Orthonormal Explicit Topic Analysis for Cross-Lingual Document Matching

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1732–1740, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Orthonormal Explicit Topic Analysis for Cross-lingual Document Matching John Philip McCrae University Bielefeld Inspiration 1 Bielefeld, Germany Philipp Cimiano University Bielefeld Inspiration 1 Bielefeld, Germany {jmccrae,cimiano,rklinger}@cit-ec.uni-bielefeld.de Roman Klinger University Bielefeld Inspiration 1 Bielefeld, Germany Abstract Cross-lingual topic modelling has applications in machine translation, word sense disam- biguation and terminology alignment. Multi- lingual extensions of approaches based on la- tent (LSI), generative (LDA, PLSI) as well as explicit (ESA) topic modelling can induce an interlingual topic space allowing documents in different languages to be mapped into the same space and thus to be compared across languages. In this paper, we present a novel approach that combines latent and explicit topic modelling approaches in the sense that it builds on a set of explicitly defined top- ics, but then computes latent relations between these. Thus, the method combines the ben- efits of both explicit and latent topic mod- elling approaches. We show that on a cross- lingual mate retrieval task, our model signif- icantly outperforms LDA, LSI, and ESA, as well as a baseline that translates every word in a document into the target language. 1 Introduction Cross-lingual document matching is the task of, given a query document in some source language, estimating the similarity to a document in some tar- get language. This task has important applications in machine translation (Palmer et al., 1998; Tam et al., 2007), word sense disambiguation (Li et al., 2010) and ontology alignment (Spiliopoulos et al., 2007). An approach that has become quite popular in re- cent years for cross-lingual document matching is Explicit Semantics Analysis (ESA, Gabrilovich and Markovitch (2007)) and its cross-lingual extension CL-ESA (Sorg and Cimiano, 2008). ESA indexes documents by mapping them into a topic space de- fined by their similarity to predefined explicit top- ics – generally articles from an encyclopaedia – in such a way that there is a one-to-one correspondence between topics and encyclopedic entries. CL-ESA extends this to the multilingual case by exploiting a background document collection that is aligned across languages, such as Wikipedia. A feature of ESA and its extension CL-ESA is that, in contrast to latent (e.g. LSI, Deerwester et al. (1990)) or genera- tive topic models (such as LDA, Blei et al. (2003)), it requires no training and, nevertheless, has been demonstrated to outperform LSI and LDA on cross- lingual retrieval tasks (Cimiano et al., 2009). A key choice in Explicit Semantic Analysis is the document space that will act as the topic space. The standard choice is to regard all articles from a back- ground document collection – Wikipedia articles are a typical choice – as the topic space. However, it is crucial to ensure that these topics cover the se- mantic space evenly and completely. In this pa- per, we present an alternative approach where we remap the semantic space defined by the topics in such a …

Content truncated to 500 words