External Lexical Information for Multilingual Part-of-Speech Tagging
This work addresses part-of-speech tagging for multilingual NLP, showing that feature-based methods can be competitive with neural approaches under certain conditions, but it is incremental as it compares existing methods.
The paper compared four part-of-speech tagging systems (two feature-based and two neural-based) across 16 languages, finding that all achieve state-of-the-art results on average, with feature-based models performing better on lexically rich datasets and neural models on less variable ones.
Morphosyntactic lexicons and word vector representations have both proven useful for improving the accuracy of statistical part-of-speech taggers. Here we compare the performances of four systems on datasets covering 16 languages, two of these systems being feature-based (MEMMs and CRFs) and two of them being neural-based (bi-LSTMs). We show that, on average, all four approaches perform similarly and reach state-of-the-art results. Yet better performances are obtained with our feature-based models on lexically richer datasets (e.g. for morphologically rich languages), whereas neural-based results are higher on datasets with less lexical variability (e.g. for English). These conclusions hold in particular for the MEMM models relying on our system MElt, which benefited from newly designed features. This shows that, under certain conditions, feature-based approaches enriched with morphosyntactic lexicons are competitive with respect to neural methods.