A Review of Knowledge Representation and Learning in Part-of-Speech Tagging

Part-of-speech tagging involves assigning the correct tag from a specified tagset to each item in a text corpus. Approaches to part-of-speech tagging can be grouped according to the type of knowledge representation they employ and whether those representations are hand-crafted or generated using machine learning. Numerical representations require machine learning as they cannot be easily understood by humans. Rule-based approaches encode knowledge that is human-readable, allowing rules to be machine-learned or manually encoded. The task of part-of-speech tagging usually involves some form of machine learning, but methods that involve no machine learning claim the best results. The trade-off for these good results are a lack of portability and rigour, and I conclude that some form of machine learning is vital in a part-of-speech tagger.