Toward Treating English Nominals Correctly

TOWARD TREATING ENGLISH NOMINALS CORRECTLY Richard W. Sproat, Mark Y. Liberman Linguistics Department AT&T BeLl Laboratories 600 Mountain Avenue Murray Hill, NJ 07974 Abstract We describe a program for assigning correct stress contours to nominals in English. It makes use of idiosyncratic knowledge about the stress behavior of various nominal types and general knowledge about English stress rules. We have also investigated the related issue of parsing complex nominals in English. The importance of this work and related research to the problem of text-to-speech is 'discussed. I. Introduction We will discuss the analysis of English expressions consisting of a head noun preceded by one or more open.class specifiers: rising prices, horse blanket, mushroom omelet, banana bread, parish priest, gurgle detector, quarterback sneak, blind spot, red herring, bachelor's degree, Planck's constant, Madison Avenue, Wall Street, Washington's birthday sale, error correction code logic, steel industry collective bargaining agreement, expensive toxic waste cleanup, windshield wiper blade replacement, computer communications network performance analysis primer, and so forth. For brevity, we will call such expressions 'nominals.' Our main aim is an algorithm for assigning stress patterns to such nominal expressions; we will also discuss methods for parsing them. Nominals are hard to parse, since their pre-terminal string is usually consistent with all possible constituent structures, so that we seem to need an analysis of the relative plausibility of the various meanings (Marcus, 1980; Finin, 1980). Even when the constituent structure is known (as trivially in the case of binary nominals), nominal stress patterns are hard to predict, and also seem to depend on meaning (Bolinger, 1972; Fudge, 1984; Selkirk, 1984). This is a serious problem for text-to-speech algorithms, since nominal expressions are common at the ends of phrases, and the location of a phrase's last accent has a large effect on its sound. Complex nominals are common in most kinds of text; for example, in the million words of the Brown Corpus (Francis and Ku~era, 1982), there are over 73,000 nominals containing more than two words. However, we have been able to make some progress on the problems of parsing and stress assignment for nominals in unrestricted text. This paper concentrates on the representation and use of knowledge relevant to the problem of assigning stress; this same knowledge turns out to be useful in parsing. For the purposes of this paper, we will be dealing with nominals in contexts where the default stress pattern is not shifted by phenomena such as as intonational focus or contrastive stress, exemplified below: (1) a. We're only interested in solvable problems. (words like only depend on stress to set their scope m otherwise, this nominal's main stress would be on its final word.) b. He's a lion-tamer, not a lion-hunter. (in a non-contrastive context, these nominals' main stresses would be on their penultimate words.) These interesting phenomena rarely I shift main phrase stress in expository text, and are 1. In our samples, only a fraction of a percent of complex nominals in phrase-final position have their main stress shifted …

Content truncated to 500 words