Statistical Sentence Condensation using Ambiguity Packing and Stochastic Disambiguation Methods for Lexical-Functional Grammar

Statistical Sentence Condensation using Ambiguity Packing and Stochastic Disambiguation Methods for Lexical-Functional Grammar Stefan Riezler and Tracy H. King and Richard Crouch and Annie Zaenen Palo Alto Research Center, 3333 Coyote Hill Rd., Palo Alto, CA 94304 {riezler|thking|crouch|zaenen}@parc.com Abstract We present an application of ambiguity pack- ing and stochastic disambiguation techniques for Lexical-Functional Grammars (LFG) to the domain of sentence condensation. Our system incorporates a linguistic parser/generator for LFG, a transfer component for parse reduc- tion operating on packed parse forests, and a maximum-entropy model for stochastic output selection. Furthermore, we propose the use of standard parser evaluation methods for auto- matically evaluating the summarization qual- ity of sentence condensation systems. An ex- perimental evaluation of summarization qual- ity shows a close correlation between the au- tomatic parse-based evaluation and a manual evaluation of generated strings. Overall sum- marization quality of the proposed system is state-of-the-art, with guaranteed grammatical- ity of the system output due to the use of a constraint-based parser/generator. 1 Introduction Recent work in statistical text summarization has put for- ward systems that do not merely extract and concate- nate sentences, but learn how to generate new sentences from ⟨Summary, Text⟩ tuples. Depending on the cho- sen task, such systems either generate single-sentence “headlines” for multi-sentence text (Witbrock and Mittal, 1999), or they provide a sentence condensation module designed for combination with sentence extraction sys- tems (Knight and Marcu, 2000; Jing, 2000). The chal- lenge for such systems is to guarantee the grammatical- ity and summarization quality of the system output, i.e. the generated sentences need to be syntactically well- formed and need to retain the most salient information of the original document. For example a sentence extraction system might choose a sentence like: The UNIX operating system, with implementations from Apples to Crays, appears to have the advan- tage. from a document, which could be condensed as: UNIX appears to have the advantage. In the approach of Witbrock and Mittal (1999), selec- tion and ordering of summary terms is based on bag- of-words models and n-grams. Such models may well produce summaries that are indicative of the original’s content; however, n-gram models seem to be insufficient to guarantee grammatical well-formedness of the system output. To overcome this problem, linguistic parsing and generation systems are used in the sentence condensation approaches of Knight and Marcu (2000) and Jing (2000). In these approaches, decisions about which material to in- clude/delete in the sentence summaries do not rely on rel- ative frequency information on words, but rather on prob- ability models of subtree deletions that are learned from a corpus of parses for sentences and their summaries. A related area where linguistic parsing systems have been applied successfully is sentence simplifica- tion. Grefenstette (1998) presented a sentence reduction method that is based on finite-state technology for lin- guistic markup and selection, and Carroll et al. (1998) present a sentence simplification system based on linguis- tic parsing. However, these approaches do not employ statistical learning techniques to disambiguate simplifi- cation …

Content truncated to 500 words