Neural disambiguation of lemma and part of speech in morphologically rich languages
This addresses a key challenge in natural language processing for morphologically rich languages, offering a novel unsupervised approach that matches or exceeds supervised methods.
The paper tackles the problem of disambiguating lemma and part of speech for ambiguous words in morphologically rich languages, achieving state-of-the-art performance without using manually annotated data.
We consider the problem of disambiguating the lemma and part of speech of ambiguous words in morphologically rich languages. We propose a method for disambiguating ambiguous words in context, using a large un-annotated corpus of text, and a morphological analyser -- with no manual disambiguation or data annotation. We assume that the morphological analyser produces multiple analyses for ambiguous words. The idea is to train recurrent neural networks on the output that the morphological analyser produces for unambiguous words. We present performance on POS and lemma disambiguation that reaches or surpasses the state of the art -- including supervised models -- using no manually annotated data. We evaluate the method on several morphologically rich languages.