Using an Ontology to Determine English Countability

Using an Ontology to Determine English Countability Francis Bond∗ and Caitlin Vatikiotis-Bateson∗∗ * bond@cslab.kecl.ntt.co.jp ** caitlinvb@yahoo.com 2-4 Hikari-dai Seika-cho, Kyoto, Japan 619-0237 NTT Communication Science Laboratories, Nippon Telegraph and Telephone Corporation Abstract In this paper we show to what degree the count- ability of English nouns is predictable from their semantics. We found that at 78% of nouns’ countability could be predicted using an ontol- ogy of 2,710 nodes. We also show how this predictability can be used to aid non-native speakers to determine the countability of En- glish nouns when building a bilingual machine translation lexicon. 1 Introduction In English, nouns heading noun phrases are typ- ically either countable or uncountable (also called count and mass). Countable nouns can be modified by denumerators, prototypically numbers, and have a morphologically marked plural form: one dog, two dogs. Uncountable nouns cannot be modified by denumerators, but can be modified by unspecific quantifiers such as much, and do not show any number dis- tinction (prototypically being singular): * one equipment, some equipment, *two equipments. Knowledge of countability is important when translating from a source language without obligatory number and countability distinctions to a target language that does make num- ber distinctions. Some examples are Japanese- to-English (Ehara and Tanaka, 1993; Bond, 2001), Japanese-to-German (Siegel, 1996), and Chinese-to-English. For a system generating English, it is impor- tant to know the countability of the head noun, as this determines whether it can become plural, and the range of possible determiners. Knowl- edge of countability is particularly important in machine translation, because the closest trans- ∗∗This research was done while the second author was visiting the NTT Communication Science Laboratories lation equivalent may have different countabil- ity from the source noun. Many languages, such as Chinese and Japanese, do not mark count- ability, which means that the choice of count- ability will be largely the responsibility of the generation component. In this paper, we measure how well seman- tic classes predict countability. Obviously, the answer depends both on how many countability distinctions are made, and how many semantic classes are used. If every sense of every word belongs to its own semantic class, then seman- tic classes will uniquely, although not usefully, predict countability. This is effectively the po- sition taken by Wierzbicka (1988), where the semantics of a noun, given in the Natural Se- mantic Metalanguage, always provides enough information to predict the countability. On the other hand, if there are only a handful of se- mantic classes, then they will have little pre- dictive power. We first define countability, and discuss its semantic motivation (§ 2). Then we present the lexical resources used in our exper- iment (§ 3), including the ontology of 2,710 se- mantic classes. Next, we describe the experi- ment, which uses the semantic classes of words in a Japanese-to-English transfer dictionary to predict their countability (§ 4). Finally, we present the results and discuss the theoretical and practical implications in (§ 5). 2 Linguistic Background Grammatical countability is motivated …

Content truncated to 500 words