<?xml version="1.0" standalone="yes"?>
<Paper uid="C92-1050">
  <Title>Incrementing Discourse with the Negation Relation</Title>
  <Section position="2" start_page="0" end_page="0" type="metho">
    <SectionTitle>
2 Semantic analysis
</SectionTitle>
    <Paragraph position="0"> Besides negation and affirmation, focus particles such as only and even express relations between an associate and frame. Here I summarize the unified semantic analysis of these particles given by Moser (1992b). To begin with, consider only and its interaction with stressfocus. A variation in stress-focus in an only sentence can lead to different truth conditions. Note the different truth conditions for (5) and (6). Since there is no discernible difference between a stress-focus in an only sentence and one in another context, this systematic effect of stress-focus on truth-conditions supports the inclusion of stress-focus as part of the formal structure of any sentence, as argued for by Klein (1991) and as will be assumed here. Stress-focus is effectively a structural feature along with the position features, the more traditional structural features relating to placement in a syntax tree. There is a formal difference between two sentences which have exactly the same sequence of words based on how they are stressed, a difference which is realized by a particular prosodic feature, that of stress. Thus, (5) and (6) are not the same sentence with different intonation, rather, they are formally different sentences.</Paragraph>
    <Paragraph position="1"> In addition to a stress-focus associate, only may occur without a stress-focus as in (7). In these eases, its associate is the adjacent constituent. The relation analysis applies independently of how oaly's associate is determined, through position or stress. The same observations hold true for the associate of negation, but without the truth conditional consequences noted for  (5) and (6).</Paragraph>
    <Paragraph position="2"> (5) Ann only gave Trim INK to Bea \[1~ unique-El \[A ink\] \[F Ax(Ann gave z to Bea)\] (6) Ann only gave the ink to BEA \[a unique-El \[a Bea\] \[e Ax(inn gave ink to x)\] (7) Ann gave only the ink to BEA  Acq'ns DE COLING-92, NANTES, 23-28 AOt3T 1992 3 1 8 PROC. OF COLING-92. NANTES, AUG. 23-28, 1992 A relation and its associate, regardless of whether determined by position or stress, may be viewed as structuring the logical proposition. That is, the relation-associate-frame structure is a formal object, an instaBce of the structured meanings discussed by Cresswell (1985) and yon Stechow (1988). Tile interpretation of these relation-associate-frame structures is a three step process. First, a set of alternatives to the associate is determined. These alternatives are things which are similar to the associate in contextually relevant ways. To determine tile alternatives, the relevant syntactic-semantic type is inferred from the ~sociate.</Paragraph>
    <Paragraph position="3"> Then, starting from the instances of this type, there may be further contextual restrictions on the members of the alternatives. For (5), the syntactic-semantic type might be art supplies. Then, tile context could be restrictive so that only the ink and paint are in tile set of alternatives. The second step in the interpretation is that tile frame is used to select a subset of these alternatives, call it the frame set. In this case, the frame set will be the subset of tile alternative supplies which Ann gave Bea. Third, the only relation nmst hold between the associate and the frame set in order for the sentence to be true. For (5) to be true, the frame set, those art supplies which Ann gave to Bea, will have the ink as their uniqBe element.</Paragraph>
    <Paragraph position="4"> The two-place relations expressed by focus particles do not have their arguments rigidly typed. Rather, the associate may be of any semantic type and the frame will be a set of elements of that type. Semautically, these particles deuote element-to-set relations. Defining them this way allows for all associate that is either a stress-focus or an adjacent constituent. The general definition of focus-particles as element-to-set relations raises the question of how the associate, the element argument to the relation, is determined. The answer l suggest is that the position of the focus particle determines an environment for the associate. Then, the associate will be the closest stress-focus in that environment or it will be the entire environment. Thus, the syntactic relationslfip between particle and associate, stress and locality all play a role in determining tile associate of a focus particle.</Paragraph>
  </Section>
  <Section position="3" start_page="0" end_page="0" type="metho">
    <SectionTitle>
3 Information structure
</SectionTitle>
    <Paragraph position="0"> In this section, I note the similarity between the relation-associate-frame structure just discussed and information structure. Conceptually, information structure is the differentiation of the content of a sentence according to tile speaker's beliefs about tile heater's knowledge or current awareness. Chafe (1976) describes this in terms of the metaphor of packaging, noting that the same logical content can be packaged iu various ways depending on what it is being used for. Information structure is a kind of non-truth-conditional meaning which is inherently pragmatic because it is defined with reference to speaker, hearer and context.</Paragraph>
    <Paragraph position="1"> However, while the meaning conveyed involves contextual factors, it does not follow that tile means of conveying it does. That is, while information structure expresses how a sentence fits in tile context, it is possible that context is not a factor in determining the information structure itself. Recent work by Vallduvl (1990) suggests that information structure lie viewed as one aspect of the fornlal structure of a sentence, on a par with predicate-argument structure and logical form.</Paragraph>
    <Paragraph position="2"> '/'he major constituents of information structure are the information-focus, the portion of content which the speaker deems will be informative or will make the point of the sentence, and the ground, the remainder of the content wltich is assumed to be uncontroversial and serves ms a vehicle for the information-focus. As a subpart, the ground contains the link, tile particular entity which the utterance is about. The features which mark the information-focus are the same position and stress features described for determining aLL associate. In fact, in Moser (1992a), I propose the major constituents of information structure be extended to include a relation between information-focus and ground, making it a tripartite structure. As evidence for this extension, I note the tendency of negation and only to accompany the information-focus in constructions which mark information structure by placing tile information-focus in an isolated position. Thus, information structure is an instance of a relation-associate-frame structure.</Paragraph>
    <Paragraph position="3"> While the structure and formal marking of information structure are identical to the relation structures already discussed, the interpretation is not. The semantic interpretation by generating alternatives, selecting a frame set and checking the relation between associate and frame set expresses nothing about the hearer's awareness. Instead, information strncture is all additional interpretation of a relation structure, made simtdtaneously with the semantic one. Vallduvi proposes that the pragmatic interpretation of information structure is effectively instructions to the hearer for updating mental lilt cards for the entities in the discourse. Here I take a slightly ditferent perspective, i.e. that tile instructions are concerned with updating a file card representation which is external to the discour~ participants.</Paragraph>
    <Paragraph position="4"> The link specifies the address at which the information of the utterance is to be stored, so its pragmatic interpretation is a C/~O-TO instruction. The ground specifies content which is currently known or under discussion, a command to RETRIEVE all open proposition front a file card. Finally, I modify Vallduvi's instructions to allow for a relation constituent as part of information structure. Rather than SUBSTITUTE tile information-focus into the ground, the more general Ac'Yzs nE COLING-92, NANTES, 23-28 Aofrr 1992 3 1 9 PROC. OF COLING-92, NAIwrEs, AUG. 23-28, 1992 RELATE information-focus to ground using a specific relation is used. Consider the example in (8). Each constituent of information structure is interpreted by an instruction; the link is used to locate a file card, the ground is used find an open proposition which is expected to be there, and the information-focus is used to fill in the open proposition depending on what relation is specified. The result is a single, unstructured proposition on the appropriate file card. If the information structure used to convey this proposition were different, the file card update would be done differently, although the same proposition would result.</Paragraph>
    <Paragraph position="5">  (8) The girls published THE BOOK \[R E\]\[a the book\] \[v \[L the girls\] published x\] (i) GO-To(the girls) (ii) RETRIEVE(_ published x) (iii) RELATE(E, the book, x)  There are two ways in which these instructions are inadequate. First, the ground may require distinct interpretations depending on the syntactic construction used. Prince (1986) discusses a variety of constructions which mark an open proposition as presupposed. The different constructions vary in the way their open proposition is presupposed, however. Second, this is a literal interpretation of information structure which does not address the more complex possibilities for its interaction with the context.</Paragraph>
  </Section>
  <Section position="4" start_page="0" end_page="0" type="metho">
    <SectionTitle>
4 Discourse incrementation
</SectionTitle>
    <Paragraph position="0"> The representation of the ongoing discourse has the structure of a set of file cards. One file card is maintained for each entity under discussion, a catalogue of the predications made about that entity. The discourse content which is being negotiated is the sum of information on all the file cards. File cards are a metaphor for a particular organization of information, one which has been independently suggested by authors concerned with several areas of inquiry in cognitive science. Each file card is arranged with two parts: (i) a marker or index, an element of pure reference distinct from any particular predication about the entity, and (ii) a list of unstructured propositions, either affirmative or negative.</Paragraph>
    <Paragraph position="1"> Ina file card representation of discourse, there is a natural distinction between the function of NPs, which pick out file cards, and the function of the sentence which NPs are embedded in, which specify predications to go on the file cards. For each NP in an utterance, a file card is activated, i.e. associated with a discourse marker with the predicates contained in the NP. The rules for this activation make use of the NP's form as pronoun, definite or indefinite, and will not be discussed here. The distinction between the NP function as file card activators and the rest of the sentence as adding predications to the file card is a kind of processing distinction. No difference in truth conditions arise from the autonomy of file card activation. However, Prince (1990) suggests some linguistic evidence for this processing distinction. In Moser (1992a), I suggest that this distinction in negative sentences may help account for the tendency of referring expressions to be understood as outside the scope of negation. Negative sentences followed by a denial of existence are argued to be cases of constructing file cards for entities which do not exist.</Paragraph>
    <Paragraph position="2"> Once the file cards are activated, the process of their updating takes the structured proposition and enters the appropriate fact(s) (=unstructured proposition) on the activated file cards. In the process of this updating, the pragmatic interpretation of tile relation structure as information structure arises and may produce inferences which connect to the preceding discourse. The result of this updating must be at least the appropriate entries on the file cards to reflect the semantics of the sentence. A model theoretic semantics for a file card representation of discourse is defined by Helm (1982) in terms of a function for embedding the set of file cards in a model. Assuming this embedding function, tile truth conditions of relation structures are sufficiently specified by entries on the file cards.</Paragraph>
    <Paragraph position="3"> Example (9) will be used to illustrate discourse incrementation. Since the activation of file cards takes place independently, the structured proposition fur the increment is (10). Assume that the file cards for the context are as shown in Figure 1. First, a set of alternatives to the information-focus is constructed. Recall that this consists of a set of contextually relevant things which match the associate (=information-focus) in semantic type. In this example, the set in (11) will be derived. The update to file cards may require these alternatives. Based on the earlier discussion, the initial formulation of the update is the set of instructions in (12). The GO-TO instruction specifies that the file card to be updated is d2 and will not be of further concern here.</Paragraph>
    <Paragraph position="4"> (9) WWII Veteran's description of Germans' surrender to his battalion by the EIbe River: They swam lhe river lo surrender to us thousands of then*-They didn't want to surrender to TRE RUSSIANS  (10) \[n C/1 \[a d41 \[~ ax (\[L as\] want t .... r. to~)\] (11) {the Russians, the British, the Americans} (12) (i) Co-To(d2) (ii) RETRIEVE(_. want to surrender to x) (iii) R\]~LATE(~, d4, x)  Speaking generally, the update instructions express a literal information structure interpretation. That is, retrieving the ground from tile prior discourt~e or from the healer's knowledge store and supplying a value for its variable correspond closely with the questiml test whidl is often cited ,as a diagnostic for informationfocus. The ground is that part of the utterance which cordd have been all inunediately preceding question and the information-focus is the answer to that question.</Paragraph>
    <Paragraph position="5"> For example, (13) could answer the question shown but not (14). Providing the answer to a question is a very literal characterization of being informative. However, the content of the information-focus is perhaps not directly informative and the content of the ground is perhaps not directly known. Instead, in virtue of being presented ,as information-focus and ground, the function of these two constituents may be achieved indirectly through inferencing. Content which is presented as informative \[nay give rise to inferences which make it informative. Similarly for tile ground, inferences may arise which transform unknown content into known content because the information was presented by ttle speaker as being known. Given tile literal interpretation of information-focus and ground, and given particular content which is inconsistent with this literal interpretation, hearers will repair the discrepancy through inferencing. Another way to state this point is that ttle information-focus conteat, the ground content and perhaps the relation will support distinct kinds of inferences. This is the central idea of Wilson and Sperher (1979).</Paragraph>
    <Paragraph position="6"> (13) Who did they surrender to? They surrendered to TIlE ~USSIANS (14) Who surrendered to tile t~ussians? As was mentioned, the ground is an open proposition which may be presupposed in a variety of senses. Prince (1986) mentions several distinct kinds of presupposition which arise front various syntactic construetions: explicit in prior discourse, inferrable from prior discourse, currently ill hearer's consciousness alld generally known. The retrieval of an open proposition from a file card, discussed earlier as tile interpretation of ground, corresponds only to tile first kind of presupposition. One modification required, theu, is that the retrieval should be from a particular source depending on the construction used, either from the file cards, from the hearer's awareness or from tile hearer's general knowledge. Assume that these other sources are available and are also structured as tile cards. Further, assume that tile ground of canonical sentences such as (9) shmdd be retrieved as inferrable from prior discourse..</Paragraph>
    <Paragraph position="7"> For this example, the retrieval will be successful. Tile prior discourse estalfiishes that the Germans were surrendering to the speaker's battalion and from this it call be inferred that who they wanted to surrender to is under discussion.</Paragraph>
    <Paragraph position="8"> Ill many cases, the retrieval of the open proposition front the required source fails. The simplest way to address tile failure is accommodation (Lewis 1979), adding tile cxpeeted ground to the source. In some cases~ accommodation is not appropriate anti an inference nlust be made to connect the ground to the source. These inferences arise from the grormd content heeansc it is intended to be recognized a~s the grmmd, i.e. as a restatement of or a connection to the discourse context.</Paragraph>
    <Paragraph position="9"> \[u (15) (adapted from the Brown corpus), for instauce, a news conlrnentator suggests the sarcastic reply to Khrushchev. As shown in (16), the subject is the informatiun-focus. The literal interpretation of the ground is that someone playing the marimba with his shoes ill tile UN is retrievable as inferrable from the prior discourse. This retrieval will fail. Accommodating the failure would add marimba playing to the discourse as if it had been discussed earlier. But this action cannot be construed as being under discussion.</Paragraph>
    <Paragraph position="10"> Because playing the marimba with shoes is marked ms being under discussion, it requires an inference to connect it with what actually is under discussion, confrontations with Khrl, shdmv. Marimba playing must be recognized as a metaphor for table banging. In ortier to make sense of this comment, the hearer recognizes, in virtue nfthe fact that it is presented as ground contest and it cannot be accounnodated, that he does know of an incident which could be described that way.</Paragraph>
    <Paragraph position="11"> The hearer will make the analogy between playing the marimba and banging on the table with a shoe and then accounnodate the ground by assuming that the shoe hanging incident is under discussion.</Paragraph>
    <Paragraph position="12"> (15) Khrushchev: l'bols! What do you think you are doing? Cornmenta~or: WE aren't playing the marimba with our shoes in the United Natioas (16) \[R q\] \[A we\] \[,.. Ax (x play the n,arimba etc.)\] This discussion suggests sew~ral components needed to interpret the ground. Since wlmt is on the file cards are simple propositions rather than open propositions, it nmst be specitled what it means to retrieve an open t)roposition from the file card. &amp;quot;file simplest view to adopt is that it will be a proposition which matches the open one, except that the. variable is instantiated with AcrEs DE COLING-92, NAN'rES, 23-28 nour 1992 3 2 1 PROC, OF COLING-92, NANTES, AUG. 23-28, 1992 one of the alternatives or a general term such as something. What is inferrable from prior discourse needs to be specified. Further, a mechanism for rejecting accommodations and inferring another ground is needed.</Paragraph>
    <Paragraph position="13"> With these components, the ground is interpreted by: *RETRIEVE ground from source (source specified by construction) *if that fails, accommodate ground *if that fails, infer other ground and accommodate it The semantics of the information-focus and its relation to ground are reflected as one or more entries on the file card. One entry will be, in either polarity, the proposition formed by substituting the information-focus into the ground. The polarity of this entry and what other entries should be made depend on the relation. The affirmative and negative relations specify the single entry in different polarities, while the only relation specifies an entry for each alternative to the information-focus. Using examples which minimally contrast with (9), the file card entries are in (17)-(19). (17) They wanted to surrender to TIlE RUSSIANS _ want to surrender to d4 (18) They didn't want to surrender to THE RUSSIANS &amp;quot;~_ want to surrender to d4 (19) They only wanted to surrender to TIlE I:~USSIANS _ want to surrender to d4 -~_ want to surrender to dn (=tile British) -1_ want to surrender to dn (=the Americans) (20) They even wanted to surrender to TIlE RUSSIANS \[R least-likely, non-unique-~\]\[a d4\] \[r Ax(d2 want to surrender to ~)\] In addition to these logical entries, there are two kinds of impiicatures based on the relation of information-focus to ground. Becanse implicatures do not affect truth conditions, they are special entries on the file cards. That is, the embedding fimetion which determines whether the discourse is true in a model must ignore the special implicature entries. First, conventional implicatures are contributed by focus particles sudl a.s even or also. These particles are analyzed as urodifiers or comments on the logical relation. Even, for example, contributes the two modifiers shown in (20). As a result, one regular and two special entries go on the d2 file card. Second, scalar implicatures (Ilirschberg 1985) may relate the alternatives to the frame. Contextual factors determine whether scalar implicatures should be generated. For instance, if the alternatives are ranked in a relevant way, then it is implicated that stronger values than the associate do not hold. Or, if the relation of the alternatives to the ground is salient, then it is implicated that the alternatives instantiate the ground with opposite polarity. For example, (21) has the information structure shown in (22). The alternatives to the associate will be the set in (23). The relation of the alternatives to the ground has been explicitly questioned, and so ttle implication that the alternatives cannot instantiate the ground will arise.</Paragraph>
    <Paragraph position="14">  (21) Q: Do you have Lana Moro or Bernat yarns? A: We have CASSINO (cited in Hirsehberg 1985:58) (22) \[n El \[a Cassino\] \[F ),x(we have x)\] (23) {Lana M .... Bernat, Ca.ssino)  Now consider the pragmatic component of relating the information-focus to the ground. Pragmatically, tile interpretation is as the informative content. As was noted earlier, the notion of informativeness expressed by the update instructions is too literal for all eases. It is the same notion which underlies the question test as a diagnostic for information-focus. Under this analysis, the earlier example has the information structure it does because it could be used to answer the question in (24). Arguably, however, it is being used to answer the question in (25). The literal view of information-focus must be extended from supplying an unknown value to making the point of the utterance, which may require inferencing ( Wilson and Sperbet 1979). In some cases, the point will be to supply the hearer with a previously unknown value (or non-value) to the ground. In other cases, the information-focus may be literally uninformative and motivate inferences.</Paragraph>
    <Paragraph position="15"> While the general mechanism which generates the inferences that supply tile point of the utterance from the relation and information-focus is ill-understood, it is related to the inferences that specify the coherence relations discussed by Hobbs (1985) and Mann and Thompson (1987). That is, making the point of the utterance overlaps with making the utterance coherent with the immediately previous discourse.</Paragraph>
    <Paragraph position="16"> (24) Who did they want to surrender to 9.</Paragraph>
    <Paragraph position="17"> They didn't want to surrender to THE RUSSIANS (25) Why did they swim the river to surrender to us? One fairly well-understood class of these inferences is the scalar implicatures mentioned above. Note that the exchange in (21) is entirely felicitous in a setting where the clerk stands in front of a conspicuous, clearly marked display of Cassino yarn. The alternative brand's filling the open proposition is marked as informative, although in the assumed setting it cannot possibly be literally informative. The information-focus does not directly provide the information requested.</Paragraph>
    <Paragraph position="18"> Therefore, the customer infers that this was marked as the point because it is the most the clerk could say.</Paragraph>
    <Paragraph position="19"> Tile customer, based on the information-focus content Acr~ DE COLING-92, NANTES, 23-28 ^ofrr 1992 3 2 2 PROC. OF COLING-92, NANTES, AUG. 23-28, 1992 and the fact that it was marked as information-focus, infers that the direct answer to her question is negative. In short, the relation-a.ssociate-frame structures which underlie sentences receive both a semantic and a pragmatic interpretation in the procesa of discourse increment ation. The truth conditions are expressed a~ a list of predications on tile cards. The pragmatic interpretation has both a literal and au extended version. The literal interpretation is that the ground is under discussion and the information-focus' relation to it is informative. When the content of the information structure constituents cannot be construed this way in a particular context, the pragmatic interpretation is extended through inferencing. Front this perspective, the information structure serves the function of differentiating the total content into increments of information for the purpose of inferencing. Given an utterance of the same basic sentence in the same context, a shift in its information-focus will affect the iufereuces it supports. null AcrEs DE COLING-92, NANTES, 23-28 AOI~'r 1992 3 2 3 PROC. OF COLING-92, NANTES, AUG. 23-28, 1992</Paragraph>
  </Section>
class="xml-element"></Paper>