Date: Tue, 23 Aug 2005 21:01:51 -0400 From: Ruixi Ressy Ai Subject: A Theory of Ellipsis
AUTHOR: McShane, Marjorie J. TITLE: A Theory of Ellipsis PUBLISHER: Oxford University Press YEAR: 2005
Ruixi Ressy Ai, Department of Linguistics, Harvard University
In this volume, McShane develops a theory of ellipsis that is quite holistic. Unlike traditional approach, which tackles only one aspect of ellipsis, e.g., with reference either to syntax, semantics, discourse, lexical semantics, prosody or stylistics, McShane deals with them all. The approach is captured under extensible parameters-and-values. For parameters, they range over syntax, lexical semantics, morphology, pragmatics, stylistics and many others. The values for these parameters are usually of multi-valance. Take the direct object ellipsis with a like antecedent for example. One of the parameters that are related to syntax of this kind of construction can be represented as syntactic structure. The values for this parameter can be: (a) VP coordination; (b) Clausal coordination (the verbs have different subjects); (c) A clause and its elaboration; (d) Main clause and gerund phrase and (e) Main clause and subordinate clause.
As we can see from this example, the parameter is quite descriptive in nature and the values are intended to exhaust all the possibilities that the parameter can predict. This parameter, combined with many others that may or may not be in the domain of syntax, e.g., in the domain of morphology, semantics, pragmatics, discourse, stylistics and so on, can not only describe what is elided but also predict what can be (sometimes, has to) be elided. Based on all these parameters and values, we can start from one parameter, follow its value setting and reach (possibly) another parameter. Following the value setting of that parameter, we can possibly reach yet another parameter. If we repeat this cycle, we can gradually reach the conclusion that co-reference between the ellipsis under investigation and some kinds of antecedents can be of high or low probability, thus resolving ellipsis. This whole process can be developed into a kind of algorithm, with great potentials to become a real computer program that can detect, interpret and generate elliptical utterances.
The focus of this book, however, is not on developing these real computer programs. Instead, it focuses on describing linguistic facts, followed by suggested algorithms for computational linguists or computer experts. As the author defines, if 'one had to name the field to which this work belongs, it might best be called descriptive computational linguistics' (p.viii).
The ellipsis that this book covers includes: (a) syntactic ellipsis; (b) semantic ellipsis and (c) morphological ellipsis. The discussion of all these elliptical phenomena is dispersed into 13 chapters, which make up the whole book. The main target language in this book is Russian, followed by (usu. a comparison with) Polish, Czech and English.
Chapter 1 Getting Started This is the cornerstone of the whole book. Instead of detailed data description, this chapter focuses on the more theoretical (sometimes philosophical) discussions. The main topics discussed in this chapter include: (a) What is a linguistic theory; (b) Why do we need a new linguistic theory for ellipsis; (c) How can we build a holistic linguistic theory for ellipsis that can cut across language modules (i.e., across syntax, semantics, pragmatics, etc.); (d) What is the methodology employed in developing such a theory; and (e) How do we evaluate such a theory eventually?
It is based on these discussions that a theory of parameters-and-values emerges. Under this theory, parameters, with their entailed values, are drawn from various language modules (e.g., syntax, semantics, morphology, discourse, pragmatics, etc.). More significantly, these parameters (with their relevant values) can interact with each other, regardless of their background (i.e., from which language module they are drawn). It is this interaction among parameters across language modules that makes this theory for ellipsis holistic. It is also based on this approach that all the elliptical data will be described in later chapters.
This chapter also differentiates two types of ellipsis: (a) syntactic ellipsis and (b) semantic ellipsis (e.g., 'I like reading Tolstoy' = 'I like reading [books written by] Tolstoy', which is not a possible utterance in Mandarin).
The latter is of great significance to NLP, though it is less studied in literature. A preliminary (yet quite complete) inventory of all elliptical phenomena is also provided in this chapter (p.6-7).
Chapter 2 Object Ellipsis: Preliminaries This is the prelude for the next five chapters which deal exclusively with object ellipsis. This is of great significance because compared with subject and verb ellipsis, object ellipsis is not widely discussed in literature. McShane argues that the study of object ellipsis should be related to case marking. Take Russian for example, all three kinds of case assignments (i.e., configurational, lexical and semantic) are all involved in assigning case to the object. Thus, the case parameter for object ellipsis should include various cases like: (a) accusative; (b) genitive; (c) instrumental; (d) dative and (e) prepositional.
These various case-bearing null objects can be related to various antecedents that may or may not bear the same case with the relevant null object. Exactly how the null object establishes the co-reference relationship with their possible antecedents under the case cue is of primary concern for the next five chapters.
Chapter 3 Direct Object Ellipsis with a Like Antecedent This chapter considers Direct Object (DO) ellipsis with a DO antecedent. In terms of case morphology, both the elided object and its antecedent bear accusative case (ACC). The licensing condition for the DO ellipsis with a DO antecedent, however, is not restricted to the case morphology alone. Instead, it is combined with syntactic, lexico-semantic and pragmatic factors. Thus, the parameters with their entailed values will be drawn from all these domains. An example of the parameter from the syntactic domain is provided at the beginning of this synopsis. An example of the parameter from the lexico-semantic domain would be something like the nature of the selectional restrictions of the verbs in the ellipsis clause, with its values set as being typical or being narrow. The pragmatic parameter would include something like stylistic force with its values set as being neutral or emphatic. All these possible parameters and values are discussed in detail with abundant data from Russian. Polish and Czech are also compared with Russian to validate this initial inventory of parameters and values. It is shown that these parameters and values can be extensible whenever something that has not been considered has turned up. In other words, in McShane's theory, new parameters and values across language modules can always be added. The theory is thus expandable by itself. Some sample processing algorithms are also provided, showing how the description derived from a theory like this can be applied in practice. All the algorithms designed are trying to resolve the co- reference between the DO ellipsis and its possible antecedents. The larger resolution of reference is excluded from these algorithms.
Chapter 4 Direct Object Ellipsis with a Nominative Antecedent In this chapter, the DO ellipsis (in ACC) is associated with a nominative (NOM) antecedent, which can be either a nominative subject or an independent nominative topic - this by itself can be a parameter with its values in resolving the co-reference between the DO ellipsis with a NOM antecedent. Other parameters include the nature of the selecting verbs, the syntactic relation between clauses and how the topic is presented. A cross-linguistic comparison with Czech and Polish, together with a sample algorithm, is also provided.
Chapter 5 Direct Object Ellipsis with an Oblique Antecedent In this chapter, the potentially elided category is a DO with configurational ACC case marking. The antecedent is a syntactically accessible object with oblique case marking. This includes: (a) complements of verbs with genitive (GEN), dative (DAT) or instrumental (INSTR) case marking and (b) complements of prepositions with GEN, DAT, INSTR, prepositional (PREP) or (loosely) ACC case marking. Other parameters that might promote ellipsis include: (a) R-expression versus pronominal antecedent (pronoun antecedents seem to incur ellipsis more readily); (b) the nature of the clause complex; (c) narrow selectional restrictions of the ellipsis-clause verb; (d) adverbs; (e) the Assertion (A) and Elaboration (E) strategy; (f) common semantic or pragmatic context and (g) rhythm, prosody.
The interaction among these factors seems to be too complex and too general to produce any predictive algorithms at the moment.
Chapter 6 Elided Lexically Case-Marked Objects Lexically case-marked objects, which can bear GEN, INSTR, and DAT case marking in Russian can be elided in verbal repetition structures, If the verb in the antecedent clause and that in the ellipsis clause are not identical, it seems that only the (lexically case-marked) Datives can undergo ellipsis. While (lexically cased-marked) GEN and INSTR will be ambiguous between ellipsis and unexpressed objects. An algorithm to determine the co-reference between the elided lexically case-marked object and its antecedent seems to be possible and it is provided in this chapter (p.113).
Chapter 7 Unexpressed Objects That Do Not or May Not Represent Syntactic Ellipsis Missing objects can be due to: (a) syntactic ellipsis; (b) the non-selection of optional objects; (c) object non-expression triggered by modality; (d) the non-expression of generalized-human referents; and (e) the non-expression of objects in series.
These five factors (or five nodes in a broad plane) can radiate and borders among them are not that clear. Since this involves both syntactic and semantic ellipsis, not only co-reference resolution algorithms, but also reference resolution algorithms - that is, to specify 'the generalized or specific object that people understand to be referred to in all such contexts' (p.127), should be developed.
Chapter 8 Head Noun Ellipsis ... or Not? Russian is significantly different from English in that adjectives in Russian can be: (a) real adjectives that can license head noun ellipsis. This is not possible in English and instead of head noun ellipsis, one-substitution is usually employed, which is an instance of semantic ellipsis since 'one(s) must be linked to a real-world referent for a full semantic representation' (p.128); (b) diachronically substantivized adjectives (deadjectival nominals) that function as full-fledged nouns.
In the latter case, no ellipsis is involved.
Chapter 9 Verbal Ellipsis with One Licensor Four types of verbal ellipsis are discussed in this chapter: (a) Gapping; (b) Stripping; (c) Sluicing; and (d) Verb Phrase Ellipsis.
Only (a) is of great significance. Compared with Gapping in English, Gapping in Russian shows the following differences: (a) it is stylistically unrestricted; (b) morphologically realized case marking in Russian expands Gapping potential; and (c) Gapping in Russian can be conjunctionless.
Only one of two basic strategies is involved in the aforementioned four types of verbal ellipsis: either interclause parallelism ((a) and (b)) or a lexical licensor (a wh-element for (c) and an auxiliary for (d)).
Chapter 10 Verb Ellipsis with a Combination of Licensors Multilicensor Verbal Ellipsis (Multi-VE) is quite productive in Russian (limited in Polish and Czech). It refers to the verb ellipsis which is licensed by the combined semantics of the overt categories (NP argument; NP adverbial; Impersonal predicate; etc.). It is the combined semantics of these overt categories, sometimes with the help of the context that ensure the recoverability of the relevant (elided) verb meaning. Orienting the description in this domain towards processing is highly possible and necessary.
Chapter 11 Ellipsis of Minor Parts of Speech This chapter describes the ellipsis of some minor parts of speech. This includes: (a) ellipsis of conjunctions and relative pronouns; (b) ellipsis of prepositions; (c) ellipsis of conditional particles; and (d) ellipsis of reciprocal and reflexive particles.
Chapter 12 Dependencies in Ellipsis: A Polish Case Study This chapter presents a clear example of dependencies in ellipsis (i.e., the interdependence among overt and elided categories). Three elidable categories in Polish: the multifunctional particle 'sie', the conditional particle or morpheme 'by'/'-by' and direct objects with default ACC case marking, are included in the discussion.
Chapter 13 More Elliptical Phenomena The last chapter discusses: (a) syntactic ellipsis with or without co-reference; (b) semantic ellipsis; (c) unexpressed morphemes; and (d) language strategies (which is used in determining special processing algorithm for that language).
Under these discussions, the resolution of ellipsis under co-reference (which is the focus for most of the book) is now brought into the larger framework of reference resolution (i.e., linking to real-world referent).
This book has made a great contribution to the theoretical and computational study of ellipsis. First of all, the parameters-and-values approach has provided us with a bird's view towards the description of the licensing condition for ellipsis - instead of a (more traditional) worm's one. As far as I know, this is the first time that the holistic view towards the licensing condition of ellipsis has been developed into a serious linguistic theory.
Second, the parameters-and-values approach does not only give rise to a new theory for ellipsis, but also provide us with a methodology for describing ellipsis. As mentioned in the synopsis (under chapter 3), the theory is self-extensible. Thus, future work on ellipsis can be easily accommodated into this framework. The only thing we have to do is to keep on adding parameters and its entailed values into this system. In this sense, this system of parameters-and-values is dynamic in nature.
Third, the sample algorithms has provided computational linguists and computer experts a huge temptation to make them into real computer programs. This is a great contribution to NLP in terms of (co-)reference resolution.
Fourth, the book has provided a detailed inventory for ellipsis that can be hardly found anywhere else in ellipsis literature. Fifth, the book is equipped with the richest examples of the relevant phenomena that I have ever seen. All in all, the book will be of great benefit to theoretical linguists, computational linguists, field/descriptive linguists and anyone who is interested in ellipsis in Slavic languages.
I have three further comments concerning the book. First, object ellipsis has been discussed extensively in the book (6 chapters in total). This is reasonable since this is not extensively discussed in literature, compared with the study of subject and verb ellipsis. This is partially because so far, the study of ellipsis is based largely on English (as the target language), which seldom allows object ellipsis (but McShane has mentioned some exceptions in the book). Thus, the extensive discussion of object ellipsis in Russian is a great contribution. The study of object ellipsis in other languages, e.g., Mandarin: Huang (1984, 1991), Li (2002); Japanese: Hoji (1997, 1998); Korean: Kim (1999), Mandarin, Japanese and Korean: Tomioka (2003); Hebrew: Doron (1999), Goldberg (2003), has also got a lot of momentum recently. Thus, it will be of great benefit to see how McShane's theory can be technically extended to those languages which might or might not have case marking.
Second, the study of VP ellipsis can be further explicated (compared with what McShane has described in the book). It is well known that VP ellipsis may not be as universal as other types of ellipsis, e.g., Sluicing. How to account for the cross-linguistic variations and how to set those parameters in VP ellipsis is going to be the next item in the ellipsis agenda. Take Mandarin for example, it has been argued that only deontic auxiliaries/modals can license VP ellipsis (Wu, 2003).
Third, in general, the book is quite clear and pleasant to read. But the lack of alignment between the Cyrillic/transliteration Russian and the English gloss does make the reading of the examples hectic sometimes (though I understand there is a good reason for McShane to have done so).
Doron, E. (1999) "V-Movement and VP-Ellipsis", in Shalom Lappin and Elabbas Benmamoun (eds.) Fragments: Studies in Ellipsis and Gapping, New York: Oxford University Press, 124-140.
Goldberg, L. (2003) "Deriving V-stranding VP Ellipsis", paper presented at NELS-34 at Stony Brook University, New York.
Hoji, H. (1997) "Sloppy Identity and Formal Dependency", in Proceedings of WCCFL 15, Stanford, CA: CSLI Publications, 209-223.
Hoji, H. (1998) "Null Object and Sloppy Identity in Japanese", Linguistic Inquiry 29, 127-152.
Huang, C.-T. J. (1984) "On the Distribution and Reference of Empty Pronouns", Linguistic Inquiry 15.4: 531-574.
Huang, C.-T. J. (1991) "Remarks on the Status of the Null Object", in Robert Freidin (ed.) Principles and Parameters in Comparative Grammar, Cambridge, MA.: MIT Press, 56-76.
Kim, S. (1999) "Sloppy/strict Identity, Empty objects, and NP ellipsis," JEAL 8, 255-284.
Li, H.-J. G. (2002) Ellipsis Constructions in Chinese, doctoral dissertation, University of Southern California.
Tomioka, S. (2003) "The Semantics of Japanese Null Pronouns and its Cross- linguistic Implications", in Kerstin Schwabe and Susanne Winkler (eds.) The Interfaces: Deriving and Interpreting Omitted Structures, Amsterdam & Philadelphia: John Benjamins Publishing Company.
Wu, H.-H. (2003) On Ellipsis and Gapping in Mandarin Chinese, MA. thesis, National Tsing-Hua University.
ABOUT THE REVIEWER:
ABOUT THE REVIEWER
Ruixi Ressy Ai is a Teaching Fellow and PhD student at the Department of Linguistics, Harvard University. His research interests lie in the areas of East Asian Languages, Theoretical Linguistics and Computational Linguistics. Particularly, he is interested in ellipsis and event structure.