Quantifier Scope Disambiguation Using Extracted Pragmatic Knowledge: Preliminary Results

Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 1465–1474, Singapore, 6-7 August 2009. c⃝2009 ACL and AFNLP Quantifier Scope Disambiguation Using Extracted Pragmatic Knowledge: Preliminary Results Prakash Srinivasan Temple University 1805 N. Broad St. Wachman Hall 324 Philadelphia, PA 19122 prakash.srinivasan@temple.edu Alexander Yates Temple University 1805 N. Broad St. Wachman Hall 324 Philadelphia, PA 19122 yates@temple.edu Abstract It is well known that pragmatic knowl- edge is useful and necessary in many dif- ficult language processing tasks, but be- cause this knowledge is difficult to acquire and process automatically, it is rarely used. We present an open information extrac- tion technique for automatically extracting a particular kind of pragmatic knowledge from text, and we show how to integrate the knowledge into a Markov Logic Net- work model for quantifier scope disam- biguation. Our model improves quantifier scope judgments in experiments. 1 Introduction It has long been a goal of the natural language processing (NLP) community to be able to inter- pret language utterances into logical representa- tions of their meaning. Quantifier scope ambigu- ity has been recognized as one particularly chal- lenging aspect of this problem. For example, the following sentence has two possible readings, de- pending on the scope of its quantifiers: Every boy wants a dog. One reading of this sentence is that there exists a single dog in the world which all boys want. The second, and usually preferred, reading is that the sentence is describing a separate “wanting” rela- tion for each boy, and that the dog in question is a function of the boy who wants it. In this read- ing, there may be as many different dogs as boys, although it leaves open the possibility that several of the boys want the same dog. In logic, these two readings can be represented as follows: 1. ∃d∈Dogs ∀b∈Boys wants(b, d) 2. ∀b∈Boys ∃d∈Dogs wants(b, d) The readings differ only in the order of the quanti- fiers. The quantifier that comes first in each ex- pression is said to have wide scope; the second quantifier has narrow scope. Linguists and NLP researchers have come up with several theories and mechanisms for automat- ically determining the scope of quantified linguis- tic expressions. Despite a long history of proposed solutions, however, researchers have for the most part abandoned this task as hopeless because of “overwhelming evidence suggesting that quanti- fier scope is a phenomenon that must be treated at the pragmatic level” (Saba and Corriveau, 2001). For example, in active voice clauses, the quantifier for the subject noun is usually preferred for wide scope over the quantifier of the predicate noun (Kurtzman and MacDonald, 1993). But such pref- erences can easily be overruled by world knowl- edge: A doctor lives in every city. 1. ∃d∈Docs ∀c∈Cities lives in(d, c) 2.(A single doctor lives in all cities.) ∀c∈Cities ∃d∈Docs lives in(d, c) (Each city has a different doctor living there.) Syntactic preferences would normally indicate that reading 1 is better, but in this particular case …

Content truncated to 500 words