In this article, we define the outlier detection task and use it to compare neural-basedword embeddings with transparent count-based distributional representations. Using the EnglishWikipedia as a text source to train the models, we observed that embeddings outperform count-basedrepresentations when their contexts are made up of bag-of-words. However, there are no sharpdifferences between the two models if the word contexts are defined as syntactic dependencies.In general, syntax-based models tend to perform better than those based on bag-of-words for thisspecific task. Similar experiments were carried out for Portuguese with similar results. The testdatasets we have created for the outlier detection task in English and Portuguese are freely available.
展开▼
