Part of Speech Tagging

Summary

Part of Speech Tagging (or POS Tagging for short) is a task in the field of computational linguistics which looks at marking each word in a text corpus with the associated word categories known as parts of speech (such as noun, verb, or adjective), based on a word's definition and context of usage.

POS tagging can be useful as a preprocessing step in tasks like Parsing, and is also useful in tasks like Word Sense Disambiguation and Speech Synthesis.

Common Approaches

Some common approaches to POS Tagging include the following:

• Hidden Markov Models based approaches, sometimes referred to as stochastic algorithms in older literature
• Transformation-based learning - Brill Tagger
• Dynamic Programming/Viterbi-like algorithms - DeRose & Church, mentioned for historical reasons

Sources of information/evidence often times used by POS taggers:

• distribution of tags for the word isolation: P(t|w)
• "Syntagmatic information"- some POS sequences are much more common than others due to syntactic constraints of the language

Example Systems

• FastTag - open source implementation of Brill Tagger
• Stanford Log-linear Part-of-Speech Tagger
• OpenNLP Tagger - based on maximum entropy
• CRF Tagger - based on conditional random fields
• LingPipe - tool kit that contains models for POS tagging

References / Links

• Webpage with links to many different POS tagger systems, from Statistical natural language processing and corpus-based computational linguistics: An annotated list of resources - [1]
• Wikipedia article on Part of Speech Tagging - [2]