Date: Mon, 8 Sep 2003 11:34:53 +0200 From: Spenader J.K. Subject: The Construction of Meaning
Beaver, David I., Luis D. Casillas Martnez, Brady Z. Clark and Stefan Kaufmann, eds. (2002), The Construction of Meaning, Center for the Study of Language and Information, (CSLI) Publications, Stanford, California. paperback ISBN 1-57586-376-6 ix+254pp, $25.00
Reviewed by Jennifer Spenader, Center for Language and Cognition Groningen, University of Groningen, The Netherlands, firstname.lastname@example.org
DESCRIPTION This book is a selection of 11 papers written by visiting speakers to Stanford university, participants in the Construction of Meaning Workshop, and the Stanford University 'Semantics fest' day of talks. Thus, the authors are all either students or faculty at Stanford University, or visitors or guests to Stanford University working on some topic related to semantics.
Below, in alphabetical order (the same order they have in the book), I give short summary of the main points of each of the papers as I understood them.
A Resource-Sensitive Semantics for Equi and Raising Ash Asudeh
Equi-NP deletion verbs, such as 'try', and a subset of raising verbs, such as 'seems', differ in their semantics in that equi-NP deletion verbs involve structural sharing. In equi-NP deletion verbs the subject of the matrix sentence and the subject of the complement are the same, but are only represented once in the output form. Raising verbs differ in that their subject arguments are only needed for their complements. These types of verbs have been analyzed as having an empty category, either PRO or trace, in the syntactic structure. LFG doesn't allow empty categories, so both verb types get a similar c-structure analysis. However, equi-NP verbs are considered to have their subject inside their VP, a so called thematic subject, while raising verbs have their subject outside the complement verb, a non-thematic subject. Raising verbs have non-thematic subjects and do not involve shared arguments. The fact that raising verbs can take expletive subjects, while equi-verbs cannot provides further support for this analysis.
Linear logic has been used in a glue semantics in recent work in LFG. Linear logic is a resource-sensitive logic, which means that in deduction, all the premises must be used, and used only once (e.g.,'linear-logic deduction is resource sensitive: each premise is consumed as it is used and all premises must be consumed in reaching a conclusion.', p. 12). This would seem to be a problem in constructions that have shared structure, such as in equi-NP deletion constructions. Asudeh's aim is to show that this is actually not a problem for this semantics, and even in equi constructions the subject argument is only needed once. The crucial point is that the information contributed by the complement is not considered to be a proposition, but a property, i.e. in an expression about trying to leave, the property of leaving, rather than the proposition that some specific individual leaves is the contribution of the complement. By analyzing the complement as a property, the subject is only required once. Complements of equi constructions have been analyzed as properties previously by Chierchia (1985), who showed that analyzing the complement as contributing a proposition leads to incorrect deductions.
Presupposition Projection and DRT David I. Beaver
This chapter looks at the binding theory of presupposition of van der Sandt (1992) and extensions of it that treat presuppositions in Discourse Representation Theory (DRT; Kamp & Reyle, 1990). These theories consider presuppositions as anaphors at the level of representation. Presupposed information can be bound with discourse given information, or if no such information is present, it can be accommodated. Accommodation can occur at the level of the main Discourse Representation Structure (DRS), also called global accommodation, at the DRS where the presupposed material is triggered, also called local accommodation, or anywhere in between, termed intermediate accommodation.
Beaver looks at a number of examples of triggered presuppositions where the binding theory is vague on how they should be handled, or where it seems to make wrong predictions, and then he gives some suggestions for how these problems might potentially be overcome. Problematic examples include cases where the presupposition is locally entailed in an embedded context, cases where the presupposed information seems to be redundant, examples where the theory predicts binding but binding seems to lead to the wrong interpretation, as well as cases where local accommodation seems to be preferred over intermediate accommodation, in contrast with the proposed preference for higher levels of accommodation over lower. In the final section, Beaver also questions the feasibility of treating all presuppositions as anaphora, arguing that the differences found in e.g. pronouns cannot be explained convincingly by the theory.
Making Use of Pragmatic Inference in the Acquisition of Meaning Eva V. Clark
This work presents observations about children's lexical acquisition based on the author's study of longitudinal data made up of recorded conversations between six children and their parents. It shows that one way in which children learn the meaning of new words is by a process of elimination. In one example the children could learn the name of a new animal by observing that all the other toy animals except one were already familiar. When an adult referred to the animal with an unknown name, the child was able to infer that the unknown word must refer to the unfamilar animal. Clark presents several similar examples of lexical acquisition, and also examines evidence of the learning of the word in the repetitions and uptakes made by the children.
Temporal Interpretation of Modals Modals for the Present and for the Past Cleo Condoravdi
This paper looks at the intepretation of non-root modals (epistemic and metaphysical: may, might, may have, might have, etc.). In particular, 'might have' (e.g. 'He might have won the race.') can be used both with an epistemic and a metaphysical modal interpretation, and one of the aims of the paper is to explain this ambiguity.
Condoravdi argues that modals contribute directly to their temporal interpretation. The ambiguity for 'might have' is explained as a result of the time at which the property the modal is applied to is initialized. Application of the modal to a property in the present or past presupposes that the issue that the property relates to is settled, and thereby excludes a metaphysical reading - but both a metaphysical and an epistemic reading are possible when the time of instantiation is in the future.
Remarks on Evidential Hierarchies Martina Faller
Evidentiality has to do with the linguistic marking of the source of information, e.g. whether it is secondhand or thirdhand reported information or visually attested information, or the result of inference-- are all evidence types. Evidentials are the grammatical markers available in languages that code one (or more) of these evidence types. Faller compares several different hierarchies of evidentiality. The analysis is from a typological perspective, and Faller considers evidential expression in Tuyuca, Kashay and Quechuan. Faller aruges that proposed evidential hierarchies are not cross-linguistically valid. Instead she proposes a partial ordering, with a non-linear hierarchy based on the directness of the information, and further that the hierarchy should be based on evidence types, not evidentials.
Event Structure and the Perfect Paul Kiparsky
This article examines the English perfect and five uses that have been attributed to it. There are two conflicting views on these uses: the 'Reichenbachian' view is that the perfect is not polysemeous, but rather pragmatic factors distinguish between its different interpretations. Other researchers have argued that some of these uses are in fact semantically different, and in favor of this latter view is typological evidence that all the different uses can be distinguished with distinct forms in other languages.
Kiparsky argues that the resultative and existential/universal meanings are distinct and tries to unify the Reichenbachian view with this proposed semantic difference. He uses tense semantics so that event structures are mapped differently into the temporal structure. If an event continues for the entire duration of the event time, the perfect gets a universal reading. If, on the other hand, the event is contained in only part of the event time then we often get the existential reading. The resultative reading is only available with change of state predicates.
Subject-Oriented 'with'-phrases in Event Semantics David A. McKercher
This paper looks at the analysis of subject-oriented adverbials, such as 'with reluctance', 'with glee', 'reluctantly' and 'intentionally'. These adverbials are used in sentences like 'Kim ate the pudding with enthusiasm' where the enthusiasm is attributed to the subject. The author tries to give the 'with'-phrases an analysis in Event Semantics, arguing that these modifiers code a relation between individuals and an event. These types of adverbials display two unusual properties. First, when passivized, the adverbial becomes ambiguous as to whether it refers to the subject or to the object, e.g. 'John was kissed by Sue with enthusiasm' is ambiguous between an interpretation where John is enthusiastic or one where Sue is enthusiastic. Second, the use of a subject-oriented adverb induces opacity on the object and it is no longer possible to make identity substitutions. McKercher argues against an analysis proposed by Wyner (1998), which would require doubling the size of the lexicon and adding proto-role properties to logical form. Instead the author uses a lambda calculus to analyze the two readings found in the passivized forms as the result of different underlying constituent analyses. This cannot be done completely compositionally however, and McKercher suggests that another meta-language may be more apt for analyzing these constructions. Further, McKercher points out that referential opacity of subject oriented modifiers occurs in other constructions where the direct object is the stimulus for a psychological state.
Spatial Representation and Shape Classifiers in Japanese and Korean Kyonghee Paik and Francis Bond
This paper gives a brief survey of shape classifiers used in Japanese and Korean. In Japanese and Korean, referring to a certain number of objects often means using a noun that classifies that object according to shape, such as 'ni-hon no manga' (Japanese, gloss: 2-book-shape GEN comic books). These classifiers can be described according to whether they are used with items that extend in one, two or three dimensions, or according to the rigidity of the object. Japanese and Korean differ as to what aspects they consider relevant in shape classification, and as to where they draw a distinction between what is considered small, or long enough to receive a certain classification. Paik & Bond use Jackendoff's (1996) work on incorporating information about spatial relationships into lexical entries. These relationships are defined using parameters functioning in human vision. This means that there is no need to argue for relevant shape information to be a part of conceptual structure because the relevant parameters are already identified as key aspects of vision. Differences in classification systems that show up between languages, such as those between Korean and Japanese, have to do with how certain parameters are set, and where distinctions are made along those parameters, with for example, Korean treating rigidity as a relevant feature.
Does English Really Have Resumptive Quantification? And Do 'Donkey Sentences' Really Express It?
Stanley Peters and Dag Westerstål
Resumptive quantification is defined by Peters and Westerstål as 'the meaning expressed by employing a monadic quantifier to bind multiple variables simultaneously in order to quantify over tuples of the entities the variables range over' (p. 182). The main claim of the paper is that determiners in English (D-quantification) can never bind multiple variables and express resumptive quantification. However, many have argued that Donkey Sentences actually are a case of resumptive quantification by a determiner. Peters and Westerstål argue that this is incorrect, that the readings found with donkey sentences can actually all be interpreted in a way that doesn't appeal to resumptive quantification, and that resumptive quantification in English is actually only found in quantification coded by adverbs.
Extended Postposing and Focus Structure in Mandarin Locatives Shiao Wei Tham
In languages with freer word order focusing can license certain word order variations. Locative expressions can occur with the location (LOC) preceding or following the thing whose location is being described, the locatum (LCM). Tham studies how word order and phonetic focus are related to the choice of verbs in locative expressions in Mandarin Chinese.
The copula 'shi' and the verb 'you' both occur with LOC > LCM word order while 'zai' occurs with LCM > LOC word order. The author analyses the LOC > LCM orders as a right word displacement, or 'extended postposing' of the LCM argument, defining postposing as a structure where the argument is realized post-verbally rather than as more normally pre-verbally. The LOC > LCM order is alway felicitious when the LCM is focussed, and Tham proposes as a condition for this that postposed material must be focused. The choice of the verbal element has to do with the preceding discourse context. In locative sentences with LOC and LCM arguments, the new argument tends to occur post-verbally. Thus, if the location is new, it occurs post-verbally and 'zai' is the preferred verb. If the locatum is new, then it is preferrably realized post-verbally, and then 'you' or 'shi' are preferred. Tham also argues that this analysis, which attributes the difference to postposing of an argument that is focussed, is preferable to an analysis based on preposing of the thematic element, that is, preposing of the more given argument.
I Wonder What Kind of Construction That This Example Illustrates Arnold M. Zwicky,
Zwicky's chapter looks as examples of interrogative WH-constructions that have a 'that'-complementizer, sentences similar to the one in his title. He has gathered a small corpus of 27 examples from speech and compares it with a number of examples collected by Seppänen and Trotta (2000). These constructions are considered ungrammatical by some speakers, but fine by others, and Zwicky attempts to analyze their properties, showing that they generally only involve interrogative WH-constructions, only involve finite clauses, are not inverted, are mainly subordinate, must have a lexical head which is generally a noun, (but for some speakers seems to be able to be an adverb, in which case it is optional), and there is no restriction on the syntactic function that the WH+that clause can have. The only previous analysis of the construction is from Radford (1988) who argues that they are parallel to constructions in several other languages. Zwicky believes that there are several reasons to reject this analysis. His own analysis is that the constructions are the result of perception and production needs of speakers and hearers where they attempt to add to structure to prevent possible confusion with similar constructions. A hearer may, on hearing a WH-interrogative clause with a lexically headed subordinate WH-interrogative, misparse the sentence as the head having a postnominal zero relative. This misparse could have been avoided if the subordinate WH-interrogative had been explicitly marked for the relative clause. Speakers then may decided to explicitly mark the relativization with that, using a WH+that clause when using a lexically headed subordinate WH-question. Hearers hear this and may be inclined to do the same thing. This analysis explains why the construction is limited to interrogative constructions, why it is excluded from relative ones, why it always involves finite constructions, and why these constructions are mainly subordinate and do not involve inversion. It also explains why the construction only occurs with WH-expressions with lexical heads.
CRITICAL EVALUATION The papers were all well written, there are only a few minor typos in the text, and most ideas are explained compactly or references are given. The book as a whole gives the reader an overview of the surprisingly broad range of semantic work that is currently being done at Stanford University, and the reader is able to learn something about a large number of topics. In the editorial introduction by David Beaver he writes 'All the papers here manifest what I would characterize as theoretical feedom to explore what different frameworks have to offer and to allow phenomena rather than theoretical predilictions to dictate the form of analysis.' (p. ix) I certainly agree.
This same strength is, however, the biggest drawback of the book. In the directions giving by Linguist list to reviewers, they advise that when reviewing an edited collection of papers the reviewer should try to point out 'how they [the papers] go together'. As far as I can tell, the papers in the collection have only two things in common, they relate to some semantic issue, and the authors are all affiliated with or were guests at Stanford University at some time. Semantics is a large field, and even the semantics work being done at one university can vary widely, further Stanford University surely attracts a great number of visitors as well as having a large faculty and a great number of students, so these two features were not highly unifying. Examining the index of the book clearly illustrates the problem: in the six page, two column index, I was only able to find nine instances of indexed terms that appeared in more than one chapter. This is because nearly all the index terms only appear in the chapter in which they are introduced; in other words, there is very little, if any, overlap in the topics or ideas discussed in the different chapters.
Because of the wide range of topics, I found the book rather challenging to review, and because I only felt able to give critical remarks on one of the papers, I decided not to single this particular paper out and have just given a short summary of all papers. Some of the papers can be read and enjoyed without the benefit of a great deal of background (Clark, Faller, Paik & Bond, Tham and Zwicky), while the others were quite demanding, in part because they were very compact, averaging at 15 pages each.
Anyone who does indeed read all eleven papers in the book will get a good idea of exactly how great the range of semantic work is, as well as a perspective on what ideas and problems are being discussed at Stanford University.
Chierchia, Gennaro (1985). Formal semantics and the grammar of predication. Linguistic Inquiry, 16(3):417-443. van der Sandt, Rob (1992). Presupposition projection as anaphora resolution. Journal of Semantics, 9:333-377. Kamp, Hans & Uwe Reyle (1990). From Discourse to Logic: Introduction to Modeltheoretic Semantics of Natural Language, Kluwer Academic Publishers. Wyner, Adam Zachary (1998). Subject-oriented adversb are thematically dependent. In Susan Rothsetain ed., Events and Grammar, pp. 333-348. Kluwer Academic Publishers, Dordrecht, The Netherlands. Kamp, Hans (1981). A theory of truth in semantic reprsentation. Reprinted in Jeroen Groenendijk et al. (eds.), 1984, Truth, Interpretation and information, Groningen-Amsterdam Studies in Semantics (GRASS) 2, Foris, Dordrecht. Heim, Irene (1983) The Semantics of Definite and Indefinite Noun Phrases. Ph.D. thesis, University of Massachusetts, Amherst. Kanazawa, Makoto (1994). Weak vs. strong readings of donkey sentences and monotonicity inference in a dynamic setting. Linguistics and Philosophy, 17:109-158. Radford, Andrew (1988). Transformational Grammar. Cambridge University Press, Cambridge. Seppanen, Aimo and Joe Trotta (2000). The wh- + that pattern in present-day English. In John M. Kirk, ed., Corpora galore: Analyses and techniques in describing English, pp. 161-175. Rodopi, Amsterdam.
ABOUT THE REVIEWER:
ABOUT THE REVIEWER Jennifer Spenader received her Ph.D. in Computational Linguistics from Stockholm University in 2002 with a dissertation entitled 'Presuppositions in Spoken Discourse'. She is currently a post-doctoral researcher at the Center for Language and Cognition at the University of Groningen in the Netherlands, working on a project studying sluicing and other types of clausal ellipsis.