Stock of Shared Knowledge - A Tool for Solving Pronominal Anaphora

STOCK OF SHARED KNOWLEDGE - - A TOOL FOR SOLVING PRONOMINAL ANAPHORA EVA HAJlt~OVA, VLADISLAV KUBOIq and PETR KUBOIq UFAL School of Computing Science Charles University Faculty of Applied Sciences Malostransk6 n~n. 25 Simon Fraser University CS-118 00 Prague Burnaby, B.C. V5A 1S6 Czechoslovakia Canada ABSTRACT The paper develops further the idea of using the notion of the stock of shared knowledge (SSK) for anaphora resolution following a more subtle treatment of the influence of the topic/focus articulation of the sentence on the degrees of salience of items of the SSK. An algorithmic evaluation procedure of the SSK is formulated taking into account the notions of contextual boundness, syntactic associations, complexity of the sentences and existence/nonexistence of possible competitors, and a general evaluating function is proposed, essential for the process of anaphora resolution. In the present paper the analysis is performed for Czech; however, the considerations are claimed to be of a universal validity, the actual relations between different factors and the values, of course, being language-dependent. (MAX-l) than the items referred to in the focus part; (iii) a pronominal reference to an item in the topic part of the utterance keeps the activation unchanged; (iv) the items not mentioned in the given utterance subtract two degrees from their previous activation. As we stated already in the above mentioned paper, this was only a tentative solution on the way to a more sophisticated approach to organizing "common knowledge" of' the speaker and the hearer. Supported by a thorough linguistic analysis of a large amount of Czech prosaic texts Hoskovec (1989), two possible improvements to the procedure have been suggested Haji~ov~t, Hoskovec, Sgall (in press), namely: 1 .INTRODUCTION In our paper at Coling'90 we followed up the investigations presented in HajEov~i, Vrbov~ (1982) and proposed an algorithm for solving pronominal anaphora with the use of "stock of shared knowledge" (SSK) - an abstract representation of the hierarchy of salience of the items of the knowledge assumed by the speaker to be shared by him and the hearer. The changes of degrees of salience were dependent solely on the bipolar division of the sentence into its topic and focus parts, respectively. In particular, the rules for computing the degrees of salience were specified as follows: (i) the items referred to in the focus part of the utterance be it by a noun or by a stressed pronoun receive the highest degree of salience (MAX); (ii) the items referred to by a noun in the topic part of the sentence are activated one degree less (i) to replace the binary account of topic/focus articulation of the sentence by a more atomic distinction between the contextually bound and non-bound elements of the sentence, thus enriching the numerical system of possible degrees of salience; (ii) to account explicitly for the empirical observations that items mentioned throughout the discourse are more likely to be referred to than those mentioned only once. In this paper we would like to argue that other important features should he taken …

Content truncated to 500 words