives in German. Journal of Germanic Linguistics 23: 101–140. Kratzer, Angelika 2000 Building statives. In Proceedings from the 26th Meeting of the Berkeley Linguistic Society, Lisa J. Conathan, Jeff Good, Darya Karitskaya, Alyssa B. Wulf, and Alan C.L. Lu (eds.), 385–399. Berkeley: University of California, Berkeley Linguistics Society. Link, Godehard 1995 Generic information and dependent generics. In The Generic Book, Gregory N. Carlson and Francis Jeffry Pelletier, 358–382. Chicago: Chicago University Press. Maienborn, Claudia 2007 Das Zustandspassiv. Grammatische Einordnung – Bildungsbeschränkung – Interpretationsspielraum. Zeitschrift für Germanistische Linguistik 35: 83–114. 2009 Building event based ad hoc properties: on the interpretation of adjectival passives. In Proceedings from Sinn und Bedeutung 13, Arndt Riester and Torgrim Solstad, 35–49. Stuttgart: University of Stuttgart. 2011 Strukturausbau am Rande der Wörter: Adverbiale Modifikatoren beim Zustandspassiv. In Sprachliches Wissen zwischen Lexikon und Grammatik. Institut für Deutsche Sprache, Jahrbuch 2010, Stefan Engelberg, Anke Holler, and Kristel Proost, 317–343. Berlin/New York: Mouton de Gruyter. Mauner, Gail 1996 The role of implicit arguments in sentence procession. Ph.D. diss., University of Rochester. Rapp, Irene 1996 Zustand? Passiv? – Überlegungen zum sogenannten “Zustandspassiv”. Zeitschrift für Sprachwissenschaft 15: 231–265.

Pragmatic inferencing and expert knowledge Petra B. Schumacher and Jörg Meibauer

1. Introduction In experimental pragmatics, there is a need to strictly separate lexical knowledge from contextual or encyclopedic knowledge. We therefore investigated the contribution of expert language – as one instantiation of encyclopedic knowledge – to a particular area of pragmatic reasoning, namely inferencing. We sought to determine whether pragmatic inferencing relies on encyclopedic knowledge or is a highly automatic operation, and which cognitive processes are affected by the computation of an expertise-based inferential relation. In the following, we first elaborate on the role of different knowledge sources for pragmatic processing, before we present a time-sensitive event-related brain potentials (ERPs) study. The division between lexical, contextual, and encyclopedic information become for instance apparent, when testing children’s competence related to promises, where the lexical meaning of the verb to promise, contextual knowledge (who promises what to whom), and encyclopedic knowledge (it is bad to break a promise) has to be taken into account (Bernicot and Laval 2004). When interpreting experimental results, certain effects could have to do with insufficient lexical learning, or with a lack of contextual or encyclopedic knowledge. The representation of a context, more often than not in the form of a little story, certainly is helpful in making the task easier for the participants, but is risky in presupposing certain knowledge that may not be accessible to participants independently of the story context. Moreover, it appears that where no explicit context is given, test persons tend to construe a more or less fitting context, and accordingly, it is hard to control for all aspects of knowledge that may contribute to a judgment. With regard to several experimental designs used in experimental pragmatics, it can be shown that contextual or encyclopedic knowledge related to lexical knowledge matters. For instance, in experimental research dealing with scalar implicatures, children and adult controls were asked whether the utterance “Some giraffes have long necks” is true (cf. Noveck 2001; Noveck and Posada 2003). “Logical” children reacted with “yes” (even if all giraffes had long necks, it is at least true that some have

232

Petra B. Schumacher and Jörg Meibauer

long necks), whereas “pragmatic” adults answered “no” (it would be underinformative to answer yes, because, as far as they know, all giraffes have long necks). But it seems that it is encyclopedic knowledge that matters here. What, if children were not more logical, but more cautious in comparison to the adults? After all, there might exist some giraffes (e.g., a certain species or baby giraffes) that have short necks indeed. Even some adults were apparently unsure about the pragmatic adequacy of the utterances. While 89 % of the responses made by children aged 7-8, and 85% of the responses by children aged 10-11 revealed “logical” agreement, 59 % of the responses of the “pragmatic” adults reflected disagreement. However, 41% of the adult responses indicated child-like behavior (i.e. agreement). More generally, then, one might want to ask whether and to what extent expert knowledge influences the generation of pragmatic inferences. In recent discussions between Neogriceans and Relevance Theorists, the question of how context influences pragmatic inferencing is crucial. Part of experimental research focusing on scalar implicatures is motivated by the attempt to decide between two competing pragmatic models, the theory of Generalized Conversational Implicature (GCI) (Default view) and Relevance theory (RT) (Context Driven view)(cf. Noveck and Sperber 2007; Geurts 2010: 83–103). While researchers such as Bott and Noveck (2004) conclude that the derivation of scalar implicatures is highly contextdependent and does not show the autonomy claimed by default approaches such as Levinson’s (2000), this does not necessarily render the notion of a GCI totally superfluous. After all, a GCI might be a type of conversational implicature that is recurrently derived in a number of contexts, as perceived by hearers. There may still be room for a pragmatic theory that views GCIs as “regularities of use that, despite being systematic, should not be confused with linguistic meanings”, as Bach (2007: 27) put it. Furthermore, when reasoning about the automaticity (or default) of deriving pragmatic inferences vis-à-vis the influence of context and contextually evoked knowledge, the problem of identifying and separating lexical and associated encyclopaedic knowledge of the participants has widely been neglected. Usually, it is simply presupposed that participants have the necessary contextual and encyclopedic knowledge that enables them to derive pragmatic inferences. If they succeed, the results are always taken to be related to their lexical knowledge (what they know about the meaning of a lexeme, for instance some and all) plus the ability to use Gricean maxims, but in most cases not as an effect of applying contextual or encyclopaedic knowledge to the respective task. We suspect, however, that the relevant knowledge could very well matter.

Pragmatic inferencing and expert knowledge

233

We therefore designed a study where expert knowledge is directly investigated with regard to pragmatic inferencing. We decided not to replicate designs dealing with scalar implicature where the (often subtle) literal meaning of the respective trigger is at stake, but to focus on lexemes of an expert language where the literal meaning of target lexemes is clearly defined within a scientific expert system. In particular, we present data from event-related brain potentials (ERPs) to examine how rich context must be and whether expert knowledge can impact pragmatic inferencing. To this end, we investigated inferential processing in short passages consisting of two sentences and varied the validity of the inferential relation (correct vs. false) and expert-specific knowledge (from linguistics vs. chemistry). An advantage of this design is that the contextually relevant knowledge is clearly related to the lexical meanings of certain lexemes. Experts have a direct access to the relevant lexical knowledge. For non-experts, there is a lack of respective knowledge. Since even experts may go wrong, and even non-experts may be happy in deriving correct relations, we are able to study how lexical knowledge becomes active in the task of deriving correct inferential relations. A further advantage of this study, as compared with research related to testing GCI- versus RT-approaches, is that there is no direct connection to Gricean maxims, e.g., the Maxim of Quantity (see, however, Huang 2010). While the Maxim of Relevance may play a role, this assumption is not necessary for anaphoric binding. 1.1. Electrophysiological measures Previous investigations of discourse processing, and pragmatic inferencing in particular, have identified two central electrophysiological signals (for an overview see Schumacher 2009): First, a negative brain potential peaking around 400 ms after stimulus-onset (N400) indexes lexical-semantic processing and is a reflection of expectations built up by previous discourse, the immediate situation of utterance, and most crucially for present purposes, lexical, encyclopedic and contextual knowledge. The amplitude of the N400 decreases with increasing predictability (cf. e.g. Kutas and Federmeier 2000) and is inversely related to a word’s frequency of occurrence and familiarity (for an overview see Kutas, van Petten and Kluender 2006). Second, a later positive deflection with an onset latency around 600 ms after stimulus-onset (which we here refer to as Late Positivity) reflects discourse updating costs, arising from either the introduction or the reorganization of information units in discourse representation (cf. Burkhardt

234

Petra B. Schumacher and Jörg Meibauer

2006, 2007). The underlying processes are labeled Discourse Linking and Discourse Updating respectively (as for instance sketched out in Schumacher 2009). Research on inferential processing has compared the comprehension of an inference-based anaphor – such as “the professor” in “Tim attended a lecture in Berlin. He said that the professor was very nice.” – to the processing of a coreferential anaphor – as in “Tim visited a professor in Berlin. He said that the professor was very nice” (e.g. Burkhardt 2006). The ERP data revealed a more enhanced N400 for inferred relations compared to repeated coreference relations, reflecting differences in the lexical fit between the critical expression and prior context, and a more pronounced Late Positivity for inferred expressions, reflecting costs from the introduction of a corresponding new discourse referent and/or the encoding of the corresponding discourse relation (e.g., elaboration, narration — as outlined in Asher and Lascarides 1983). These findings and follow-up investigations indicated that lexical and contextual knowledge facilitates the early Discourse Linking stage (i.e., coreference links are easier than inferential links). Subsequently, discourse-internal maintenance demands for the introduction of an independent discourse representation corresponding to “the professor” in the inferential case, resulting in more enhanced Discourse Updating costs. While there is ample evidence in the literature for co-textually induced N400-modulations (e.g., van Berkum, Hagoort and Brown 1999; Burkhardt 2006), it has also been shown that encyclopedic knowledge (what speakers know about the world) can modulate the N400. Hagoort et al. (2004) reported N400-effects for world knowledge violations (“Dutch trains are white.”); van Berkum and colleagues (2008, 2009) found N400-effects to manipulations of stereotypical knowledge (a child / adult uttering “Every evening I drink some wine before I go to bed.”) and hearer-specific values (a conservative person reading “I think euthanasia is an acceptable / unacceptable course of action”). We therefore predicted that individual expertise, as part of hearer-specific knowledge, should affect the processes underlying the N400, since it facilitates lexical retrieval required for linking an incoming word to the wider discourse. As far as processes of pragmatic inferencing are concerned that are required for discourse-internal operations, two hypotheses must be tested. On the one hand, expert knowledge may not only license linking processes, but also pragmatic inferencing, and successful inferencing would thus be reflected in a Late Positivity effect. On the other hand, inference drawing may be a rather automatic process that occurs independent of encyclopedic

Pragmatic inferencing and expert knowledge

235

knowledge, because the addressee assumes correct and cooperative behavior on the part of the speaker; such an automatic process should result in no Late Positivity differences as a function of expertise. Following this line of reasoning, the generation of an inference should be closely tied to the associated evaluative processes, and previous research has reported modulations of positive potentials as a function of task demands. For instance, when required to judge an antonymy relation (“black – white”) or to carry out a sensicality task on a sentence expressing an antonymy (“The opposite of black is white”.), highly expected continuations (“white” vs. “yellow”) registered a positive deflection (P300), reflecting the ease of evaluating the critical antonym (Roehm et al. 2007; Kretzschmar 2010). Task-dependent positivities have further been observed for expected completions of idioms and collocations in figurative expressions (Vespignani et al. 2010; Molinaro and Carreiras 2010). It thus appears possible that the evaluation of an inferential relation has consequences for discourse-internal operations. More generally, expertise has been shown to influence cognitive mechanisms, for instance in studies on various aspects of music processing and rhythmic performance comparing trained musicians to laypersons (e.g., Krampe et al. 2000; Koelsch, Schmidt and Kansok 2002; Regnault, Bigand and Besson 2001) or object recognition by experts for birds or dogs (Tanaka and Curran 2001). The general tenor has been that training and experience enhance expert-specific skills and modulate electrophysiological responses. Expertise can further be considered a function of the amount of training and experience, which may account for certain developmental changes as well (cf. e.g., Carey 1992 for findings of face recognition improving with age). This corroborates the idea that the source of the divergent performance observed between children and adults may be more general in nature and represent lack of experience, which makes children more cautious in certain decisions. Expertise may thus also influence inferential processing, and in particular both the processes underlying the Discourse Linking and the Discourse Updating stage. Expectations built up on the basis of encyclopedic knowledge should be guided by individual knowledge, i.e. expert-based relations should be more familiar and hence easier to retrieve than nonexpert relations, resulting in a more pronounced N400-amplitude for nonexpert over expert relations. In addition, if the construction and maintenance of discourse representation structure is driven by task-specific instructions, assessing an expertise-supported inference might be easier and hence engender a more pronounced Late Positivity.

236

Petra B. Schumacher and Jörg Meibauer

2. The Current Study 2.1. Methods 2.1.1. Participants Twenty-eight students from the University of Mainz (15 women) participated in this investigation. Half of them majored in linguistics and half of them in chemistry. All participants were monolingual native speakers of German. Their ages ranged from 19-29 years (mean age: 23.1), they were right-handed (assessed by a German version of the Edinburgh handedness test) and reported normal or corrected-to-normal visual acuity. Five participants had to be discarded from the analysis because of poor performance in the expert knowledge test assessed after EEG recording; an additional two participants were excluded from the analysis because they showed excessive ocular artifacts. 2.1.2. Materials One hundred and sixty pairs of stimuli were constructed with the factors expert knowledge (linguistics/chemistry) and validity of the inferential relation (correct/false). Eighty sets each were constructed on the basis of expert knowledge from linguistics and chemistry respectively. Examples are provided in (1) and (2) below. Context sentences introduced a particular subfield (e.g., phonology, stereoisomers) and target sentences referred to a specific aspect of this area, allowing for a correct, expertise-based inferential link in the (a)-cases below (e.g., phonology - sonority, stereoisomers - chirality). For the false relation ((b)-cases), an illegitimate pairing was constructed (e.g., semantics - sonority, sodium chloride – chirality). The materials were constructed and checked for accuracy by graduate students in the two disciplines. The stimuli were distributed across lists, so that each participant saw 80 stimuli from the linguistics set and 80 from the chemistry set, representing 40 correct and 40 false relations each. Additional 160 filler trials were constructed and interspersed with the 160 critical trials, including coherent and incoherent mini-discourses, as well as some mini-discourses requiring inferences based on general world knowledge (e.g., accident-casualty).

Pragmatic inferencing and expert knowledge

237

(1a) Linguistics knowledge – Correct inferential relation: Flora hat gestern mit ihrer Freundin über Phonologie geredet. Lange Zeit haben sie über die Sonorität diskutiert. ‘Yesterday, Flora talked with a friend about phonology. For a long time, they discussed the sonority.’ (1b) Linguistics knowledge – False inferential relation: Flora hat gestern mit ihrer Freundin über Semantik geredet. Lange Zeit haben sie über die Sonorität diskutiert. ‘Yesterday, Flora talked with a friend about semantics. For a long time, they discussed the sonority.’ (2a) Chemistry knowledge – Correct inferential relation: Tim will sein Wissen über Stereoisomere vertiefen. Am Freitag will er über Chiralität recherchieren. ‘Tim wants to deepen his knowledge on stereoisomers. On Friday, he wants to explore chirality.’ (2b) Chemistry knowledge – False inferential relation: Tim will sein Wissen über Kochsalz vertiefen. Am Freitag will er über Chiralität recherchieren. ‘Tim wants to deepen his knowledge on sodium chloride. On Friday, he wants to explore chirality.’ The participants’ task was a sensicality judgment. They first had to indicate as quickly and accurately as possible whether the two sentences fitted, by pressing a yes- or no-button on a gamepad. Next, they were asked to indicate their certainty on a four-point scale by pressing one of four buttons. Following the ERP recording, each participant’s knowledge in their field of studies was assessed in a questionnaire, in which all correct and false pairs of words from the ERP study were presented and participants were asked to rate the connection between the two critical words (e.g., phonology – sonority) on a 7-point-scale. 2.1.3. Procedure Participants were instructed to read the material for comprehension and to respond to a sensicality and a certainty judgment task at the end of each trial. Stimuli were presented visually on a computer screen in yellow letters against a blue background. Each trial began with a fixation star displayed for 500 ms and followed by a blank screen of 150 ms. Each stimulus was

238

Petra B. Schumacher and Jörg Meibauer

then displayed in segments (single words for 400 ms and two word segments for 450 ms), with an intersegment blank screen of 150 ms. Following a blank screen of 150 ms, questions for the two tasks were presented and participants were required to respond as quickly and accurately as possible by pressing a button on a gamepad. Following a 500 ms blank screen, the next trial started. After participants were prepared for the experiment, they completed a short training session to get acquainted with the experimental procedure and were then presented with eight blocks of 40 trials each. After recording, participants filled out a questionnaire that assessed their knowledge of linguistic and chemistry-based relations on a 7-point-scale. 2.1.4. Electrophysiological Measures and Preprocessing The electroencephalogram (EEG) was recorded from 27 Ag/AgCl scalp electrodes mounted in an elastic cap (ground: AFz; reference: left mastoid). Signals were rereferenced offline to linked mastoids. Electroculograms (EOGs) were recorded by means of two sets of electrode pairs placed at the outer canthus of each eye and above and below the participant’s right eye. Electrode impedances were kept below 5 kΩ. All EEG and EOG channels were amplified and digitized with a rate of 500 Hz. The EEG data were bandpass-filtered offline (0.3-20 Hz) to exclude slow drifts. Furthermore, EOG rejections were performed automatically (±40 mV) and manually. In addition to trials with excessive artifacts, trials that registered an incorrect or timed-out response to the sensicality task were also excluded from the ERP data analysis. 2.1.5. Data Analysis The ERP analyses are based on trials that registered correct responses to the sensicality task. Grand average ERPs were time-locked to the onset of the critical NP (the sonority/the chirality in (1) and (2)). Repeatedmeasures analyses of variance (ANOVAs) were performed with the factors Expertise (2 levels: expert vs. non-expert) and Validity (2 levels: correct vs. false inferential relation) for lateral and midline electrodes separately. The lateral analyses included the factor region of interest (ROI) (with 4 levels: left anterior (F3/FC1/FC5/C3), right anterior (F4/FC2/FC6/C4), left posterior (CP1/CP5/P3/P7), right posterior (CP2/CP6/P4/P8)) and the midline analysis included the factor ELECtrode (with 2 levels: anterior (Fz/FCz/Cz), posterior (CPz,Pz,POz)). All analyses were carried out in a

Pragmatic inferencing and expert knowledge

239

hierarchical manner on the mean amplitude value per condition, i.e., only significant interactions (p < .05) were resolved. To avoid excessive type 1 errors from violations of sphericity, we applied Huynh and Feldt correction when the analysis involved factors with more than one degree of freedom in the numerator. Analyses were carried out for the time windows between 300 and 450 ms (N400) and between 500 and 650 ms (Late Positivity). 2.2. Results 2.2.1. ERP Data Figure 1 illustrates main effects of EXPERTISE, reflected in a more pronounced negativity between 300 and 450 ms for non-expert knowledge and a Late Positivity starting around 500 ms for expert knowledge. These patterns were confirmed by statistical analyses. The ANOVA in the time range from 300-450 ms post-onset revealed a main effect of EXPERTISE (lateral: F(1,20)=8.19, p