Automatic analysis of semantic similarity in comparable text through syntactic tree matching

Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010), pages 752–760, Beijing, August 2010 Automatic analysis of semantic similarity in comparable text through syntactic tree matching Erwin Marsi TiCC, Tilburg University e.c.marsi@uvt.nl Emiel Krahmer TiCC, Tilburg University e.j .krahmer@uvt.nl Abstract We propose to analyse semantic similar- ity in comparable text by matching syn- tactic trees and labeling the alignments according to one of five semantic simi- larity relations. We present a Memory- based Graph Matcher (MBGM) that per- forms both tasks simultaneously as a com- bination of exhaustive pairwise classifica- tion using a memory-based learner, fol- lowed by global optimization of the align- ments using a combinatorial optimization algorithm. The method is evaluated on a monolingual treebank consisting of com- parable Dutch news texts. Results show that it performs substantially above the baseline and close to the human reference. 1 Introduction Natural languages allow us to express essentially the same underlying meaning as many alterna- tive surface forms. In other words, there are of- ten many similar ways to say the same thing. This characteristic poses a problem for many nat- ural language processing applications. Automatic summarizers, for example, typically rank sen- tences according to their informativity and then extract the top n sentences, depending on the re- quired compression rate. Although the sentences are essentially treated as independent of each other, they typically are not. Extracted sentences may have substantial semantic overlap, result- ing in unintended redundancy in the summaries. This is particularly problematic in the case of multi-document summarization, where sentences extracted from related documents are very likely to express similar information in different ways (Radev and McKeown, 1998). Therefore, if se- mantic similarity between sentences could be de- tected automatically, this would certainly help to avoid redundancy in summaries. Similar arguments can be made for many other NLP applications. Automatic duplicate and pla- giarism detection beyond obvious string overlap requires recognition of semantic similarity. Au- tomatic question-answering systems may benefit from clustering semantically similar candidate an- swers. Intelligent document merging software, which supports a minimal but lossless merge of several revisions of the same text, must handle cases of paraphrasing, restructuring, compression, etc. Recognizing textual entailments (Dagan et al., 2005) could arguably be seen as a specific in- stance of detecting semantic similarity. In addition to merely detecting semantic simi- larity, we can ask to what extent two expressions share meaning. For instance, the meaning of one sentence can be fully contained in that of another, the meaning of one sentence can overlap only partly with that of another, etc. This requires an analysis of the semantic similarity between a pair of expressions. Like detection, automatic analy- sis of semantic similarity can play an important role in NLP applications. To return to the case of multi-document summarization, analysing the semantic similarity between sentences extracted from different documents provides the basis for sentence fusion, a process where a new sentence is generated that conveys all common information from both sentences without introducing redun- dancy (Barzilay and …

Content truncated to 500 words