Automatic extraction of paraphrastic phrases from medium-size corpora

Automatic extraction of paraphrastic phrases from medium size corpora Thierry Poibeau Laboratoire d’Informatique de Paris-Nord – CNRS UMR 7030 Av. J.B. Clément – F-93430 Villetaneuse thierry.poibeau@lipn.univ-paris13.fr Abstract This paper presents a versatile system intended to acquire paraphrastic phrases from a representative corpus. In order to decrease the time spent on the elaboration of resources for NLP system (for example Information Extraction, IE hereafter), we suggest to use a knowledge acquisition module that helps extracting new information despite linguistic variation (textual entailment). This knowledge is automatically derived from the text collection, in interaction with a large semantic network. 1 Introduction Recent researches in NLP have promoted a now widely-accepted shallow-based analysis framework that has proven to be efficient for a number of tasks, including information extraction and question answering. However, this approach often leads to over-simplified solutions to complex problems. For example, the bag-of-words approach fails in examples such as: Lee Harvey Oswald, the gunman who assassinated President John F. Kennedy, was later shot and killed by Jack Ruby (example taken from Lin and Katz, 2003). In this case, it is essential to keep track of the argument structure of the verb, to be able to infer that it is Jack Ruby and not John Kennedy who is the murderer of Lee Harvey Oswald. A wrong result would be obtained considering too shallow analysis techniques or heuristics, based for example of the proximity between two person names in the sentence. Several studies have recently proposed some approaches based on the redundancy of the web to acquire extraction patterns and semantic structures. However, these methods cannot be applied to medium size corpora. Moreover, existing structured knowledge contained in dictionaries, thesauri or semantic networks can boost the learning process by providing clear intuition over text units. In this paper, we propose a knowledge rich approach to paraphrase acquisition. We will firstly describe some related work for the acquisition of knowledge, especially paraphrases, from texts. We then describe how semantic similarity between words can be inferred from large semantic networks. We present an acquisition process, in which the semantic network is projected on the corpus to derive extraction patterns. This mechanism can be seen as a dynamic lexical tuning of information contained in the semantic network in order to generate paraphrases of an original pattern. In the last section, we propose an evaluation and some perspectives. 2 Related work This section presents some related works for the acquisition of extraction patterns and paraphrases from texts. 2.1 IE and resource acquisition IE is known to have established a now widely accepted linguistic architecture based on cascading automata and domain-specific knowledge (Appelt et al, 1993). However, several studies have outlined the problem of the definition of the resources. For example, E. Riloff (1995) says that about 1500 hours are necessary to define the resources for a text classification system on terrorism1. Most of these resources are variants of extraction patterns, which have to be manually established. 1 We estimate that the development of resources for IE …

Content truncated to 500 words