Lexical Pragmatics - Journal of Semantics

Reinhard Blutner 121(c) a fast book(d) a fast driver[one that can be read in a short time][one who drives quickly]Examples of this kind suggest that the adjective modifies a specificconceptual component connected with the noun, namely its purpose orfunction. With regard to this component the adjective seems to make aunitary contribution; it qualifies this component (the act of moving, typing,reading or driving) in a specific and unitary way. In the general case theprinciple of pragmatic compositionality says that it is possible to decomposethe lexical items in conceptual components and that these componentsdetermine the conceptual interpretation of the whole expression. In somecases it seems that the conceptual components of a lexical item areassociated semantically with it, i.e. those components determine the meaningof the lexical item. In other cases the association of the conceptualcomponents is via general world knowledge and this information isdetachable from the word meaning. The former case is exemplified bythe information concerning the purpose or function of artefact terms.The latter case may be illustrated by the information determining thedesignation of the spatial properties of spatial object terms (e.g. Lang 1989).As an exercise that demonstrates a typical case of the notion ofcompositionality just under discussion let us calculate the 'conceptualinterpretation' of the expression fast car. In (3 a) the semantic analysis ofthe noun car is sketched in some relevant aspects. The analysis states that theconcept related with cars is characterized (beside other things) by a telic role(purpose or function) that qualifies a situation type s associated with cars asa moving process. The semantic analysis of the adjective fast is given in (b)expressing that this adjective affects the telic role only. In (c) the expressiongiven in (a) and (b) are combined by the intersection operation, and in (d) theresulting interpretation (a car that moves quickly) is obtained by unifyingthe free variables.(3) (a) car. Ax [CAR(x) A TELIC(x,s) A MOVE(s) A . . .](b) fast:Ax [TELIC(x, s') A FAST(s')](c) fast car. Ax [CAR(x) A TELIC(x,s) A MOVE(s) ATELIC(x,s') A FAST(s')...](d) unification ~> Ax [CAR(x) A TELIC(x,s) A MOVE(s) AFAST(s)A...]Downloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014It is a straightforward how the analysis can be extended to the other casesgiven in (2).It seems to me that this kind of analysis works well in cases like (2). Theproblems of the account become visible, however, when we consider othertypes of adjectives, for example colour and taste adjectives. Suppose that we

122 Lexical Pragmaticswant to describe that a red apple is one whose surface is red (but notnecessarily its inside), and a red grapefruit is one having a red inner volume(but not necessarily a red surface). According to the account just sketchedwe can try to describe this by assuming an application condition for redsaying that a salient part of the object is wholly reddish: Ax [SAL-PART(x,y') A RED(y')]. Furthermore, we have to postulate that the salientpart of an apple is its surface and the salient part of a grapefruit is its innervolume. However, what counts as salient part with regard to colour is notnecessarily salient with regard to other aspects. What counts as the salientpart of an apple with regard to taste, for example, seems to be the innervolume and not the peel. Thus, according to the view under discussion wehave to change the semantic entry for red to something like that shown in(4a), whereas the entry for sweet would look like (4b). Furthermore, in orderto make the procedure pragmatically compositional, we would have topostulate a conceptual analysis for apple as sketched out in (4c).(4) (a) red: Ax [SAL-PART COLOUR (x, y') A RED(y')](b) sweet: Ax [SAL-PART TASTB (x,y') A SWEET(y')](c) apple: Ax [APPLE(x) A SAL-PART TASTE (x,y,) A PULP(y,) ASAL-PART COLOOR (x,y 2 ) A PEEL(y 2 ) A . . .]Although such an 'analysis' gives the right input for satisfying pragmaticcompositionality, it doesn't describe the data more systematically than thefunctional view sketched before. Obviously, both views describe the dataonly by enumerating them. What can be learned from this observation issimple that compositionally always can be retained, but at the expense ofsystematicity. As a matter of fact, in the former case the notion of semanticcompositionality is saved by enumerating the adjectives' applicabilityconditions for different objects. In the second case, on the other hand,the notion of pragmatic compositionality is saved by enumerating thesalient parts of objects with regard to any aspect (colour, taste . . .). In thiscase, the cumbersome has to do with the relational notion of saliencewhich obviously cannot be analysed in a way which is systematic andcompositional at the same time.Another issue which conflicts with a systematic compositional treatmentconcerns the fact that the colour vocabulary gets reduced when applied tothings that come in a limited range of colors (e.g. red/white wine). A curiousobservation is that the same colour sometimes gets expressed in differentways, depending on what it contrasts with. For example, in Japanese,aka-zatoo 'brown sugar' (lit 'red sugar') comes in the same range of coloursas shira-miso, lit. 'white bean paste'. 1Next, let us consider a problem which I will call the problem ofpragmatic anomaly. In (5) the examples (a-d) seem to make correct (butDownloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014

Reinhard Blutner 123not necessarily true) statements about a conceivable state of affairs. Theexamples (e-h), on the other hand, are somehow defective, but they aredefective for different reasons: (g) and (h) represent true category violations,while (e) and (f) represent the phenomenon of pragmatic anomaly:(5) (a) The tractor is red(b) The tractor is defective(c) The tractor is loud(d) The tractor is gassed up(e) ?The tractor is pumped up(f) ?The tractor is sweet(g) *The tractor is pregnant(h) *The tractor is bald-headedThe importance of this distinction has been noted, for example by Keil(1979). Category mistakes can be explained on grounds of an ontologicalcategory violation as described by Sommers (1959, 1963) and Keil (1979).Pragmatic anomaly, on the other hand, has only indirectly to do with theso-called ontological level describing the basic categories of existence interms of which we conceptualize our everyday world. That sweet is not anappropriate attribute of tractors can't be explained on grounds of anontological category violation. A tractor can be sweet, by the way. Tasteone: it might surprise you.The problem of pragmatic anomaly has to do (i) with the distinctionbetween category mistakes and pragmatic anomaly and (ii) with the formaltreatment of the latter kind of deviance. Both the phenomena of categoryviolations and pragmatic anomaly are intimately based upon the conceptualinformation associated with lexical items. As a consequence, both kinds ofphenomena should find a proper place within Lexical Pragmatics. Sincealmost nothing is known about the proper treatment of pragmatic anomaly,this phenomenon seems to be an especially exciting challenge for LexicalPragmatics.A third problem that Lexical Pragmatics has to deal with concerns thephenomenon called lexical blocking. This phenomenon has been demonstratedby a number of examples where the appropriate use of a givenexpression formed by a relatively productive process is restricted by theexistence of a more 'lexicalized' alternative to this expression. One case inpoint is due to Householder (1971). The adjective pale can be combinedwith a great many colour words: pale green, pale blue, pale yellow. However,the combination pale red is limited in a way that the other combinations arenot. For some speakers pale red is simply anomalous, for others it picksup whatever part of the pale domain of red pink has not pre-empted. ThisDownloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014

Reinhard Blutner 125the structure of the (mental) lexicon may be derived. As an example for thiskind of research I refer to seminal work of Caramazza & Grober (1977).These authors investigated twenty-six senses of the word line. Clusteringand scaling analyses revealed five major groupings of these senses for whichclear descriptions can be given. Distinguishing a core meaning level fromthe level of conceptually salient senses, Caramazza & Grober have proposedEXTENSION as the underlying abstract core meaning of line and they havepresumed 'instruction rules' of the form REALIZE X AS Y (where X and Ycorrespond to semantic representations) in order to produce the abstractmeanings of the five clusters. Thereafter, the application of subsequentinstructions produce the senses realized in specific contexts. As an exampleof how the instruction rules might work, Caramazza & Grober (1977)consider the derivation of the sense of line in draw a line under the title of thebook. The 'linguistic dictionary' correlates the sound part of line with its coremeaning EXTENSION. Using the instruction REALIZE EXTENSIONAS UNIDIMENSIONAL EXTENSION, an output corresponding to acertain cluster of concrete senses would be realized. Applying a furtherinstruction, say REALIZE UNIDIMENSIONAL EXTENSION AS VISUALPERCEPTIBLE, would help to isolate the intended surface sense.Though investigations of that kind may be extremely instructive andinteresting from a psycholinguistic point of view, they leave an importantpoint out of consideration: the isolated investigation of singular lexical unitscloses one's eyes to certain regularities and restricting conditions that mayarise alone from investigating the semantic relations between different butin some aspects similar lexical units.The lexicologist must show the relationships between all the senses of a lexical item (a taskwhich conventional dictionaries have done well) and also the relationships of related sensesof different lexical items (a task which recently linguists have begun doing), but few studiesattempt to do both of these tasks (Lehrer 1978: 95).Following this methodological insight, Lehrer has investigated cookingwords (Lehrer 1968), temperature words (Lehrer 1970), and sensory wordsfor taste, smell, feel (Lehrer 1978), and she has suggested the followinghypothesis:Downloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014If there is a set of words that have semantic relationship in a semantic .field (where suchrelationships are described in terms of synonymy, autonymy, hyponymy, etc., and if one ormore items pattern in another semantic field, then the other items in the first field areavailable for extension to the second semantic field. Perceived similarity is not necessary(Lehrer 1978: 96).As a necessary condition on semantic transfer—the transfer of words with ameaning in one domain to another—this principle is certainly interesting.However, as Lehrer suggested herself, not each potential transfer of

126 Lexical Pragmaticsmeaning that satisfies the principle is realized and becomes permanent inthe usage of the speech community. As an example let us consider theEnglish touch word sharp which transfers to taste, smell, and sound.However, not each touch word related to sharp in the touch domain maytransfer to the other domains. Blunt, for example, may transfer neither totaste nor to smell. Consequently, the quoted hypothesis cannot be thewhole story. There must be some kind(s) of extra conditions whichdetermine which kind of potential transfers of meaning remain in thelanguage and which disappear.Cast in Caramazza's & Grober's (1977) framework of instruction rules, itis the question of restrictions on rules that becomes important. Why is theapplication of an instruction rule possible in the case of sharp but impossiblein the case of blunt? And what is the theoretic status of these restrictions?Are they reflecting only idiosyncratic properties of the lexicon? Or are theymore systematic and perhaps related to properties of the conceptualdomains under discussion? The complex of questions raised in thisconnection I call the restriction problem of polysemy. Lexical Pragmaticssuggests that part of the restriction problem can be treated in terms ofpragmatic restrictions.In the previous subsection I discussed two further problems in orderto demonstrate some challenges for Lexical Pragmatics: the fallacy of(pragmatic) compositionality and the problem of handling lexical blocking.Both problems have interesting consequences within the domain ofsystematic polysemy as well. Let's consider first lexical blocking. Take thewell-known phenomenon of 'conceptual grinding', whereby ordinary countnouns get a mass noun reading denoting the stuff the individual objects aremade of, as in There is fish on the table or There is dog all over the street. Thereare several factors that determine whether 'grinding' may apply, and, morespecific, what kind of 'grinding' (meat grinding, fur grinding, universegrinding . . .) may apply. Some of these factors have to do with theconceptual system, while others are language-dependent (cf. Nunberg &Zaenen 1992; Copestake & Briscoe 1995).Lexical blocking is a language-dependent factor. It refers to the fact thatthe existence of a specialized item can block a general/regular process thatwould lead to the formation of an otherwise expected interpretationequivalent with it. For example, in English the specialized mass termspork, beef, wood usually block the grinding mechanism in connection withthe count nouns pig, cow, tree. This explains the contrasts given in (6).Downloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014(6) (a) I ate pork/?pig(b) I like beef/?cow(c) The table is made of wood/?tree

128 Lexical Pragmatics(9) The Radical Underspecification View(a) Every lexical unit determines an underspecified representation (i.e. arepresentation that may contain, for example, place holders andrestrictions for individual and relational concepts).(b) The combinatorial system of language determines how lexical unitsare combined into larger units (phrases, sentences).(c) There is a system of type and sortal restrictions which determineswhether structures of a certain degree of (under)specification arewell formed.(d) There is a mechanism of contextual enrichment (pragmaticstrengthening based on contextual and encyclopedic knowledge).This inferential mechanism is controlled by cost factors and doesn'tneed triggering by type or sort violations.The mechanism of contextual enrichment carries the main burden inexplaining restrictions on interpretation. Because of its inferential character,this mechanism is structured non-compositionally. Inferential processingmust be controlled by cost factors. Such cost factors may reflect nonrepresentationalmeans as salience and relevance.* As we shall see in section 4,the idea of (radical) underspecification and contextual enrichment nicelyfits in the picture of mono tonic processing (Alshawi & Crouch 1992; vanDeemter & Peters 1996). Moreover, it is this feature of processing whichcrucially is involved in explaining pragmatic anomaly.3 LEXICAL PRAGMATICS AND THE THEORY OFCONVERSATIONAL IMPLICATUREBased on the material presented in the last section, the task of this section ismainly to evolve and propose a set of guidelines for realizing LexicalPragmatics as a proper theoretic settings. As we shall see, these guidelinesaim at a straightforward formulation of the notion of conversationalimplicature as a necessary prerequisite to develop Lexical Pragmatics.Before we are ready to speculate on what a proper treatment ofconversational implicatures might be, let me first make some remarksconcerning present accounts of conversational implicatures.For Griceans, conversational implicatures are those non-truth-functionalaspects of utterance interpretation which are conveyed by virtue of theassumption that the speaker and the hearer are obeying the cooperativeprinciple of conversation, and, more specifically, various conversational maxims:maxims of quantity, quality, relation, and manner.While the notion of conversational implicature doesn't seem hard toDownloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014

Reinhard Blutner 129grasp intuitively, it has proven difficult to define precisely. The generalityof the cooperative principle and the conversational maxims makes itdifficult to specify just which maxims are involved in particular implicatures.Essential concepts mentioned in the maxims are left undefined(what is relevance, adequate evidence, etc.). However, before we can start to'flesh out' something like the maxims, Grice's view of implicature raiseseven more basic questions. Are there just the maxims Grice mentioned, ormight others be needed (as he suggested himself)? Or could the number ofmaxims be reduced? Sperber & Wilson (1986) are an extreme case insuggesting one only, the maxim of relevance. And what is the rationalebehind the cooperative principle and the maxims? Are they norms whichspeakers and hearers must know in order to communicate adequately (asGrice and most followers suggest)? Or are they generalizations about certainforms of inferential behavior which speakers and hearers need no more toknow to communicate than they need to know the principles of digestionto digest (Sperber & Wilson's position, a position which is also adopted inthe present account).An important step in reducing and explicating the Gricean frameworkhas been made by Atlas & Levinson (1981) and Horn (1984). TakingQuantity as a starting point they distinguish between two principles, theQ-principle and the I-principle (termed R-principle by Horn 1984).* Simplebut informal formulations of these principles are as follows:(10) Q-principle: Say as much as you can (given I) (Horn 1984: 13).Make your contribution as. informative (strong) aspossible (Matsumoto 1995: 23).Do not provide a statement that is informationallyweaker than your knowledge of the world allows,unless providing a stronger statement would contravenethe I-principle (Levinson 1987: 401).I-principle: Say no more than you must (given Q) (Horn 1984: 13).Say as little as necessary, i.e. produce the minimallinguistic information sufficient to achieve your communicationalends (bearing the Q-principle in mind)(Levinson 1987: 402)Read as much into an utterance as is consistent withwhat you know about the world. (Levinson 1983: 146-7)Obviously, the Q-principle corresponds to the first part of Grice's quantitymaxim (make your contributions as informative as required), while it can beargued that the countervailing I-principle collects the second part of thequantity maxim (do not make your contribution more informative than isrequired), the maxim of relation and possibly all the manner maxims. AsDownloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014

130 Lexical PragmaticsHorn (1984) seeks to demonstrate, the two principles can be seen asrepresenting two competing forces, one force of unification minimizingthe Speaker's effort (I-principle), and one force of diversification minimizingthe Auditor's effort (Q-principle).Conversational implicatures which are derivable essentially by appeal tothe Q-principle are called Q-based implicatures. A standard example isgiven in (11). As a general characteristic, 'these implicata limit what is said byshrinking the range of possible states of affairs associated with what is said toa smaller range of those states of affairs associated with what is communicated.What is communicated is MORE DEFINITE than what is said' (Atlas & Levinson1981: 35).(11) Some of the boys are at the party—• Not all of the boys are at the partyConversational implicatures which are derivable essentially by appeal tothe I-principle are called I-based implicatures. These implicatures can begenerally characterized as 'enriching what is said by reshaping the range ofthe possible states of affairs associated with what is communicated. What iscommunicated is MORE PRECISE than what is said.' (Atlas & Levinson 1981: 36).A standard examples is given in (12).(12) John said 'Hello' to the secretary and then he smiled—> John said 'Hello' to the female secretary and then he smiled3.1 Towards the proper treatment of conversational implicatureI believe that the proper treatment of conversational implicatures cruciallydepends on the proper formulation of the Q- and the I-principle. As I willdemonstrate subsequently, such a formulation also has to account for theinterplay of these two principles and their interaction with the two qualitymaxims.Let me start with giving a more explicit formulation of the Q- and I-principle. In order to do that, we need a distinction between those aspects ofan utterance that can be described in purely linguistic (conventional) termsand a description of the intended content of the utterance. 6 Let usabbreviate the linguistic part of an utterance by a where a includesphonological, syntactic, and semantic information—indicated by phon(a),syn(a), and sem(a), respectively. For the sake of explicitness, we identifythe intended content of an utterance (better: a possible interpretativehypotheses about the content of the utterance) with a (partial) statedescription and denote it with m. 7 Furthermore, we have the generalDownloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014

132 Lexical Pragmaticspr(m|sem(a)) is the conditional probability that an instance of theproposition sem(a) is an instance of the (partial) state description m in agiven space of eventualities or possible worlds. Using a simple factorialanalysis for the total cost function c(a, m), we get the following ansatz:(13) c(a, m) = compl(a) • c(sem(a),m), where compl(a) is a positivereal number and c(sem(a),m) = — log 2 pr(m|sem(a))Now let me suggest the following first approximation to the Q- and theI-principle, respectively:(14) (a) [a, m] satisfies the Q-principle iff there is no [c/,m] £ C suchthat c(a',m) < c(a, m)(b) [a, m] satisfies the I-principle iff there is no [a, m'] 6 C such thatc(a, m') < c(a, m).In this formulation, the Q- and the I-principle constrain the set C ofpossible [sem(a), m]-pairs in two different ways. The I-principle constrainsthe set by selecting the minimal surprising state descriptions with respect to agiven semantic content sem(a) 9 and the Q-principle constrains the set byblocking those state descriptions which can be grasped more economicallyby an alternative linguistic input a'. I should also add that I have tried toformulate the Q- and the I-principle from the perspective of languagecomprehension. Due to the very symmetric formulation of the principles,switching to the production perspective may be realized simply byswitching Q and I.Before we come to a closer inspection of the formulation (14) we have tointroduce the notion of common ground in our theoretic framework and wehave to investigate the effects of the maxims of quality. According toconventional wisdom, a common ground eg is an information state thatcontains all the propositions that are shared by several participants (forexample, S and H). In more formal terms this means that an informationstate eg (non-empty set of possible worlds) counts as common ground iff foreach proposition it holds: eg |= (f>

Reinhard Blutner 13 3condition may be seen as related to the second maxim of quality (Do not saywhat you lack evidence for).If that is right, then crucial consequences of the maxims of quality andtheir very special status within the overall theory can be formulated interms of conditions on updating the common ground. Let us use theabbreviation PQ(O;) for the set of possible state descriptions that areconstraint both truth-conditionally (by means of C) and by means of theQ-principle, i.e. PQ{CX) =def { m: [ a , m ] £ C and [a, m] satisfies Q-principle}. Analogously we have the definition p\(a) =j e f {m: [a, m] G Cand [a, m] satisfies I-principle}. We simply write p(a) referring to theintersection of both conditions: p(a) =&( pq(a) C\ p,(a). Using thisnotation we can state the two conditions related to the quality maximsas follows:(15) (a) Quality 1: for each m G p(a): m is consistent with cg[a](b) Quality 2: p(a) is a non-empty set and Vp(a) holds in cg[a]"Let us call an update pragmatically licensed iff it satisfies the conditions(i5)(a, b). Now we call an utterance a pragmatically anomalous iff there is nopragmatically licensed update for it. Furthermore, a proposition (j> is called aconversational implicature of a iff eg [a] f= for each pragmatically licensedupdate. If this relationship holds for each common ground eg we may speakof generalized implicatures. Restricting the corresponding notions tospecific classes of common grounds, we may define implicatures of theparticularized variety.Let us now consider some simple examples to see how the proposedmechanism is working. First consider Moore's paradox exemplified by thecontrast between (16a) and (16b).(16) (a) The cat is on the mat, but John doesn't know it.(b) ?The cat is on the mat, but I don't know it.The absurdity of (16b) falls out straight away as a case of pragmaticanomaly. The explanation immediately results from the formulation of thequality maxims in (15) and the conditions on common grounds. To see thecrucial point, we have to show first that whenever a has a pragmaticallylicensed update cg[a], then the proposition Ks(sem(a)) must be logicallyconsistent. This assertion follows from the fact that the proposition Vp(ar)logically entails the proposition sem(o:) and the fact that Vp(a) iscontained in the common ground cg[a]. The latter directly results fromthe condition (15b). As a consequence sem(a) must be contained in cg[a].From our condition on common grounds it follows that Ks(sem(a)) mustalso be contained in eg [a] and therefore must be consistent.Hintikka (1962) calls a proposition (f> epistemically indefensible just in caseDownloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014

134 Lexical Pragmaticsthe proposition K$ is inconsistent (with regard to his epistemic logicsystem). Using this notion, we can summarize our argumentation as follows:There can be no pragmatically licensed update for a in case the propositionsem(o;) is epistemically indefensible. Utterances with epistemicallyindefensible sem(a) come out as pragmatically anomalous utterances accordingto the definition given above. It is a simple exercise to show thatexpressions of the form 0A->K s are epistemically indefensible (i.e.Ks($A->Ks) is inconsistent). Consequently, the absurdity of (16b)comes out as a case of pragmatic anomaly. Other cases of epistemicallyindefensible statements are discussed by Karttunen (1972) and Gazdar (1979)and demonstrate the importance of this subcase of pragmatic anomalies.Next, consider a simple example showing the generation of scalar andclausal implicatures. We consider the expression a = S t or S 2 and thecompeting expression a' = S 1 and S 2 and we assume that both expressionsare of the same linguistic complexity: compl(a) = compl(a'). Thederivation of the clausal and scalar implicatures of a is schematized in()(17) (a) a:

Reinhard Blutner 135true. Clearly, as in the binary case we get for any number of arguments(AND, OR) as a Horn-scale which predicts that (18a) implies (18b).(18) (a) ORfS,, S a , . . . , S n )(b) NOT AND(S,, S 2) ..., S,,)Unfortunately this prediction is too weak. The conjunction of (18a) and(18b) yields an formula which is true if any number of disjuncts smallerthan n is true. This is correct for n = 2, but wrong for more argumentssince a general account of the exclusive interpretation would have topredict the interpretation according to which it is true in case one (and onlyone) disjunct is. The utterance of (19) certain does not invite you to takeeither one or two of the items mentioned.(19) With the salmon you can have fries, rice or a baked potato.It is easy to check that the current account yields the right result. As anexample consider the case of three disjuncts a = OR(S,,S 2 ,S 3 ). Thederivation of the exclusive interpretation runs as above, but now basedon the following alternatives a' o = AND(S n S 2 ,S 3 ), a[ =AND(S,,S 2 ),a' 2 — AND(S,, S 3 ), c^ = AND(S 2 , S 3 ). Again the central point is that thestronger expressions realkes the relevant state descriptions with higherprobability than the weaker expressions thereby blocking them for theinterpretation of OR(Si, S2, S3).It should be noted that we did not include the single disjuncts amongthe alternatives. This is motivated by the independent requirement (whichany theory of Q-based implicatures has to make, but which is notoriouslydifficult to formalize) that the alternatives must contrast in view of anelement which is qualitatively similar in a relevant sense. This is a generalphenomenon. In spite of the entailment relation licensed by existentialgeneralization a proper name as 'John' does not form a contrast class with aquantifier like 'some'. 'All' being a quantifier itself does.' 3The next class of examples deals with the phenomenon of (partial) lexicalblocking. Aronoff (1976) has shown that the existence of a simple lexical itemcan block the formation of an otherwise expected affixally derived formsynonymous with it. In particular, the existence of a simple abstract nominalunderlying a given -ous adjective blocks its nominalization with -ity.(20) (a) curious-curiositytenacious-tenacity(b) furious-*furiosity-furyfallacious-*fallacity-fallacyWhile Aronoff's formulation of blocking has been limited to derivationalprocesses, Kiparsky (1982) notes that blocking may also extend to inflectionalDownloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014

136 Lexical Pragmaticsprocesses and he suggests a reformulation of Aronoff's blocking as a subcaseof the Elsewhere Condition (special rules block general rules in their shareddomain). However, Kiparsky cites examples of partial blocking in order toshow that this information is too strong. According to Kiparsky, partialblocking corresponds to the phenomenon that the special (less productive)affix occurs in some restricted meaning and the general (more productive). affix picks up the remaining meaning (consider examples like refrigerantrefrigerator,informant-informer, contestant-contester). To handle these and othercases Kiparsky (1982) formulates his general condition Avoid Synonymy citedabove.Working independent of the Aronoff-Kiparsky line, McCawley (1978)collects a number of further examples demonstrating the phenomenon ofpartial blocking outside the domain of derivational and inflectionalprocesses. For example, he observes that the distribution of productivecausatives (in English, Japanese, German, and other languages) is restrictedby the existence of a corresponding lexical causative. Whereas lexicalcausatives (e.g. (21a)) tend to be restricted in their distribution to thestereotypic causative situation (direct, unmediated causation throughphysical action), productive (periphrastic) causatives tend to pick up moremarked situations of mediated, indirect causation: For example, (21b) couldhave been used appropriately when Black Bart caused the sheriff's gun tobackfire by stuffing it with cotton.(21) (a) Black Bart killed the sheriff(b) Black Bart caused the sheriff to dieLet me now demonstrate how the theory developed so far accounts fortotal blocking. Consider two expressions a and a' that are semanticallyequivalent, i.e. C(cv) = C(cv'). In case that the expression a is less complexlinguistically than the expression a', i.e. compl(a) < compl(a'), it results(from (13)) that c(a, m) < c(a', m) for each m G C(a). Consequently, weobtain pq(a') = 0. The latter implicates p(a') = 0 and we canconclude that there is no pragmatically licensed update for a'. In otherwords, the existence of a linguistically simpler (less marked) expression aequivalent to a' has totally blocked the more complex one. When theexpressions a and a' are of comparable linguistic complexity, i.e.compl(a) = compl(a'), the result is c(a, m) = c(a', m) for eachm € C(a). From this we get PQ(CV) = PQ(

Reinhard Blutner 137expressions like icebox-refrigerator, synonymy-synonymity, persuade to not—dissuade from can coexist within a single idiolect despite their referentialequivalence.' 4 Furthermore, Horn argues that in these cases the inherentcomplexities (as demonstrated by psycholinguistic evidence) are approximatelyof the same size. Though it is not completely clear which factorsinfluence the inherent complexity of a linguistic expression, Horn'scounterexamples suggest that our prediction is of the right kind.The theory developed so far has been shown to predict that the morecomplex one of two semantically equivalent expressions must be blocked inall its interpretations. However, this prediction seems too strong and wouldconflict with examples like (21). There are several possibilities of avoidingthis conclusion. First, we could stipulate that a and a 1 would not besemantically equivalent in such cases. Essentially we had to stipulate thatthe less complex expression applies semantically to the stereotypic causativesituation only, whereas the more complex expression is not restrictedsemantically in a related way but gets its restriction (to the indirectcausative situation) by way of the Q-principle. Another way out of thedilemma of total blocking would accept the semantic equivalence ofexpressions like (2i)(a,b) but would claim that the principles (i4)(a,b)have default character only, with a preference to the Q-principle in the caseof conflicts. I think that both 'solutions' are unsatisfactory for conceptualreasons and would not really explain the general tendency that 'unmarkedforms tend to be used for unmarked situations and marked forms formarked situations' (Horn 1984: 26)—a tendency that Horn (1984: 22) calls'the division of pragmatic labor'.I think a better solution to this problem and a real explanation of 'thedivision of pragmatic labor' has to start with a reformulation of the I- andQ-principle. The informal formulation of these principles as documentedin (10) stresses a kind of partial circularity: in expressing the Q-principlereference to the I-principle has been made and vice versa. I think we have tolive with this kind of partial circularity, but at the same time we must give aprecise formulation for it in order to see its consequences. The following isan attempt in this direction:Downloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014(22) (a) [a, m] satisfies the Q-principle iff there is no [a',m] £Csatisfying the I-principle such that c(a', m) < c(a, m)(b) [a, m] satisfies the I-principle iff there is no [a, m'] G C satisfyingthe Q-principle such that c(a, m') < c(a, m).Let me now demonstrate how this explication of the Q- and I-principleexplains Horn's 'division of pragmatic labor'. Let me keep the previousnotations but give them a slightly changed content by referring to theprinciples in (22) instead of those in (14). Consider again two expressions a

138 Lexical Pragmaticsand a! that are semantically equivalent, i.e. C(a) = C(a') and let usassume furthermore that C(a) and C(a'), respectively, contain exactly twoelements mai r and m^dir of different complexity, say c(sem(a),mair) =c(sem(a / ),mdi r )

Reinhard Blutner 139implausible to assume that I have only one book or one finger. Only in theformer case can the more precise genitive form (/ slept on my book; I slept inmy car) block the corresponding interpretation. In treating the examples onthe right-hand side, we have to take into account also those interpretationswhere more than one book or one finger is involved; in this case there is noalternative expression that may block the ego-centred interpretations andthey may be selected by means of the I-principle.Summarizing, the present account of conversational implicature triesto give a real unification of the two competing 'forces' expressed by theQ-principle and the I-principle, respectively. This approach contrasts withmost recent theoretical accounts (e.g. Hirschberg 1991; Matsumoto 1995)that have focused on single classes of conversational implicatures only. Incontrast, the present account tries to address the integration of differentkinds of conversational implicatures.The main problems addressed in this paper are problems of LexicalPragmatics. Before I come to a detailed treatment of some typical examples,I want to discuss four theses that are designed to characterize LexicalPragmatics from a methodological point of view.3.2 Four theses of Lexical PragmaticsLexical Pragmatics is a research field that tries to give a systematic andexplanatory account of pragmatic phenomena that are intimately connectedwith the semantic underspecification of lexical items. The approachcombines a compositional semantics with a general mechanism of conversationalimplicature. Starting off from a underspecified semanticrepresentation, a mechanism of information enrichment (abduction) isinvoked to yield the appropriate specification with regard to thecommon ground.In section 2 the range of Lexical Pragmatics has been characterized byseveral examples, and some general, typical problems have been discussed.The present subsection tries to sharpen the rather impressionistic picturethat has evolved and attempts to illuminate it from a methodologicalperspective.Downloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014Thesis i: Lexical Pragmatics is non-compositionalIn section 2 I argued against the principle of pragmatic compositionality. Thisprinciple says that it is possible to decompose the lexical items of ancompound expression into conceptual components which combinedtogether determine the conceptual interpretation of the whole expression.

140 Lexical PragmaticsI have tried to make it clear that neither the influence of salience nor thephenomenon of lexical blocking can be approached in this way and that therelevant data should not be described only by enumeration but in a moresystematic way. The problem behind it has to do with the relational natureof salience, the existence of blocking effects, i.e. the influence of itemsthat don't occur in the expression under discussion but nevertheless areimportant to determine its interpretation, and the whole idea of inferentialreasoning.The main doctrine of Lexical Pragmatics aims at the combination of acompositional semantics with a general mechanism of conversationalimplicature. It is the second part of this doctrine that accounts for thenon-compositional character of Lexical Pragmatics. Almost everything inthe formulation of conversational implicature has this non-compositionalcharacter: both the formulation of the Q-principle as well as the I-principleare 'holistic' in addressing a whole range of alternative expressions; theconceptions of informativeness, surprise (measured in terms of conditionalprobability), and linguistic complexity are non-combinatorial and cannot bereduced to the corresponding properties of the parts of an expression; themechanism of information enrichment C as based on abductive inference(see section 4) is non-compositional.Thesis 2: Lexical Pragmatics crucially involvesnon-representational meansBeside the question 'Is Lexical Pragmatics compositional?' we have therelated question 'Is Lexical Pragmatics combinatorial?' The basic intuitionunderlying the combinatorial approach is that a cognitive activity is aprocess of manipulating representations, typically a sequential procedureconsisting of discrete steps in accordance with definite criteria. Thecombinatorial approach contrasts with the connectionist approach (e.g.Rumelhart, McClelland, & the PDP Research Group 1986), viewing thecognitive system as a network of units collected to each other through linksof various strengths. The cognitive activity in these systems consists of aparallel spread of activation instead of the combinatorial sequentialprocedure.Perhaps there are arguments against the view that lexical pragmatic iscombinatorial (cf. the discussion in Lahav 1993 with regard to thepragmatics of adjectives). However, I want to be cautious with regard tothis issue, which is difficult to decide, and I will ask a different but relatedquestion instead, namely the question of whether Lexical Pragmaticsinvolves non-representational means of manipulating representations. TheDownloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014

Reinhard Blutner 141answer to this question, I claim, is clearly affirmative. Notions such assalience, cue validity, diagnostic value, informativeness, surprise, relevance,frequency of use, and so on are candidates for such non-representationalmeans. It would be a fallacy to assume—according to a superficial reflectionof the concept of mental representation—that each quantity that is involved indetermining our mental behaviour must be mentally represented in orderto become effective. Parameters like salience and cue validity need not berepresented mentally in order to exist and to determine our cognitiveactivity. Instead, such parameters involve the non-representational dimensionof our computational system. IS With regard to the theoretical framework ofLexical Pragmatics, the whole conception ofcost function is interspersed withnon-representational means. A problem for ordinary symbolism: we need away of manipulating non-representational elements.Thesis 3: Lexical Pragmatics crucially involvesEconomy principlesEconomy principles are crucially involved in determining how nonrepresentationalparameters control the selection and suppression ofrepresentations. With Zipf (1949) as a forerunner we have to acknowledgetwo basic and competing forces, one force of unification, or Speaker'seconomy (I-principle), and the antithetical force of diversification, orAuditor's economy (Q-principle). The two opposing economies are inextreme conflict, and we have reformulated this conflict in a way thatmakes it possible to obtain definite outcomes with regard to the selection ofinterpretative hypotheses. However, I feel that there must be an independentjustification of this kind of Economy principles, perhaps onederived from the general economics involved in defining connectionistnetwork behaviour (e.g. Rumelhart, McClelland, & the PDP ResearchGroup 1986).Downloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014Thesis 4: Lexical Pragmatics has to explain whenconversational implicatures are cancellableand when notDoes cancellability constitute a necessary condition for conversationalimplicatures? Grice himself notes that cancellability doesn't hold for allkinds of conversational implicatures and mentions implicatures based onthe Quality maxim as an exceptional case. Our discussion of Moore'sparadox has demonstrated this case in the context of the present theory.Another type of conversational implicature constituting a counterexample

142 Lexical Pragmaticsagainst the claim that cancellability is necessary for conversationalimplicatures has been pointed out by Sadock (1978) in his seminalpaper attacking the usefulness of the cancellation test for conversationalimplicature.Grice states explicitly that generalized conversational implicatures, those that have little todo with context, are cancelable. But is it not possible that some conversational implicaturesare so little dependent on context that cancellation of them will result in somethingapproaching invariable infelicity? In a paper in preparation, I argue that sentences of theform almost P only conversationally entail not P, contrary to the claim made by Karttunenand Peters (1979). The implicature is straightforwardly calculable and highly nondetachablebut, unfortunately for my thesis, just about uncancelable. The sentence Gertrude notonly almost swam the English Channel, in fact she swam it is, I admit, pretty strange (Sadock1978: 293).Langendoen's (1978) analysis of reciprocals give rise to another kind ofexamples, suggesting that cancellability is not necessary for conversationalimplicatures. Langendoen assumes that the reciprocal makes a uniformsemantic contribution on every occasion of its use. This semantic contribution,he assumes, must be reflected by truth conditions included inevery, instance of use of the reciprocals. His analysis, then, rules out all butthe weakest meaning he discusses, Weak Reciprocity, as the correctmeaning of the reciprocal. However, as discussed in Dalrymple et al.(1994), Langendoen wouldn't deny that expressions like (24) appear toexpress Strong Reciprocity.(24) Willow School's fifth-graders know each otherAn advocate of this position needs to explain why such examples appear to meansomething stronger than Weak Reciprocity. Doing so will presumably involve appeal topragmatic strengthening of the proposition that according to them is the sentence's literalmeaning. One way such strengthening might occur is through conversational implicature(Dalrymple et al. 1994: 76).As observed by Dalrymple et al, the added strength of (24) over WeakReciprocity does not seem to be cancellable, as evidenced by the infelicity ofthe following cancelling expression:Downloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014(25) *Willow School's fifth-graders know each other, but the oldest onedoesn't know the youngest.Hirschberg (1991), Dalrymple et al. (1994), and others insist on takingcancellability as a hallmark of conversational implicature. They wouldreject an analysis of these examples in terms of conversational implicature.It seems, however, that this claim is not so much based on theirtreatment of conversational implicature but is rather a consequence of theold dictum semantics is strong and pragmatics is weak. The present account of

Reinhard Blutner 143conversational implicature suggests that the borderline between semanticsand pragmatics (conversational implicature) cannot be drawn by thecondition of cancellability. The discussion of blocking has shown thatpragmatic anomaly isn't necessarily connected with inconsistency. In asimilar vein, non-cancellable implicatures can arise in spite of theconsistency of the corresponding cancelling expressions. To take a typicalexample from the pragmatics of adjectives, (26a) would suggest (26b) as anconversational implicature.(26) (a) This apple is sweet(b) Its pulp is sweet(c) ?This apple is sweet, but its pulp is not (perhaps, its peel is)The corresponding cancelling expression (26c), however, can be shown to bepragmatically anomalous for very general classes of common grounds (inspite of its semantic consistency). The theoretic treatment of such examples(see section 4) strongly suggests that a proper treatment of conversationalimplicatures may explain when an implicature is cancellable and when notIn this sense, the dictum semantics is strong (leading to non-cancellableinferences in each case) and pragmatics is weak (justifying cancellableinferences only) must be abandoned. 16 Instead, it is an important taskfor Lexical Pragmatics to explain when conversational implicatures arecancellable and when not.4 UNDERSPECIFICATION AND ABDUCTIONIn the former section, we have introduced the general constraint C. defining the range of possible refinements of an underspecified semanticrepresentation. However, we have only considered a rather provisionalexplication of this constraint. In this section we consider a realization ofC that seems refined enough to analyse some specific phenomena ofLexical Pragmatics. The main idea is to consider C(a) as the set ofabductive variants that can be generated from sem(a) by means of aspecific common ground that includes crucial aspects of world anddiscourse knowledge. It is shown how the incorporation of the abductivecomponent in the general pragmatic framework solves some of theproblems in connection with the pragmatics of adjectives and thephenomenon of systematic polysemy.Downloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014

144 Lexical Pragmatics4.1 Cost-based abduction: an extension of the basic mechanismFor the sake of explicitness, let us consider sem(a) as a conjunction ofpositive literals and let us propose weighted abduction (Stickel 1989; Hobbset al. 1993) as a general method to specify sem(a) by exploiting Horn clauseknowledge bases. The use of weighted abduction allows us to pair theabduced variants m, with its proof costs. The earlier measure of the globalcosts c(a, m,) (formula (13)) should then be replaced by an explicit accountof those proof costs.For the present purpose, we adopt Stickel's (1989) PROLOG-likeinference system for generating abductive specifications and his mechanismfor computing proof costs in a slightly simplified way. It is taken for grantedthat every literal in the initial formula is annotated with (non-negative)assumption costs q: q^1,...q^n.The knowledge base is assumed to provideformulas of the form p^1,. . . , p^* 1 —> q, where the literals pj in theantecedent are annotated with weights uiyThere are four inference rules that constitute abductive proofs anddetermine the assignment of concrete proof costs (for details, see Stickel1989):Resolution with a fact: If a current goal clause contains a literal that isunifiable with a fact in the knowledge base, then this literal is marked asproved. (The retention of a proved literal allows its use in future factoring).Resolution with a rule: Let the current goal clause be ... q' c ... and letthere be an axiom p^1,. . . p^11 —> q in the knowledge base. If q' and q areunifiable with most general unifier a, then the goal clause ...p c l' wi a,...,p c n um a,q'

Reinhard Blutner 145the probabilistic interpretation of costs, but we will not refrain from usingfactoring. Factoring some literals obtained by backward chaining can beproved to be a very useful operation in natural language interpretation (cf.Stickel 1989).It is now possible to incorporate the abductive component in the generalpragmatic framework viewing natural language interpretation as inferencesto pragmatically licensed updates. For simplicity's sake let me illustrate theincorporation of abduction by way of an elementary example. This gives methe opportunity to discuss some crucial differences between the presentapproach and the Hobbs-Stickel account where natural language interpretationis viewed as abductive inference to the best explanation. In order tosimplify matters, I will exclude effects of blocking via the Q-principle. Thatmeans, I will assume that there are no expression alternatives a' that mayblock any interpretation of a.Let us assume a knowledge base as presented in (27) and let us accept thatall axiom costs are zero.(*7)Thethe(28)eg: C 2-> sD 1D 1A A O S -> CA B 0 5 - > ADdiagram (28) shows the abductive inference graph when (27) is taken ascommon ground and sem(a) = S is taken as the starting clause.D 1TDAD 1TAB 0 - 5 ->A 051 1A->CC 2 -t 1g$IOT [Assuming C:2 • $10 = $20][Assuming A:[Assuming B:0.5 • $10 = $5]Downloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014The resulting set of abductive variants is presented in (29a) and the costsassociated with these variants are given in (29b).(29) (a) C(S) = {A,B,C}(b) c(S,A) = $io,c(S,B) = $5,c(S,C) = $20(c) p(S) = {B}(d) cg[S] = eg U {B}

146 Lexical PragmaticsSince we have assumed that iliere are no blocking alternatives, thecondition (22a) becomes vacuous and the set p(S) is the set of costminimalvariants, given in (29c). Since the expression B is consistent with eg,a pragmatically licensed update exists (satisfying the Quality conditions(15 a, b)). It is given in (2ad).The Hobbs-Stickel account is looking for minimal explanations, thatmeans it selects the cost-minimal variants from the set of the consistentabductive variants. This contrasts with the former view which first selectsthe cost-minimal variants from the set of all abductive variants and thenchecks them with regard to consistency. However, in the present case thismakes no difference, since the minimal variant B is consistent with eg, andconsequently it is at the same time the minimal explanation of S. Theupdating of eg by the minimal explanation gives the same result as alreadypresented in (2od).Now consider the common ground eg' given in (30), which is egaugmented by the clause B —» (i.e. by ->B).(30) eg': C 2 -* s 'D 1 A A 05 -> CD 1 A B os -+ AB ->In this case we have the same abductive inference graph as shown before in(28), and we get the same abductive variants and the same costs associated withthem. But now the cost-minimal variant B is inconsistent with eg'. From thisfact it follows that there is no pragmatically licensed update for S with regardto eg'. In other words, S becomes pragmatically anomalous with regard to eg'.Downloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014Now look at the Hobbs-Stickel account. It gives A as the minimalexplanation (cf. the diagram (31)). This leads to the postulation of eg' U {A}as update. Consequently, there is an important difference between theHobbs-Stickel account and the present one. On the Hobbs-Stickel view

Reinhard Blutner 147there is an update in each case when the starting clause sem(a) is consistentwith eg. The present account, on the other hand, yields a much morerestricted notion of update. There is a pragmatically licensed update onlywhen one of the cost-minimal abduced variants is consistent with eg. If allcost-minimal variants are inconsistent with eg they can be seen as 'blocking'any interpretation of the starting clause. As shown in the next section, thisdevice is appropriate to capture cases of pragmatic anomalies in naturallanguage interpretation.From a computational point of view, the present approach looks well if itis assumed that the abductive machine generates the abductive variants inthe order of its (estimated) costs. In this case, we have to assume simply thatthe abductive system stops if it has completed its first abductive proof. Theresult is then given to the consistency checker. If the result is consistent, thesystem has found an interpretation. If not, the system may indicate that itdoesn't understand—the only interpretation it can find is a faulty one.Perhaps, there is a mechanism of accommodating the knowledge base thatrestores interpretability after all, but even then there is no possibility toaccess other variants than the cost-minimal ones.The overall architecture of the Hobbs-Stickel account sets out to accessnon-minimal variants when the minimal ones do not provide explanations.This feature makes processing less efficient, and it makes it difficult todiscriminate between 'good' and 'bad' interpretations. In contrast, thepresent view of interpretation connects an efficient processing architecturewith the possibility of providing an explanation of pragmatic anomalies.Straightforwardly, this way of realizing efficient processing conforms withthe realization of the monotonicity property of language processing (e.g.Alshawi & Crouch 1992; cf. section 2.2), and it is this idea that helps us toexplain pragmatic anomaly.There is yet another important feature distinguishing the presentaccount from the Hobbs-Stickel approach, the possibility of having noncancellableimplicatures. Let us call a conversational implicature of anutterance a in eg contextually cancellable iff there is a strengthening eg' of egsuch that a is interpretable in eg' but is not longer a conversationalimplicature of a in eg'.Obviously, the entailment eg U {sem(a)} f= excludes the contextualcancellability of

148 Lexical Pragmaticsproposition B. For example, if we strengthen eg by adding ->B (as in (30), Swill be pragmatically anomalous in the new context. This shows that on thecurrent account cancellability is not a necessary feature of conversationalimplicature; some implicatures may be non-cancellable. The Hobbs-Stickelapproach, on the other hand, is in agreement with the standard view(resting on the highly defeasible notion of minimal explanation).The usefulness of cancellation as a test for conversational implicature hasbeen challenged in connection with the thesis 4 of section 3.2. The presentaccount of conversational implicature suggests that a proper treatment ofconversational implicature may explain when an implicature is cancellableand when not. The next subsection provides an analysis of the pragmatics ofadjectives that gives further evidence for this view.4.2 Abduction and the pragmatics of adjectivesOne part of speech is especially suited to demonstrating the phenomenon ofsemantic underspecification: the adjective. In section 2.1 we considered thefree variable view as an especially promising account of treating the meaningof adjectives. It had been stressed that the specification of free variables isnecessary for a full interpretation of an utterance. I will now demonstratehow the current theory yields an appropriate mechanism of (contextual)specification by applying it to the kind of examples discussed by Quine(i960) and Lahav (1993) (cf. section 2.1).(32) (a) The apple is red(b) Its peel is red(c) Its pulp is red(d) APPLE(d) A PART(d,x) A COLOUR(x,u) A u = red(e) APPLE(d) A PART(d,x) A PEEL(x) A COLOUR(x,u) A u = redMy claim is that (32b) but not (32c) can be construed as a conversationalimplicature of (32a). Input of the analysis is the underspecified semanticrepresentation given in (32d). One of the abductive specifications of thissemantic input specifies x as the peel part of the apple (see (32c)). For thecalculation of the corresponding costs we start with assumption costs asgiven in the first line of (3 3). Note that we take the assumption cost forthe 'slots' PART(d.x) and COLOUR(x,u) as negligible with regard to thecosts of the more 'specific' elements of the representation. 17 This contrastswith corresponding stipulations by Hobbs et al. (1993) but it agrees withthe general picture that specificity is the primary determinator of theassumption costs. Furthermore, we refer to axioms of the formareq

Reinhard Blutner 149certain conditional probabilities: u>, oc prob(q|pi) (cf. Hobbs et al. 1993). Ifthe pi are necessary conditions for q, then we have u, + u> 2 = i, and theweights U{ can be interpreted to estimate the saliences of the featurecomplexes p; with regard to p.(33) APPLE(d) $1 APART(d,x) $0 A C0L0UR(x,u) S0 Au = red 11APPLE(d)

150 Lexical Pragmatics(3$) (a) The apple is sweet(b) Its pulp is sweet(c) Its peel is sweetIt should be added that the present account evaluates utterances as (36) aspragmatically anomalous (assuming the former axioms and weights)(36) ?This apple is red, but its peel is not (perhaps, its pulp is)This qualifies implicatures like (32b) and (35b) as non-cancellable (undernormal circumstances—neglecting the possibility of genetic engineering).Finally consider the contrast between (37a) and (37b):(37) ( a ) ?The tractor is pumped up(b) The tyres of the tractor are pumped up(c) ?The coachwork of the tractor is pumped up(d) TRACTOR(d) A PART(d, x) A PRESSURE(x, u) A u = pumped up(e) TRACTOR(d) A PART(d,x) A TYRES(x) A P-STATE(x,u) Au = pumped up(f) TRACTOR(d) A PART(d, x) A MOTOR(x) A P-STATE(x, u) Au = pumped upThe present account predicts (37a) as pragmatically anomalous. Thisprediction results from the fact that those parts of tractors that may bepumped on (the tyres) are only marginally diagnostic for identifyingtractors and therefore the corresponding interpretation (37b) can be blockedby specifications that refer to more salient parts, for example as shown in(37c). However, the latter specifications suffer from sort conflicts andtherefore violate the condition (15).To make the argument explicit, let us start with (37d) as underspecifiedrepresentation of (37a), and let us compare the cost for calculating the twoenrichments (37c) and (371") (related to (37b) and (37c)). The diagram (38)presents the corresponding abductive inference graph that is relevant forabducing the pumped up fyra-interpretation (37c).(38) TRACTOR(d) $1 APART(d,x) M A P-STATE(x,u) $0 A u = up $1Downloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014TRACTOR(d)

Reinhard Blutner 151Note that this graph has practically the same structure as that given in (3 3). Inthe present case the factoring operation unifies the part and pressure-stateslots arising from the predicate complex of the utterance (37a) with thosethat emerge while conceptually decomposing the subject term the tractor.Next let us consider an abductive inference that corresponds to anenrichment referring to more salient parts as the tyres, say the motor of thetractor as it is given in (39).( 39 ) TRACT0R(d) $1 A PART(d, x) $0 A P-STATE(x, u) $0 A u = up $1TRACTOR(d)

152 Lexical PragmaticsThis utterance does not have the highly marked status of (37a). The presentaccount explains this by making the plausible assumption that the tyres ofbicycles are one of the most salient parts of bicycles. Consequently, in thiscase the pumped up tyres- interpretation comes out as a cost minimal one, andit doesn't suffer from sort conflict.Needless to say, the present considerations regarding the amounts of theparameters have to be supported by careful empirical studies. However, as afirst step considerations of this kind may be valuable. They may demonstrateat least which kinds of influence are conceivable, and this again maybe tested empirically.4.3 Abduction and systematic polysemyIn this section I will demonstrate how the ideas put forward in section 4.1may provide a mechanism for generating the range of the conceptuallysalient senses of institute-type words—a mechanism that solves the restrictionproblem of polysemy. Adopting the radical underspecification view(section 2.2), I will show how the extended mechanism of conversationalimplicature is capable of giving a principled account.The general idea that leads us to underspecified representations in thecase of institute-type words is as follows. Suppose there are certain entitieswhich can be understood as conceptual frames or schemata and can beclassified according to the variety of institute-types (government, school,parliament, etc.). Suppose further that these entities can be consideredunder different perspectives. These perspectives are assumed to provide moreconcrete realizations of the rather abstract concept of a certain institute-typee, perhaps realized as building, process, or institution property. However,the particular perspective adopted and, consequently, the concreterealization of the intended institute-type remains semantically open.In a first approximation, the semantic representation of institute-typenominals may look like (41a, b).Downloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014(41) (a) Ax 3e[SCHOOL(e) A REALIZE(e, x)](b) Ax 3e[GOVERNMENT(e) A REALIZE(e, x)]Note that the specification of x as building, process, or institution properhas not been specified in the lexicon. That means that the variety ofdifferent interpretations has not been treated by stipulating semanticambiguities. Note furthermore that the different restrictions on interpretativevariants, for example for school and government, are no longer treatedsemantically. As a consequence, the restriction problem of polysemy has tobe analysed pragmatically.

Reinhard Blutner 15 3In the previous subsection we used an axiom of the formq •— p 1^1 A p^2to abduce, for instance, the existence of peel parts ofan assumed apple from the existence of the apple. In a similar vein, wenow use axioms of this form in order to abstract, for instance, theexistence of a building and/or an institution realization from theexistence of an entity of type school or government. Weighted abductionrules that provide the corresponding decompositions are presented in(42a, b) for the case of school:(42) (a) SCHOOL(e) «- REALIZE(e,x) a7 A BUILD(x) Q ( I - 7) A etc 1 ""(b) SCHOOL(e) «- REALIZE(e, x) a ' 7 A INSTIT(x) Q '('~ 7) A etc 1 ""'Analogously to the case discussed before, the parameters may beinterpreted as follows: a as the salience of the building realization of aschool, a' as the salience of the institution realization, and 7 as thesalience of the realization slot (7 < 1). It is important to note that theassumptions about the weights in (42) are correct only when thecondition on the left side of

154 Lexical Pragmatics(45) (a) 3e[SCHOOL(e) A REALIZE(e,x) A BUHDING(x) AHAS_A_FLAT_ROOF(x)j(b) 3e[GOV(e) A REALIZE(e,x) A BUHDING(x) AHAS_A_FLAT_ROOF(x)]In both cases, the first two conjuncts result from the lexical inputs of theinstitute-type nominals, and the remaining ones correspondingly to thepredicate complex. The expression BUILDING(x) is due to the assumed sortrestriction provided by the predicate complex, and it is singled out as animportant representational element in the present analysis.The diagram (46) shows the part of the abductive inference graph that isrelevant for abducing the building-interpretation (43b) starting with (43a)(in its pre-analysed form (45a)). Note that there is no real abduction in thisvery crude and simplifying analysis. The graph shows a 'conceptualdecomposition' of the subject term and a factoring operation that unifiesthe occurrence of BUILD(x) resulting from this decomposition with itsoccurrence resulting from the predicate complex. This effects a saving inassumption costs (by an amount of a).(46) SCH00L(e) $1 AREALIZE(e,x)$O A BUILD(x) $0 A . . . s "SCHOOL(e)

Reinhard Blutner 155of comparable amount, since the building and institution reading of schoolcan be seen as realizing concepts of both the basic level of buildings andthat of institutions. Consequently, the condition a/a' > 7 (with 7

156 Lexical Pragmaticson the way how the restriction problem of polysemy may be solved byconsidering the probabilistic nature of conceptual knowledge.5 CONCLUSIONOne aim of this paper was to collect some general problems that have aprima facie claim on the attention of linguists interested in LexicalSemantics. These problems had to do with the utterance of words withinconcrete conceptual and contextual settings and went beyond the aspects ofmeaning typically investigated by a contrastive analysis of lexemes withinthe Katz-Fodor tradition. Three groups of problems were considered: (i)pragmatic compositionality, (ii) blocking, and (iii) pragmatic anomaly. Theproblems came to the fore in connection with the pragmatics of adjectivesand the phenomenon of systematic polysemy. The same points can be madewith regard to word formation in general (e.g. Aronoff 1976; Bauer 1983)and the interpretation of compounds in particular (e.g. Meyer 1993; Wu1990). Moreover, the investigation of kinds of polysemy other than thosefound with institute-tyipe. words may be helpful in order to see the uniquityof these problems (cf., for instance, Lakoff's (1987) study on Englishprepositions and Sweetser's (1990) investigation of English perceptionverbs). Furthermore, Fabricius-Hansen's (1993) research on how theinterpretation of noun-noun compounds is affected by a genitive attributemay raise the same problems in a more complex area.The second aim of this paper was to sketch a new approach called LexicalPragmatics that deals with these problems in an explanatory way and tries togive a systematic account of the phenomena under discussion. Theparadigm is based on two simple principles: (i) an adequate representationof lexical items has to be given in a semantically underspecified format, and(ii) their actual interpretation has to be derived by a pragmatic strengtheningmechanism. The basic pragmatic mechanism rests on conditions ofupdating the common ground and allows to give a precise explication ofnotions such as generalized conversational implicature and pragmaticanomaly. The fruitfulness of the basic account was initially established byits application to a variety of recalcitrant phenomena, among which itsprecise treatment of Atlas & Levinson's Q- and I-principles and theformalization of the balance between informativeness and efficiency innatural language processing (Horn's division of pragmatic labour) deserveparticular mention. The basic mechanism was subsequently extended by anabductive reasoning system that is guided by subjective probability. Theextended mechanism turned out to be capable of giving a principledaccount of lexical blocking, the pragmatics of adjectives, and certain typesof systematic polysemy.Downloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014

Reinhard Blutner i$7I find it important to apply the ideas to other possibly more complex andmore realistic examples than those considered here. Moreover, methods areneeded that allow one to measure the values of the probabilistic parametersthat control and organize conceptual knowledge.Seen from a moderately distant viewpoint, the standard accounts ofLexical Semantics may appear as an incoherent research field which is atodds with itself. As an endeavour that has to access Grammar, Semantics,and aspects of utterance interpretation at the same time it multiplies thediversion of. these disciplines. Overstretched by the task of theoryformation, it either combines theoretical rigour with descriptive poverty,or, more predominantly, it leads to linguistic anecdotalism, collecting prettyand curious observations without theoretic control. I sense that it is theinadequacy or the lack of a genuine pragmatic component that has led tothis situation. In so far as Lexical Pragmatics tries to take pragmaticsseriously—especially the conception of conversational implicature—and inso far as it is explicit about this component, it may substantiate a division oflabour between grammatical and pragmatic aspects of the lexicon. This maybroaden the way for overcoming the unfortunate situation just mentioned.Perhaps most details concerning the main ideas of the present account inconcrete terms may prove false in the future. This may concern, first at all,the Economy principles and their interaction. In order really to justify thedetails of these principles we need more empirical evidence and studies. Butit is also crucial to discover the reasons that explain why the principles arejust as they are. This brings us to a reductionist programme as is currentlypursued in the domain of Integrative Connectionism (e.g. Smolensky 1995). Afirst attempt at achieving a full reduction of Speaker's economy (I-principle)and Hearer's economy (Q-principle) to connectionist principles is currentlyunder way.AcknowledgementsI am grateful to Manfred Bierwisch, Paul David Doherty, Bart Geurts, Gerhard Jager,Thomas Jiingling, Annette Lefimdllmann, Chris Pirion, and Rob van der Sandt for usefulcomments on earlier versions of this paper. Special thanks go to an anonymous referee ofthe JS. I do not, however, intend to imply by this that they endorse my approach. Inparticular, Thomas and Manfred don't believe a word of it.Downloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014REINHARD BLUTNER Received: 18.10.97Humboldt University, Berlin Final version received: 05.04.98Jagerstrasse 10-1110117 BerlinGermanye-mail: blutner@german.hu-berHn.de

158 Lexical PragmaticsNOTES1 I have to thank an anonymous referee ofthe JS for this example.2 An anonymous referee notes that forhim beef sounds equally good as cow in(7). Consequently, the relevant differencein acceptability between (6b)and (7) is in cow (which sounds fine)more than in beef. This fact is sufficientto illustrate the phenomenon of deblocking,which is the relevant one in thepresent context3 These problems concern (i) the restrictivenessof the coercion mechanism, (ii)the apparent inflation of shiftingoperations, (iii) the stipulation of anadditional checking mechanism thatdiminishes the use of monotonicprocessing, and (iv) problems with theanalysis of co-predication in case oflogical polysemy. For details and furthercriticism, see Copestake & Briscoe('99S)> Fodor & Lapore (to appear),and Blutner (to appear).4 In general, cost factors relate to the(estimated) cost of accessing the differentinterpretations. In section 4.1 anexplicit account is provided whichbrings our system as close as possibleto a Bayesian network and takes costsas negative logarithms of certain conditionalprobabilities.5 Sperber & Wilson's (1986) extremeposition of reducing the maxims tojust one—the maxim of relevant—isn'trelevant in the present context. Asargued by Levinson (1989), Sperber &Wilson try the 'impossibility of reducingcountervailing principles to onemega-principle'. They concentrate onthe phenomena of classic particularizedRelevance implicatures illustrated byGrice, and they fail to account for thewhole range of generalized conversationalimplicatures—the implicaturesthat are most important for lexicalpragmatics.6 See, for example, Bierwisch (1983) andJackendofT (1983) for similar distinctions.The important point of this distinctioncorrelates with Grice's proposalin his William James Lectures to make adistinction within the 'total signification'of a linguistic utterance betweenwhat a communicator has said (in acertain favoured, and maybe to somedegree artificial, sense of said), and whata communicator has meant beyond it(what she has implicated, indicated,suggested).7 Mental model and conceptual representationare more psychologically colouredterms; information state is the favouredterm used in formal semantics.8 See, for example, McEliece (1977).9 In section 4 a more refined cost-functionis developed which sometimes allows theselection of state descriptions that are notminimal surprising. However, these statedescriptions can be characterized as the'better interpretations' because they aremore unifying and, in some sense, morerelevant than less surprising ones. Thisformulation brings us closer to the ideaof Atlas & Levinson (1981) where theI-principle is intended as inference to'the best interpretation' (with 'bestinterpretation' informally understoodas interpretation which prefers coreferentialreadings of entities, making useof stereotypical relations betweenreferents or events. However, it shouldbe added that the way in which Atlas &Levinson (1981) try to formalize theirPrinciple of Informativeness seemsrather misleading.10 Here K is the epistemic operator indexedto H and S, respectively. The epistemiclogic I assume is Hintikka's (1962). Asdiscussed by Zeevat (1997), this conditionon common grounds is only a necessaryone. Developing a more refined definitionof common ground, Zeevat formulatesalso an update operation for commongrounds. His conception, however,Downloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014

Reinhard Blutner 159ignores the effects of conversationalimplicatures which also influence thecommon ground. The present accountseeks to grasp these effects in a first andrather sketchy manner. Furthermore, itshould be noticed that I use the notionof information state sometimes as referringto a set of possible worlds andsometimes as a representational structure(state description or disjunction ofstate descriptions).It should be added that in the presentcontext we read K as know for sure (thisdoesn't presuppose the complementproposition). However, there is a problemwith this formulation that at leastshould be mentioned. In the presentformulation, the common groundcannot include propositions that someor all of the participants know to befalse. However, there are kinds ofconversations where this formulation isunsatisfying. For instance, consider aChristmas-time conversation wherethe proposition that there is a SantaClaus may be common ground, even ifsome or all of the participants knowthat there isn't any Santa Claus.11 Identifying state descriptions with setsof possible worlds and p(a) with afamily of sets of possible worlds, wecan write this condition in the followingway: Up(a) D cg[a].12 Beside the epistemic operator K weneed its dual P =j e f -1K-1. Hintikkareads K a 0 as a knows that (j>. It isimportant to note that Hintikka isusing the verb know in a technicalsense without the usual factive presupposition.In this vein, -iK,^ can be readas what a knows is not that p, andP a 0(=d e f -iK-2~i) can be read as forall a knows, it is possible that p.13 The presented arguments showing theadvantages of the present approachover the traditional approach based onHorn-scales are due to Rob van derSandt I thank him for allowing me toinclude his considerations in this article.14 An anonymous referee notes that atleast for the pair persuade to not-dissuadefrom the equivalence is spurious, sincedissuade presupposes that the personpreviously intended the complementThe referee's example: If John wasundecided about whether to vote forClinton, Mary could persuade John notto vote for Clinton, but she couldn'tdissuade him from voting for him.15 Only under very special conditions is itpossible to construct representationalpendants to the non-representationalelements, for example when we developintuitions about our cognitive system orabout our mental activity. In this sense,the representation of salience, relevance,and so on is possible. Usually, theserepresentations are comparative in characterand are not quantitatively scaled.16-With regard to the first part of thedictum, seeing non-cancellability as anecessary condition of entailment (orseeing cancellability as a sufficient conditionof conversational implicature), Iagree with Hirschberg (1991) andassume that it is right, at least if it ispossible to discriminate cancellationfrom suspension (the calling into questionof an asserted proposition), from contextualdisambiguation, and from certainforms of speaking loosely (for carefuldiscussion, cf. Hirschberg 1991: 28 ff).17 As assumption cost of the latter units Ihave stipulated $1. You may see thisstipulation as fixing the $-unit.18 An anonymous referee has suggestedmore minimal contrasts in order tobring out the relevant difference:?The government has a flat roof vs. TheMinistry of Justice has aflat roof.?The university has a flat roof vs. Thecollege of engineering has aflat roof.Downloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014

i6o Lexical PragmaticsREFERENCESAlshawi, H. & Crouch, R. (1992),'Monotonic semantic interpretation', inProceedings of ACL, Delaware, 32-9.Aronoff,M(i976), WordFormation in GenerativeGrammar, MIT Press, Cambridge, MA.Atlas, J. & Levinson, S. (1981), 7f-clefts,informativeness and logical form', in P.Cole (ed.), Radical Pragmatics, AcademicPress, New York, 1-61.Bauer, L. (1983), English Word-Formation,Cambridge University Press, Cambridge.Bierwisch, M (1983), 'Semantische undkonzeptuelle Representation lexikalischerEinheiten', in W. Motsch &R Ruzicka (eds), Untersuchungen zurSemantik, Akademie Verlag, Berlin,61-99.Bierwisch, M. (1989), The semantics ofgradation', in M. Bierwisch & E. Lang(eds), Dimensional Adjectives, Springer-Verlag, Berlin etc., 71-261.Blutner, R, (to appear), 'Lexical semanticsand pragmatics', in Linguistische Berichte.Caramazza, A. & Grober, E. (1977), 'Polysemyand the structure of the subjectivelexicon', in C. Rameh (ed.), GeorgetownUniversity Roundtable on Language andLinguistics, Georgetown UniversityPress, Washington, DC.Carnap, R. (1947), Meaning and Necessity,University of Chicago Press, Chicago &London.Charniak, E. & Shimony, E. S. (1990),'Probabilistic semantics for cost basedabduction', Technical Report CS-90-02,Computer Science Department, BrownUniversity.Clark, E. V. (1990), 'On the pragmatics ofcontrast', Journal of Child Language, 17,417-31.Copestake, A & Briscoe, T. (1995),'Semi-productive polysemy and senseextension', Journal of Semantics, 12,1S-67.Cruse, D. A. (1986), Lexical Semantics,Cambridge University Press, Cambridge.Deane, P. D. (1988), 'Polysemy andcognition', Lingua, 75, 325-361.Dalrymple, M., Kanazawa, M., Mchombo,& Peters, S. (1994), "What do reciprocalsmean?', in M. Harvey & L. Santelmann(eds), Proceedings of the Fourth Semanticsand Linguistic Theory Conference: SALTIV, Cornell University, Ithaca.van Deemter, K. & Peters, S. (1996), SemanticAmbiguity and Underspecification, CSLIPublications, Stanford, CAFabricius-Hansen, C. (1993), 'Nominalphrasenmit Kompositum als Kern',Beitrdge zur Geschichte der deutschenSprache und Literatur, 115, 193-243.Fodor, J. A & Pylyshyn, Z. W. (1988),'Connectionism and cognitive architecture:a critical analysis', Cognition,28, 3-71.Fodor, J. A & Lapore, E. (to appear), Theemptiness of the lexicon: critical reflectionson J. Pustejovsky's The GenerativeLexicon'.Gazdar, G. (1979), Pragmatics, AcademicPress, New York.Grice, P. (1968), Logic and Conversation, textof Grice's William James lectures atHarvard University, published in Grice(1989).Grice, P. (1989), Studies in the Way of Words,Harvard University Press, Cambridge,MAHintikka, J. (1962), Knowledge and Belief: AnIntroduction to the Logic of the Two Notions,Cornell University Press, Ithaca & NewYork.Hirschberg, J. (1991), A Theory of ScalarImplicature, Garland Publishing, Inc.,New York & London.Hobbs, J. R., Stickel, M E., Appelt, D. E.,& Martin, P. (1993), 'Interpretation asabduction', Artificial Intelligence, 63,69-142.Horn, L. R. (1984), Toward a new taxonomyfor pragmatic inference: Q-based and R-based implicatures', in D.Schifrrin (ed.), Meaning, Form, and Use inContext, Georgetown University Press,Washington, DC, 11-42.Downloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014

Reinhard Blutner 161Householder, F. W. (1971)- Linguistic Speculations,Cambridge University Press,London & New York.Jackendoff, R (1983), Semantics andCognition, MIT Press, Cambridge, MA.Kamp, H. (1975), Two theories about adjectives',in E. L. Keenan (ed.), FormalSemantics/or Natural Language, CambridgeUniversity Press, Cambridge, 123-55-Karttunen, L. (1972), 'Possibly and Must', inJ. P. Kimball (ed), Syntax and Semantics 1,Seminar Press, New York, 1-20.Karttunen, L. & Peters, S. (1979), 'Conventionalimplicature', in C.-K. Oh &D. A. Dinneen (eds), Syntax and Semantics11: Presupposition, Academic Press, NewYork, 1-56.Keenan, E. L. (1974), The functionalprinciple: generalizing the notion ofSubject of, in Papers from the TenthRegional Meeting of the Chicago LinguisticSociety, Chicago, IL, 298-310.Keil, F. C. (1979), Semantics and ConceptualDevelopment, Harvard University Press,Cambridge, MA.Kiparsky, P. (1982), 'Word-formation andthe lexicon', in F. Ingeman (ed.), Proceedingsof the 1982 Mid-America LinguisticConference.Lahav, R. (1989), 'Against compositionality:the case of adjectives', PhilosophicalStudies, 55, 111-29.Lahav, R (1993), The combinatorialconnectionistdebate and the pragmaticsof adjectives', Pragmatics and Cognition, I,71-88.Lakoff, G. (1987), Women, Fire, andDangerous Things: What Categories RevealAbout the Mind, University of ChicagoPress, Chicago.Lang, E. (1989), The semantics of dimensionaldesignation of spatial objects', inM Bierwisch & E. Lang (eds), DimensionalAdjectives, Springer-Verlag, Berlin,71-261.Langendoen, D. T. (1978), The logic ofreciprocity', Linguistic Inquiry, 9, 177-97.Lehrer, A. (1968), 'Semantic cuisine', fournalof Linguistics, 5, 39-56.Lehrer, A. (1970), 'Static and dynamic elementsin semantics: hot, warm, cool, cold',Papers in Linguistics, 3, 49-74.Lehrer, A. (1978), Structures of the lexiconand transfer of meaning, Lingua, 45,95-123.Levinson, S. (1983), Pragmatics, CambridgeUniversity Press, Cambridge.Levinson, S. (1987), 'Pragmatics and thegrammar of anaphora', fournal ofLinguistics, 23, 379-434-Levinson, S. (1989), 'Relevance', fournal ofLinguistics, 25, 455-72.McCawley, J. D. (1978), 'Conversationalimplicature and the lexicon', in P. Cole(ed.), Syntax and Semantics 9: Pragmatics,Academic Press, New York, 245-59.McCawley, J. D. (1993), Everything thatLinguists have Always Wanted to Knowabout Logic but were Ashamed to Ask, 2ndedn, University of Chicago Press, Chicago.McEliece, R. (1977), Theory of Informationand Coding, Addison-Wesley, Reading,MA.Matsumoto, Y. (1995), The conversationalcondition on Horn scales', Linguistics andPhilosophy, 18, 21-60.Meyer, R. (1993), Compound Comprehensionin Isolation and in Context, Max NiemeyerVerlag, TiibingeaMontague, R (1970), 'Universal Grammar',Theoria, 36, 373-98.Nunberg, G. (1979), The non-uniquenessof semantic solutions: Polysemy',Linguistics and Philosophy, 3, 143-84.Nunberg, G. (1995). Transfers of meaning',fournal of Semantics, 12, 109-132.Nunberg, G. & Zaenen, A. (1992), 'Systematicpolysemy in lexicology andlexicography', in K. Varantola, H.Tommola, T. Salmi-Tolonen, J. Schopp(eds), Euralex II, Tampere, Finland.Partee, B. (1984), 'Compositionality', in F.Landman & F. Veltman (eds), Varietiesof Formal Semantics, Foris, Dordrecht,281-311.Pustejovsky, J. (1989), Type coercion andselection', paper presented at WCCFLVIII, April 1989, Vancouver, BC.Downloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014

162 Lexical PragmaticsPustejovsky, J. (1991), The generative lexicon',Computational Linguistics, 17, 4,409-41.Pustejovsky, J. (1993), T)T e coercion andlexical selection', in J. Pustejovsky (ed.),Semantics' and the Lexicon, Kluwer,Dordrecht, 73-96.Pustejovsky, J. (1995), The GenerativeLexicon, MIT Press, Cambridge, MA.Pustejovsky, J. & Boguraev, B. (1993):'Lexical knowledge representation andnatural language processing', ArtificialIntelligence, 63, 193-223.Quine, W. V. O. (i960), Word and Object,MIT Press, Cambridge, MA.Rumelhart, D. E., McClelland, J. L. and thePDP Research Group (1986), ParallelDistributed Processing: Explorations in theMicrostructure of Cognition, Vol. 1: Foundations,MIT Press/Bradford Books,Cambridge, MA.Sadock, J. M. (1978), 'On testing forconversational implicature', in P. Cole(ed.), Syntax and Semantics, Vol. 9:Pragmatics, Academic Press, New York,281-97.Smolensky, P. (1995), 'Constituent structureand explanation in an integratedconnectionist/symbolic cognitive architecture',in C. Macdonald & G.Macdonald (eds), Connectionisnv Debateson Psychological Explanation, Vol. 2, BasilBlackwell, Oxford, 221-90.Sommers, F. (1959), The ordinary languagetree', Mind, 68, 160-85.Sommers, F. (1963), Types and ontology',Philosophical Review, 72, 327-63.Sperber, D. & Wilson, D. (1986), Relevance,Blackwell, Oxford.Stickel, M. E. (1989), 'Rational and methodsfor abductive reasoning in naturallanguage interpretation', in R. Studer(ed.), Natural Language and Logic,Springer-Verlag, Berlin, 233-52.Sweetser, E. E. (1990), From Etymology toPragmatics, Cambridge University Press,Cambridge.Thomason, R H. (1990), 'Accommodation,meaning, and implicature: interdisciplinaryfoundations for pragmatics', in P.R Cohen, J. Morgan, & M. E. Pollack(eds), Intentions in Communication, MITPress, Cambridge, MA.Wu, D. (1990), 'Probabilistic unificationbasedintegration of syntactic andsemantic preferences for nominal compounds',Proceedings of the 13th InternationalConference on ComputationalLinguistics (COLING go), Helsinki,413-18.Zeevat, H. (1997), The common groundas a dialogue parameter', in A. Benz &G. Jager (eds), MunDial'97: Proceedingsof the Munich Workshop on FormalSemantics and Pragmatics ofMunich.Dialogue,Downloaded from http://jos.oxfordjournals.org/ by guest on September 12, 2014