← Lessons

quiz vs the machine

Silver1050

Machine Learning

The Part Of Speech Tagging Deep

Assigning grammatical categories like noun and verb to every word.

4 min read · intro · beat Silver to climb

What POS tagging does

Part of speech tagging labels each word with its grammatical category: noun, verb, adjective, determiner, and so on. It is a building block for parsing, NER, and many downstream tasks.

Why context matters

The same surface word takes different tags in different sentences.

  • In I book a flight, book is a verb.
  • In I read a book, book is a noun.

So a tagger cannot just look up a word in a dictionary; it must read the surrounding words.

Classic and modern approaches

  • Hidden Markov models treat tags as hidden states and words as emissions, decoded with the Viterbi algorithm.
  • Neural taggers encode the sentence and predict a tag per token, capturing long context.
  • Standard tag sets include the Penn Treebank tags and the cross language Universal POS tags.

How it is evaluated

Accuracy per token is the usual metric, and strong models exceed ninety seven percent on clean English news text. Errors cluster on rare words and ambiguous function words.

Key idea

POS tagging assigns a grammatical category to each word using sentence context, since one word form can be a noun in one sentence and a verb in another.

Check yourself

Answer to earn rating on the learn ladder.

1. Why can POS tagging not be a simple dictionary lookup?

2. Which algorithm decodes the best tag sequence in a hidden Markov tagger?