Section: New Results

Unsupervized lexical semantics

Participant : Marianna Apidianaki.

Unsupervized word sense induction and disambiguation

Word sense induction (WSI) is the task aimed at automatically identifying the senses of words in texts, without the need for handcrafted resources or annotated data. Up till now, most WSI algorithms extract the different senses of a word 'locally' on a per-word basis, i.e. the different senses for each word are determined separately. In collaboration with Tim van de Cruys, at Alpage in 2010, now at University of Cambridge [19] , [50] , we have compared the performance of such algorithms to a new algorithm that uses a 'global' approach, i.e. the different senses of a particular word are determined by comparing them to, and demarcating them from, the senses of other words in a full-blown word space model. The induction step and the disambiguation step are based on the same principle: words and contexts are mapped to a limited number of topical dimensions in a latent semantic word space. The intuition is that a particular sense is associated with a particular topic, so that different senses can be discriminated through their association with particular topical dimensions; in a similar vein, a particular instance of a word can be disambiguated by determining its most important topical dimensions. We evaluated our model on the SemEval-2010 word sense induction and disambiguation task. All systems that participated in this task use a local scheme for determining the different senses of a word. We obtain state-of-the-art results.

Unsupervized cross-lingual lexical substitution

Cross-Lingual Lexical Substitution (CLLS) is the task that aims at providing for a target word in context several alternative substitute words in another language. The proposed sets of translations may come from external resources or be extracted from textual data. In 2011, we have introduced a new approach for this task [18] , namely the use of an unsupervised cross-lingual word-sense induction method. This method identifies the senses of words by clustering their translations according to their semantic similarity. We evaluated the impact of using clustering information for CLLS on the SemEval-2010 CLLS data set. Our system performs better on the 'out-of-ten' measure than the systems that participated in the SemEval task.