Kathleen Dahlgren

The Cognitive Structure of Social Categories

Support for the prototype theory of categorization was found in a study of the structure of social categories. Though occupational terms such as DOCTOR are socially defined, they do not have the classical structure their clear definitional origins would predict. Conceptions of social categories are richer and more complex than those of physical object categories and subjects agree upon them. Comparison of various instructions for eliciting attributes of categories showed that whether subjects are asked to define a term, give characteristics, or describe ways they recognize members of categories, the attributes they list contribute to a prototype structure. These data provide evidence against the view that prototype structure is relevant only to an identification procedure and not to the core of concepts, as has been suggested.

Kind Types in knowledge representation

This paper describes Kind Types (KT), a system which uses commonsense knowledge to reason about natural language text. KT encodes some of the knowledge underlying natural language understanding, including category distinctions and descriptions differentiating real-world objects, states and events. It embeds an ontology reflecting the ordinary persons top-level cognitive model of real-world distinctions and a database of prototype descriptions of real-world entities. KT is transportable, empirically-based and constrained for efficient reasoning in ways similar to human reasoning processes.

Discourse Coherence and Segmentation

This paper explores the basis of discourse structure, cognitive mechanisms for recovering it, and computational algorithms designed to mimic human discourse structure recovery for text. We argue that structure is recovered to the extent that the reader can build a coherent cognitive model of the eventuality (situation) the discourse describes from the reader’s interpretation of the semantic content of the discourse. In empirical studies of newspaper commentary and narrative text we found that discourse structure is infrequently marked by cue phrases, and that paragraph shift, tense shift and focus shift do not add up to sufficient information for the location and recovery of discourse segment boundaries. A discourse theory which would rely solely upon these elements could not account for intuitions of discourse structure. In contrast, an adequate theory accounts for discourse in terms of coherence in addition to the above-mentioned elements. Asher (1993) and Asher and Kamp (1995) develop a similar theory.

ITP: description of the Interpretext system as used for MUC-3

The ITP System for MUC3 is diagrammed in Figure 1. The three major modules handle different units of processing: the Message Handler processes a message unit; the ITP NLU Module processes a sentence and builds a Cognitive Model of the message; and the MUC3 Template Reasoning Module processes a segment of discourse.

SOCIAL TERMS AND SOCIAL REALITY

The diachronic semantics of individual words Iias been described äs erratic and unpredictable. Williams (1975) suggests that only three processes Iiave been identified: narrowing, widening and semantic shift. Here it will be argued that new generalizations emerge when a sociohistorical methodology is employed to trace the interaction between semantics and social history. A more refined approach to the semantics of various types of vocabulary increases the number of historical predictions which can be made. In particular, social kind terms, such äs KING and FARMER, which denote groups of people with particular functions and positions in the social hierarchy, are fundamentally different from natural kind terms. The difference originates in the essential. natures of the classes of objects they denote, and leads to significant psychological differences affecting lexical semantics. As a result, social kind terms change meaning in predictably different ways First, the distinct semantics of social kind terms will be analized briefly. Then the particular diachrony of several Anglo-Saxon terms will be traced. Finally the predictions that can be made about the likely semantic changes in social terms relative to other Mnds of terms will be summarized.

Natural Language Query in the Biochemistry and Molecular Biology Domains Based on Cognition Search

MOTIVATION With the increasing volume of scientific papers and heterogeneous nomenclature in the biomedical literature, it is apparent that an improvement over standard pattern matching available in existing search engines is required. Cognition Search Information Retrieval (CSIR) is a natural language processing (NLP) technology that possesses a large dictionary (lexicon) and large semantic databases, such that search can be based on meaning. Encoded synonymy, ontological relationships, phrases, and seeds for word sense disambiguation offer significant improvement over pattern matching. Thus, the CSIR has the right architecture to form the basis for a scientific search engine. RESULT Here we have augmented CSIR to improve access to the MEDLINE database of scientific abstracts. New biochemical, molecular biological and medical language and acronyms were introduced from curated web-based sources. The resulting system was used to interpret MEDLINE abstracts. Meaning-based search of MEDLINE abstracts yields high precision (estimated at >90%), and high recall (estimated at >90%), where synonym, ontology, phrases and sense seeds have been encoded. The present implementation can be found at http://MEDLINE.cognition.com. CONTACT Elizabeth.goldsmith@UTsouthwestern.edu Kathleen.dahlgren@cognition.com.

The Autonomy of Shallow Lexical Knowledge

The question of what is “purely linguistic” is considered in relation to the problem of modularity. A model is proposed in which parsing has access to world knowledge, and both contribute to the construction of a discourse model. The lexical semantic theory of naive semantics, which identifies word meanings with naive theories, and its use in computational text interpretation, demonstrate that a shallow, constrained layer of knowledge which is linguistic can be identified.

ITP Interpretext system: MUC-3 test results and analysis

Intelligent Text Processing is a small start-up company participating in the MUC-3 exercise for the first time this year. Our system, Interpretext, is based on a prototype text understanding system. With three full-time and three part-time people, dividing time between MUC-3 and other contract projects, ITP made maximum use of modest resources.

Word Sense Disambiguation

Computational lexical approaches to disambiguation divide into syntactic category assignment such as whether farm is a noun or a verb (Milne, 1986) and word sense disambiguation within syntactic category.9 The latter problem is the subject of this chapter. Assuming that word senses are listed together under one lexical entry in a given syntactic category, the problem is to select the correct one. One computational method of disambiguation is pattern matching where the surrounding words frequently associated with a sense are used to disambiguate a word. Such methods are powerful and can be used to eliminate 70% of the ambiguity (Black, 1986). A second method employs a rich syntactic lexicon which includes selectional restrictions (Gross, 1985). A third method uses a combination of structural and conceptual analysis for disambiguation (Black, 1986). In the present work a method is proposed which combines three types of information to disambiguate: fixed and frequent phrases, syntactic information and commonsense reasoning. It is similar to Black’s approach, but it differs in using a psycholinguistically motivated word meaning representations as the basis of a generalized disambiguation procedure. The advantage of the method is that it employs computationally expensive commonsense reasoning only for the difficult cases, and not for simpler cases.