Sense and Deduction: The Power of Peewees Applied to the SENSEVAL-2 Swedish Lexical Sample Task

Sense and Deduction: The Power of Peewees Applied to the SENSEVAL-2 Swedish Lexical Sample Task Torbjorn Lagert and Natalia Zinovjevat tDepartment of Linguistics, Uppsala University +Hapax Information Systems AB, Stockholm Abstract This paper describes our use of Prolog Word Experts (PWEs) in the SENSEVAL-2 competi- tion. We explain how we specify our PWEs as sequences of transformation rules and how they can be trained on sense tagged corpus data. We give a semantics of PWEs by translating them into first order predicate logic, and we describe how PWEs can be compiled into Prolog pro- cedures. We finally present our results for the Swedish lexical sample task: 63% (fine-grained score) for our best PWE, and a second place in the ranking. 1 Introduction Word experts are small expert system-like mod- ules for processing a particular target word based on neighboring words. Typically, a word expert uses rules that test the identity and rela- tive position of words in the context in order to infer the role of the target word in the passage (Berleant, 1995). In this paper, we describe the development of various kinds of word experts in a logic programming framework, dealing with word sense disambiguation in the context of the SENSEVAL-2 competition. In a logic programming framework, the task of engineering a word (sense) expert can be specified as follows. Given a suitable represen- tation of a text, we want to define a predicate sense/2 such that sense (P, S) is true iff the word at position P in the text has the senseS. In the remainder of the paper, we will refer to this kind of word expert as a Pro log Word Expert (or PWE for short - "Peewee" to its friends). This is to distinguish it from other kinds of word ex- perts, and to emphasize the fact that it is 'pro- grammed in logic'. 99 2 The Anatomy of a Peewee 2.1 Peewee Specifications In the present paper, a word expert's knowledge will be expressed, not as Prolog clauses defining sense/2 directly, but as a sequence of transfor- mation rules. For example, here is how we spec- ify a word expert which is able to disambiguate occurrences of interest:1 word_expert sense := sense:add 6 <- word:interest@[O] o sense:6>1 <- word:in@[l] o sense:1>5 <- word:'/.'@[-1] o end. The first rule works as a default rule, which sim- ply assigns the most frequent sense to the word interest ( 6 in this case). If no other rules apply, this is the tag that the word will eventually get. The other rules dictate when based on the context - a word should have its tag changed. The second rule is to be read "replace the tag for sense 6 with the tag for sense 1, if the next word is in". The third rule says "replace the tag for sense 1 with the tag for sense 5, if the pre- vious 'word' is '%'." The o-symbol is a compo- sition operator, and (R o Rs) basically means …

Content truncated to 500 words