A Quantitative Evaluation of Linguistic Tests for the Automatic Prediction of Semantic Markedness

A Quantitative Evaluation of Linguistic Tests for the Automatic Prediction of Semantic Markedness Vasileios Hatzivassiloglou and Kathleen McKeown Department of Computer Science 450 Computer Science Building Columbia University New York, N.Y. 10027 {vh, kathy}~cs, columbia, edu Abstract We present a corpus-based study of methods that have been proposed in the linguistics liter- ature for selecting the semantically unmarked term out of a pair of antonymous adjectives. Solutions to this problem are applicable to the more general task of selecting the positive term from the pair. Using automatically collected data, the accuracy and applicability of each method is quantified, and a statistical analysis of the significance of the results is performed. We show that some simple methods are indeed good indicators for the answer to the problem while other proposed methods fail to perform better than would be attributable to chance. In addition, one of the simplest methods, text frequency, dominates all others. We also ap- ply two generic statistical learning methods for combining the indications of the individual methods, and compare their performance to the simple methods. The most sophisticated complex learning method offers a small, but statistically significant, improvement over the original tests. 1 Introduction The concept of markedness originated in the work of Prague School linguists (Jakobson, 1984a) and refers to relationships between two complementary or antonymous terms which can be distinguished by the presence or absence of a feature (+A versus --A). Such an opposition can occur at various linguistic levels. For example, a markedness contrast can arise at the morphology level, when one of the two words is derived from the other and therefore contains an explicit formal marker such as a prefix; e.g., prof- itable-unprofitable. Markedness contrasts also ap- pear at the semantic level in many pairs of grad- able antonymous adjectives, especially scalar ones (Levinson, 1983), such as tall-short. The marked and unmarked elements of such pairs function in dif- ferent ways. The unmarked adjective (e.g., tall) can be used in how-questions to refer to the property de- scribed by both adjectives in the pair (e.g., height), but without any implication about the modified item relative to the norm for the property. For exam- ple, the question How tall is Jack? can be answered equally well by four or seven feet. In contrast, the marked element of the opposition cannot be used generically; when used in a how-question, it implies a presupposition of the speaker regarding the rela- tive position of the modified item on the adjectival scale. Thus, the corresponding question using the marked term of the opposition (How short is Jack?) conveys an implication on the part of the speaker that Jack is indeed short; the distinguishing feature A expresses this presupposition. While markedness has been described in terms of a distinguishing feature A, its definition does not specify the type of this feature. Consequently, sev- eral different types of features have been employed, which has led into some confusion about the meaning of the term markedness. Following Lyons (1977), we distinguish …

Content truncated to 500 words