A Syntax-Free Approach to Japanese Sentence Compression

Proceedings of the 47th Annual Meeting of the ACL and the 4th IJCNLP of the AFNLP, pages 826–833, Suntec, Singapore, 2-7 August 2009. c⃝2009 ACL and AFNLP A Syntax-Free Approach to Japanese Sentence Compression Tsutomu HIRAO, Jun SUZUKI and Hideki ISOZAKI NTT Communication Science Laboratories, NTT Corp. 2-4 Hikaridai, Seika-cho, Soraku-gun, Kyoto 619-0237 Japan {hirao,jun,isozaki}@cslab.kecl.ntt.co.jp Abstract Conventional sentence compression meth- ods employ a syntactic parser to compress a sentence without changing its mean- ing. However, the reference compres- sions made by humans do not always re- tain the syntactic structures of the original sentences. Moreover, for the goal of on- demand sentence compression, the time spent in the parsing stage is not negligi- ble. As an alternative to syntactic pars- ing, we propose a novel term weighting technique based on the positional infor- mation within the original sentence and a novel language model that combines statistics from the original sentence and a general corpus. Experiments that involve both human subjective evaluations and au- tomatic evaluations show that our method outperforms Hori’s method, a state-of-the- art conventional technique. Because our method does not use a syntactic parser, it is 4.3 times faster than Hori’s method. 1 Introduction In order to compress a sentence while retaining its original meaning, the subject-predicate rela- tionship of the original sentence should be pre- served after compression. In accordance with this idea, conventional sentence compression methods employ syntactic parsers. English sentences are usually analyzed by a full parser to make parse trees, and the trees are then trimmed (Knight and Marcu, 2002; Turner and Charniak, 2005; Unno et al., 2006). For Japanese, dependency trees are trimmed instead of full parse trees (Takeuchi and Matsumoto, 2001; Oguro et al., 2002; Nomoto, 2008)1 This parsing approach is reasonable be- cause the compressed output is grammatical if the 1Hereafter, we refer these compression processes as “tree trimming.” input is grammatical, but it offers only moderate compression rates. An alternative to the tree trimming approach is the sequence-oriented approach (McDonald, 2006; Nomoto, 2007; Clarke and Lapata, 2006; Hori and Furui, 2003). It treats a sentence as a se- quence of words and structural information, such as a syntactic or dependency tree, is encoded in the sequence as features. Their methods have the potential to drop arbitrary words from the original sentence without considering the boundary deter- mined by the tree structures. However, they still rely on syntactic information derived from fully parsed syntactic or dependency trees. We found that humans usually ignored the syn- tactic structures when compressing sentences. For example, in many cases, they compressed the sen- tence by dropping intermediate nodes of the syn- tactic tree derived from the source sentence. We believe that making compression strongly depen- dent on syntax is not appropriate for reproducing reference compressions. Moreover, on-demand sentence compression is made problematic by the time spent in the parsing stage. This paper proposes a syntax-free sequence- oriented sentence compression method. To main- tain the subject-predicate relationship in the com- pressed sentence and retain fluency without …

Content truncated to 500 words