Unsupervised Ranked Cross-Lingual Lexical Substitution for Low-Resource Languages

Unsupervised Ranked Cross-Lingual Lexical Substitution for Low-Resource Languages Stefan Ecker, Andrea Horbach, Stefan Thater Saarland University Department of Computational Linguistics Saarbr¨ucken, Germany {stefane, andrea, stth}@coli.uni-saarland.de Abstract We propose an unsupervised system for a variant of cross-lingual lexical substitution (CLLS) to be used in a reading scenario in computer-assisted language learning (CALL), in which single-word translations provided by a dictionary are ranked according to their appropriateness in context. In contrast to most alternative systems, ours does not rely on either parallel corpora or machine translation systems, making it suitable for low-resource languages as the language to be learned. This is achieved by a graph-based scoring mechanism which can deal with ambiguous translations of context words provided by a dictionary. Due to this decoupling from the source language, we need monolingual corpus resources only for the target language, i.e. the language of the translation candidates. We evaluate our approach for the language pair Norwegian Nynorsk–English on an exploratory manually annotated gold standard and report promising results. When running our system on the original SemEval CLLS task, we rank 6th out of 18 (including 2 baselines and our 2 system variants) in the best evaluation. Keywords: cross-lingual lexical substitution, low-resource languages, computer-assisted language learning 1. Introduction Over the last few years, online learning platforms for for- eign languages have emerged to complement and some- times even replace classroom-based courses. These plat- forms are particularly important for those languages which are not offered at local language schools. One technique suitable for self-study CALL applications is Intensive Reading, where you try to gain a deep understand- ing of a small amount of text with the help of a dictionary. Yet dictionary usage itself poses a challenge to (at least be- ginning) language learners: if a word has several transla- tions with different meanings, how does one know which is the right one in a particular context? A desirable feature of a CALL application for Intensive Reading would thus be a ranking mechanism that ranks the translations of a certain word depending on its context. In the general case, one way to address this problem is to make use of the advances in cross-lingual lexical substi- tution (Sinha et al., 2009): When given a word with its context in language A, the task is to substitute the word with one or more words from language B that are good translations in that particular context. This task is rather resource-intensive: most system variants that took part in the SemEval-2010 Cross-Lingual Lexical Substitution Task use some sort of parallel corpora or a machine translation system (McCarthy et al., 2013). For low-resource lan- guages, aligned parallel corpora or off-the-shelf machine translation programs are often not of comparable size and quality to the language pair Spanish–English used in that task, if such resources are available at all. A common choice for the parallel corpus used in the CLLS systems is the sentence-aligned Europarl corpus for the lan- guage pair English–Spanish. Releases of comparable size are available for 9 additional languages …

Content truncated to 500 words