Extracting Hyponyms of Prespecified Hypernyms from Itemizations and Headings in Web Documents

Extracting Hyponyms of Prespecified Hypernyms from Itemizations and Headings in Web Documents Keiji Shinzato Kentaro Torisawa Japan Advanced Institute of Science and Technology, 1-1 Asahidai, Tatsunokuchi-machi, Nomi-gun, Ishikawa, 923-1292 JAPAN {skeiji,torisawa}@jaist.ac.jp Abstract This paper describes a method to acquire hy- ponyms for given hypernyms from HTML doc- uments on the WWW. We assume that a head- ing (or explanation) of an itemization (or list- ing) in an HTML document is likely to contain a hypernym of the items in the itemization, and we try to acquire hyponymy relations based on this assumption. Our method is obtained by ex- tending Shinzato’s method (Shinzato and Tori- sawa, 2004) where a common hypernym for ex- pressions in itemizations in HTML documents is obtained by using statistical measures. By us- ing Japanese HTML documents, we empirically show that our proposed method can obtain a sig- nificant number of hyponymy relations which would otherwise be missed by alternative meth- ods. 1 Introduction Hyponymy relations can play a crucial role in var- ious NLP systems, and there have been many at- tempts to develop automatic methods to acquire hy- ponymy relations from text corpora (Hearst, 1992; Caraballo, 1999; Imasumi, 2001; Fleischman et al., 2003; Morin and Jacquemin, 2003; Ando et al., 2003). Most of these techniques have relied on par- ticular linguistic patterns, such as “NP such as NP.” The frequencies of use for such linguistic patterns are relatively low, though, and there can be many ex- pressions that do not appear in such patterns even if we look at large corpora. The effort of searching for other clues indicating hyponymy relations is thus significant. Our aim is to extract hyponyms of prespecified hy- pernyms from the WWW. We use itemizations (or lists) in HTML documents, such as the one in Fig- ure 1(A), and their headings (‘Car Company List’ in the figure). In a similar attempt, Shinzato and Tori- sawa proposed an automatic method to obtain a com- mon hypernym of expressions in the same itemiza- tions in HTML documents (Shinzato and Torisawa, 2004) by using statistical measures such as document frequencies and inverse document frequencies. In the following, we call this method the Algorithm for Hyponymy Relation Acquisition from Itemizations (AHRAI). On the other hand, the method we propose in this paper is called Hyponym Extraction Algorithm ・ Car Company List ・ Toyota ・ Honda ・ Nissan ・ Car List ・ Toyota ・ Honda ・ Nissan (A) (B) Figure 1: Examples of itemization from Itemizations and Headings (HEAIH). The difference between AHRAI and HEAIH is that HEAIH uses the headings attached to item- izations, while AHRAI obtains hypernyms without looking at the headings. This difference has a sig- nificant consequence in the acquisition of hyponymy relations. A hyponym tends to have more than one hypernym. For instance, “Toyota” can have at least two hypernyms, “company” and “car.” AHRAI may be able to obtain “company,” for instance, from the itemizations presented in Figure 1(A), but it can- not simultaneously obtain “car.” Consider the item- ization …

Content truncated to 500 words