Search

The need for deeper semantic processing of human language by our natural language processing systems is evidenced by their still-unreliable performance on inferencing tasks, even using deep learning techniques. These tasks require the detection of subtle interactions between participants in events, of sequencing of subevents that are often not explicitly mentioned, and of changes to various participants across an event. Human beings can perform this detection even when sparse lexical items are involved, suggesting that linguistic insights into these abilities could improve NLP performance. In this article, we describe new, hand-crafted semantic representations for the lexical resource VerbNet that draw heavily on the linguistic theories about subevent semantics in the Generative Lexicon (GL). VerbNet defines classes of verbs based on both their semantic and syntactic similarities, paying particular attention to shared diathesis alternations. For each class of verbs, VerbNet provides common semantic roles and typical syntactic patterns. For each syntactic pattern in a class, VerbNet defines a detailed semantic representation that traces the event participants from their initial states, through any changes and into their resulting states. The Generative Lexicon guided the structure of these representations. In GL, event structure has been integrated with dynamic semantic models in order to represent the attribute modified in the course of the event (the location of the moving entity, the extent of a created or destroyed entity, etc.) as a sequence of states related to time points or intervals. We applied that model to VerbNet semantic representations, using a class's semantic roles and a set of predicates defined across classes as components in each subevent. We will describe in detail the structure of these representations, the underlying theory that guides them, and the definition and use of the predicates. We will also evaluate the effectiveness of this resource for NLP by reviewing efforts to use the semantic representations in NLP tasks.

Dalrymple and Lødrup (2000) propose that (tensed) clausal complements have the OBJ function in some languages, the COMP function in others and both functions in still another set. They treat Swedish as a mixed language. In this chapter, Zaenen and Engdahl look more closely at Swedish data discussed in the chapter and supplement them as more electronic data have become available and some of the criteria used have been refined (as is also documented in more recent work by Lødrup (2002), Lødrup (2004), and Lødrup (2012) for Norwegian). Their investigation also illustrates that two related issues need further study: the status of agent phrases in the various forms of passive in Swedish and the status of DP complements in some of the alternations with sentential complements.

This paper announces the release of a new version of the English lexical resource VerbNet with substantially revised semantic representations designed to facilitate computer planning and reasoning based on human language. We use the transfer of possession and transfer of information event representations to illustrate both the general framework of the representations and the types of nuances the new representations can capture. These representations use a Generative Lexicon-inspired subevent structure to track attributes of event participants across time, highlighting oppositions and temporal and causal relations among the subevents.

We review the various non-canonical positions in which the thematically highest argument can occur in Icelandic and in Swedish. We show that NPs in these non-canonical positions have subject properties in both languages. We summarize the restrictions that we are aware of holding on the various positions and discuss whether they are configurational or thematic/semantic in nature.

Hearers and readers make inferences on the basis of what they hear or read. These inferences are partly determined by the linguistic form that the writer or speaker chooses to give to her utterance. The inferences can be about the state of the world that the speaker or writer wants the hearer or reader to conclude are pertinent, or they can be about the attitude of the speaker or writer vis-à-vis this state of affairs. The attention here goes to the inferences of the first type. Research in semantics and pragmatics has isolated a number of linguistic phenomena that make specific contributions to the process of inference. Broadly, entailments of asserted material, presuppositions (e.g., factive constructions), and invited inferences (especially scalar implicatures) can be distinguished. While we make these inferences all the time, they have been studied piecemeal only in theoretical linguistics. When attempts are made to build natural language understanding systems, the need for a more systematic and wholesale approach to the problem is felt. Some of the approaches developed in Natural Language Processing are based on linguistic insights, whereas others use methods that do not require (full) semantic analysis. In this article, I give an overview of the main linguistic issues and of a variety of computational approaches, especially those stimulated by the RTE challenges first proposed in 2004.

In this chapter, we describe the method and process of transforming the theoretical formulations of a linguistic phenomenon, based on empirical observations, into a model that can be used for the development of a language annotation specification. We outline this procedure generally, and then examine the steps in detail by specific example. We look at how this methodology has been implemented in the creation of TimeML (and ISO-TimeML), a broad-based standard for annotating temporal information in natural language texts. Because of the scope of this effort and the richness of the theoretical work in the area, the development of TimeML illustrates very clearly the methodology of the early stages of the MATTER annotation cycle, where initial models and schemas cycle through progressively mature versions of the resulting specification. Furthermore, the subsequent effort to convert TimeML into an ISO compliant standard, ISO-TimeML, demonstrates the utility of the CASCADES model in distinguishing between the concrete syntax of the schema and abstract syntax of the model behind it.

Up to rather recently Natural Language Processing has not given much attention to modality. As long as the main task was to determined what a text was about (Information Retrieval) or who the participants in an eventuality were (Information Extraction), this neglect was understandable. With the focus moving to questions of natural language understanding and inferencing as well as to sentiment and opinion analysis, it becomes necessary to distinguish between actual and envisioned eventualities and to draw conclusions about the attitude of the writer or speaker towards the eventualities referred to. This means, i.a., to be able to distinguish ‘John went to Paris’ and ‘John wanted to go to Paris’. To do this one has to calculate the effect of different linguistic operators on the eventuality predication.

Treebanks are language resources that provide annotations at various levels of linguistic structure starting from the word level. They typically provide syntactic constituent or dependency structures for sentences, but increasingly extend to annotation beyond syntactic structure, including semantic, pragmatic and rhetorical annotation, or go beyond a single language, as in parallel treebanks. Experience in building treebanks has shown that there is a close relation between formal linguistic theory and the design and practice of annotation. With increasing complexity of annotations, the design of annotation schemes becomes more and more theory-dependent. At the same time, linguistically motivated treebank annotations have become crucially important for the development of data-driven approaches to natural language processing and for linguistic research in general. Treebanks therefore constitute an important link between linguistic theory and computational linguistics. The International Workshop on Treebanks and Linguistic Theories provides a forum for researchers working on treebanks from both perspectives. The present volume presents the contents of the 10th edition of this workshop series, held in 2012 at the University of Heidelberg.

In addition to describing facts and events, texts often communicate information about the attitude of the writer or various participants towards material being described. The most salient clues about attitude are provided by the lexical choice of the writer but, as discussed below, the organization of the text also contributes information relevant to assessing attitude. We argue that the current work in this area that concentrates mainly on the negative or positive attitude communicated by individual terms (Edmonds and Hirst, 2002; Hatzivassiloglou and McKeown, 1997; Turney and Littman, 2002; Wiebe et al., 2001) is incomplete and often gives the wrong results when implemented directly. We then describe how the base attitudinal valence of a lexical item is modified by lexical and discourse context and propose a simple, “proof of concept” implementation for some contextual shifters.

This paper argues that local textual inferences come in three well-defined varieties (entailments, conventional implicatures/presuppositions, and conversational implicatures) and one less clearly defined one, generally available world knowledge. Based on this taxonomy, it discusses some of the examples in the Pascal text suite and shows that these examples do not fall into any of them. It proposes to enlarge the test suite with examples that are more directly related to the inference patterns discussed.

We present an application of ambiguity packing and stochastic disambiguation techniques for Lexical-Functional Grammars (LFG) to the domain of sentence condensation. Our system incorporates a linguistic parser/generator for LFG, a transfer component for parse reduction operating on packed parse forests, and a maximum-entropy model for stochastic output selection. Furthermore, we propose the use of standard parser evaluation methods for automatically evaluating the summarization quality of sentence condensation systems. An experimental evaluation of summarization quality shows a close correlation between the automatic parse-based evaluation and a manual evaluation of generated strings. Overall summarization quality of the proposed system is state-of-the-art, with guaranteed grammaticality of the system output due to the use of a constraint-based parser/generator.

G. Nunberg et A. Zaenen : La polysémie systématique dans la description lexicale The phenomenon of systematic polysemy offers a fruitful domain for examining the theoretical differences between lexicological and lexicographic approaches to description. We consider here the process that provides for systematic conversion of count to mass nouns in English (a chicken /chicken, an oak / oak etc.). From the point of view of lexicology, we argue, standard syntactic and pragmatic tests suggest the phenomenon should be described by means of a single unindividuated transfer function that does not distinguish between interpretations (rabbit = « meat » vs. « fur »). From the point of view of lexicography, however, these more precise interpretations are made part of explicit description via the inclusion of semantic « licences », a mechanism distinct from lexical rules., Nunberg Geoffrey, Zaenen Annie. La polysémie systématique dans la description lexicale. In: Langue française, n°113, 1997. Aux sources de la polysémie nominale, sous la direction de Pierre Cadiot et Benoît Habert. pp. 12-23.

A. Zaenen & M. Dalrymple : The syntactic analysis of French causative constructions relies on the distinction, crucial in the LFG formalism, between the constituent structure (hierarchical : faire takes a VP complement) and the functional structure (flat : faire and the infinitival V form a complex predicate in a unique functional structure). The semantic representation is constructed from the f-structure, lexical semantic information (which specifies semantic roles associated with each predicate), and independent rules which map functions onto semantic roles. This construction takes the general form of a deduction in linear logic, whose property of resource sensitivity is exploited to give the effect of completeness and coherence, which are general constraints on linguistic structures., Zaenen Annie, Dalrymple Marie. Les verbes causatifs « polymorphiques » : les prédicats complexes en français. In: Langages, 30ᵉ année, n°122, 1996. Nouveaux raisonnements syntaxiques, sous la direction de Anne Abeillé et Danièle Godard. pp. 79-95.

The relation between the meaning and the syntax of lexical items is among the more frustrating issues in linguistics: on the one hand it seems clear that the meaning of a lexical item determines at least to some degree the syntactic behavior of its participant roles; on the other hand, attempts to characterize the relation explicitly tend not to be very successful. In this paper I look at the unaccusative/unergative distinction in Dutch as an example of a distinction for which it is felt that some semantic correlates should be available (e.g. in traditional grammar, Pollmann, 1975; Perlmutter, 1978; Zaenen, 1988) but for which it has also been argued that no semantic criteria can be found (Hoekstra, 1984; and more generally, Rosen, 1984). Rosen (1984) remarks a propos of semantic criteria for unaccusativity that “one cannot state a semantic criterion that actually works: not animacy of the argument, not agentive or volitional meaning, not existential or presentential meaning”. I will show that even this state of affairs does not mean that one has to reject all forms semantic grounding of a particular syntactic behavior. Following Pesetsky (1987), I will call the hypothesis that such grounding can be found, the “transparency hypothesis”.

Systems and methods for analyzing speech containing at least one lexical item, said analysis to determine an attitude of a speaker towards an entity, comprising determining at least one actual valence for the at least one lexical item by analyzing the at least one lexical item in context; determining the attitude based on the at least one actual valence; associating the speaker, the entity and the attitude; and wherein the at least one lexical item encodes attitude information about the entity.