WordNet, a lexical database for English that is extensively used by computational linguists, has not previously distinguished hyponyms that are classes from hyponyms that are instances. This note describes an attempt to draw that distinction and proposes a simple way to incorporate the results into future versions of WordNet.

Of the various problems that natural language processing has revealed, polysemy is probably the most frustrating. People deal with polysemy so easily that potential abiguities are overlooked, whereas computers must work hard to do far less well. A linguistic approach generally involves a parser, a lexicon, and some ad hoc rules for using linguistic context to identify the context-appropriate sense. A statistical approach generally involves the use of word co-occurrence statistics to create a semantic hyperspace where each word, regardless of its polysemy, is represented as a single vector. Each approach has strengths and limitations; some combination is often proposed. Various possibilities will be discussed in terms of their psychological plausibility.

Two tasks involving lexical semantic sense tagging are described. Different task requirements made it necessary to select different corpora to be tagged and to develop different tagging interfaces to achieve the desired result. A vocabulary-building task required sequential tagging of connnected text, whereas a word-sense identification task required targeted tagging of many instances of common polysemous words. Advantages and drawbacks of both are compared.

How lexical information should be formulated, and how it is organized in computer memory for rapid retrieval, are central questions for computational linguists who want to create systems for language understanding. How lexical knowledge is acquired, and how it is organized in human memory for rapid retrieval during language use, are also central questions for cognitive psychologists. Some examples of psycholinguistic research on the lexical component of language are reviewed with special attention to their implications for the computational problem.

No abstract available.

This correspondence addresses itself to the question of why computers have had so little practical success in taking over information processing tasks traditionally performed by people. This question is examined in different fields of psychology, education, and linguistics. The major reason why computers have not made more progress is that twe have not had a satisfactory psychological theory of cognitive organization. Some suggestions are made as to how we might converge on a general theory of cognitive organization.

It has been shown experimentally that speech intelligibility is a function of grammatical content. This fact implies that automatic speech recognizers may well need to include information about linguistic structure.

The amount of selective information in a message can be increased either by increasing the variety of the symbols from which it is composed or by increasing the length of the message. Psychological experiments indicate that the variety of the symbols is far less important than the length of the message in controlling what human subjects are able to remember. Two messages equal in length but differing in the amount of information per symbol are equally easy to memorize. This fact provides an opportunity for the effective use of recoding procedures and reveals the mental economy involved in organizing the materials we want to remember. An apparent exception to the rule that length, not variety, is the limiting factor in human memory occurs in the case of redundant messages. If two messages of the same length differ because one contains redundancy familiar to the learner and the other does not, the redundant message will usually be easier to learn and remember. In terms of the theory of information, redundancy can be viewed equally well as a reduction in the information per symbol or as a reduction in the effective length of the message. Psychologically, however, these two alternatives are not equivalent; redundancy permits a reorganization into familiar sequences in a way that effectively shortens the length of the message and so makes it easier to memorize, but this is not psychologically equivalent to reducing the amount of information per symbol. It is as if each storage register could accept any one of a tremendous variety of alternative symbols, but the number of registers available was quite limited. If we use these registers to store binary symbols, the storage is inefficient. If we group the binary symbols into sequences, give each sequence a different name, and store the recoded names, we can make much more efficient use of the registers. Familiar redundancy is helpful because it enables us to recode more efficiently. These results for human memory are all the more striking in view of the fact that the amount of information per symbol is a critically important variable controlling the accuracy of our perceptions.