While there has been much research into using selectional preferences for word sense disambiguation (WSD), much difficulty has been encountered. To facilitate study into this difficulty and aid in WSD in general, a database of the selectional preferences of non-pertainym prenomial adjectives extracted from the Google Web 1T 5-gram Corpus is proposed. A variety of methods for computing the preferences of each adjective over a set of noun categories from WordNet have been evaluated via simulated disambiguation of pseudohomonyms. The best method of these involves computing for each noun category the ratio of single-word common (i.e. not proper) noun lemma types which can co-occur with a given adjective to the number of single-word common noun lemmata whose estimated frequency is greater than a threshold based on the frequency of the adjective. The database produced by this procedure will be made available to the public.

We describe the automatic construction of a semantic network1, in which over 3000 of the most frequently occurring monosemous nouns2 in Wikipedia (each appearing between 1,500 and 100,000 times) are linked to their semantically related concepts in the WordNet noun ontology. Relatedness between nouns is discovered automatically from co-occurrence in Wikipedia texts using an information theoretic inspired measure. Our algorithm then capitalizes on salient sense clustering among related nouns to automatically disambiguate them to their appropriate senses (i.e., concepts). Through the act of disambiguation, we begin to accumulate relatedness data for concepts denoted by polysemous nouns, as well. The resultant concept-to-concept associations, covering 17,543 nouns, and 27,312 distinct senses among them, constitute a large-scale semantic network of related concepts that can be conceived of as augmenting the WordNet noun ontology with related-to links.

This paper describes a linguistic text mining tool for analyzing problem reports in aerospace engineering and safety organizations. The Semantic Trend Analysis Tool (STAT) helps analysts find and review recurrences, similarities and trends in problem reports. The tool is being used to analyze engineering discrepancy reports at NASA Johnson Space Center. The tool has been augmented with a statistical natural language parser that also resolves parsing gaps and identifies verb arguments and adjuncts. The tool uses an aerospace ontology augmented with features of taxonomies and thesauruses. The ontology defines hierarchies of problem types, equipment types and function types. STAT uses the output of the parser and the aerospace ontology to identify words and phrases in problem report descriptions that refer to types of hazards, equipment damage, performance deviations or functional impairments. Tool performance has been evaluated on 120 problem descriptions from problem reports, with encouraging results.

A method for the definition of verb predicates is proposed. The definition of the predicates is essentially tied to a semantic interpretation algorithm that determines the predicate for the verb, its semantic roles and adjuncts. As predicate definitions are complete, they can be tested by running the algorithm on some sentences and verifying the resolution of the predicate, semantic roles and adjuncts in those sentences. The predicates are defined semiautomatically with the help of a software environment that uses several sections of a corpus to provide feedback for the definition of the predicates, and then for the subsequent testing and refining of the definitions. The method is very flexible in adding a new predicate to a list of already defined predicates for a given verb. The method builds on an existing approach that defines predicates for WordNet verb classes, and that plans to define predicates for every English verb. The definitions of the predicates and the semantic interpretation algorithm are being used to automatically create a corpus of annotated verb predicates, semantic roles and adjuncts.

A method has been developed and implemented that assigns syntactic roles to commas. Text that has been tagged using a part-of-speech tagger serves as the input to the system. A set of Finite State Automata first assigns temporary syntactic roles to each comma in the sentence. A greedy learning algorithm is then used to determine the final syntactic roles of the commas. The system requires no training and is not domain specific. The performance of the system on numerous corpora is given.

An algorithm for semantic interpretation that integrates the determination of the meaning of verbs, the attachment and meaning of prepositions, and the determination of thematic roles is presented. The parser does not resolve structural ambiguity, which is solely the task of the semantic interpreter. Lexical semantic information about nouns and verbs is applied to the resolution of verb polysemy and modifier attachment. Semantic interpretation is centered on the representation of the meaning of the verb, called verbal concept. Verbal concepts are organized into a classification hierarchy. As long as the meaning of the verb remains unknown, parsing proceeds on a syntactic basis. Once the meaning of the verb is recognized, the semantic component makes sense of the syntactic relations built so far by the parser and of those still to be parsed. The algorithm has been implemented and tested on real–world texts.

In recent years, natural language processing (NLP) has experienced a dramatic research shift by focusing on the processing of real-world texts, rather than on restricted domains. This shift of focus has been an acid test for the core components of NLP, namely parsing and semantic interpretation. For the last few years, we have been working on knowledge acquisition from texts (F. Gomez, 1985; F. Gomez and C. Segani, 1989). The research started as a set of theoretical ideas and, then, gradually we built a system that embodies the theory. The system could be at first called a "toy system", that works in restricted domains. Recently, we have extended every component of the system to handle "real-world" texts. The model has been implemented in a program that reads unedited texts from The World Book Encyclopedia (1994), and acquires new concepts and conceptual relations about topics dealing with the dietary habits of animals, their classifications and habitats. The program is also able to answer an ample set of questions about the database that it has automatically acquired. Some of the major lessons that derive from the research are presented.>

A method for representing texts describing biological systems is presented. The representation is that which an educated layman would build from reading a text describing a biological system. This representation is viewed as a bridge between the English description of a biological system and deeper representations of such a system. The representation is based on temporal relations which are used to index and organize knowledge about biological systems (circulatory, respiratory, etc.) in a hierarchy of events. Several reasoning methods which the proposed representation permits are explained. A method for drawing diagrams depicting the function of the systems represented is also included.>

A representation formalism for N-ary relations, quantification, and definition of concepts is described. Three types of conditions are associated with the concepts: (1) necessary and sufficient properties, (2) contingent properties, and (3) necessary properties. Also explained is how complex chains of inferences can be accomplished by representing existentially quantified sentences, and concepts denoted by restrictive relative clauses as classification hierarchies. The representation structures that make possible the inferences are explained first, followed by the reasoning algorithms that draw the inferences from the knowledge structures. All the ideas explained have been implemented and are part of the information retrieval component of a program called Snowy. An appendix contains a brief session with the program.>

In understanding a novel scientific text, we may distinguish the following processes. First, concepts are built from the logical form of the sentence into the final knowledge structures. This is called concept formation. While these concepts are being formed, they are also being recognized by checking whether they are already in long-term memory (LTM). Then, those concepts which are unrecognized are integrated in LTM. In this paper, algorithms for the recognition and integration of concepts in scientific texts are presented. It is shown that the integration of concepts in understanding scientific texts is essentially a classification task, which determines how and where to integrate them in LTM. In some cases, the integration of concepts results in a reclassification of some of the concepts already stored in LTM. All the algorithms described here have been implemented and are part of SNOWY, a program which reads short scientific paragraphs and answers questions.

An automatic programming system must embody not only a methodology for program synthesis, but also a formalism whereby problems can be specified by the user. The authors have designed TQ as a language for specifying programming problems and built a problem solver based on the methodology of classification problem solving. TQ is an informal specification language that allows a problem description to be written in nonprocedural terms by users who have no knowledge of programming. TOAD, the prototype system, demonstrates the feasibility and potential of the proposed approach and its suitability for end users. The authors compare their approach to those of other researchers in automatic programming and to the fourth generation languages.>

An outline of a theory of comprehension of declarative contexts is presented. The main aspect of the theory being developed is based on Kant's distinction between concepts as rules (we have called them conceptual specialists) and concepts as an abstract representation (schemata, frames). Comprehension is viewed as a process dependent on the conceptual specialists (they contain the inferential knowledge), the schemata or frames (they contain the declarative knowledge), and a parser. The function of the parser is to produce a segmentation of the sentences in a case frame structure, thus determininig the meaning of prepositions, polysemous verbs, noun group etc. The function of this parser is not to produce an output to be interpreted by semantic routines or an interpreter, but to start the parsing process and proceed until a concept relevant to the theme of the text is recognized. Then the concept takes control of the comprehension process overriding the lower level linguistic process. Hence comprehension is viewed as a process in which high level sources of knowledge (concepts) override lower level linguistic processes.

A diagnostician, when he arrives at a diagnosis or diagnoses, has invoked some concepts. They can be diseases, causes of them, or other notions that are relevant to the diagnosis. These concepts form a hierarchical structure similar to a botanical or zoological classification. The diagnostician's knowledge is distributed through this hierarchy. The concepts in the hierarchy provide the criteria to organize under them small pieces of knowledge represented in the form of production rules. Thus concepts may be viewed as clusters of production rules. They extend the capabilities of production rules to more complex problem solving situations. The rules under each concept are further organized into three subgroups: exclusionary, confirmatory, and recommendation rules. During the problem solving process, the concepts are taken as specialists. They interact and communicate among themselves by means of a blackboard.