Part-of-Speech (POS) Tagging

Part-of-speech (POS) tagging is the NLP task of labelling every word in a text with its grammatical category: noun, verb, adjective, adverb, pronoun and so on. It is one of the basic steps of syntax analysis, and it helps a machine work out what role each word plays, for example whether “book” is a thing or an action.

How it works

Text is first split into tokens (tokenization). A tagger then assigns each token a tag, using the word itself and its context. Early taggers were rule-based or statistical; modern ones are neural models trained on large annotated corpora. Many tools use the Universal Dependencies tag set (NOUN, VERB, ADJ, ADV, PROPN, DET, ADP and others).

Tools that do it:

• Google Cloud Natural Language API‘s syntax analysis returns, for each token, a detailed part-of-speech object (including features such as tense, number and case), a dependency label and the lemma.
• spaCy and NLTK in Python.

POS tags usually feed other steps: lemmatisation uses them to find the right base form, dependency parsing uses them to build sentence structure, and keyword extraction uses them to keep noun phrases.

How it’s used in SEO and marketing

• Keyword and phrase extraction: keep nouns and adjective-noun patterns (“waterproof hiking boots”) and drop filler words.
• Query analysis: separate the main noun (the thing searched for) from modifiers and verbs that hint at intent (“buy”, “fix”, “compare”).
• Content analysis: compare the share of nouns, verbs and adjectives across pages to spot style differences, as in MLforSEO’s syntax analysis Looker Studio visualisations.
• Product and review mining: pull out adjectives used with product features to understand how customers describe them.
• Cleaning transcripts and text before topic modelling.

Example

Run the sentence “Book a table at our new restaurant” through a syntax analysis template and the API would typically return “Book” as a VERB with imperative mood, “table” as a NOUN, “new” as an ADJ and “restaurant” as a NOUN. The verb in first position signals an action, which is useful when classifying queries and calls to action.

Limitations and things to watch

• Short, ungrammatical text (search queries, social posts) is harder to tag than full sentences.
• Ambiguous words can be mis-tagged, and errors carry into later steps.
• Tag sets differ between tools, so map them before comparing outputs.
• POS tags describe grammar, not meaning; combine them with entity and semantic analysis.

Learn it on MLforSEO

• How to do Syntax Analysis with Google’s Natural Language API in Google Sheets (Apps Script)
• Google Sheets Syntax Analysis Template
• Sentiment Analysis: A Deep Dive Into the Theory, Methods, and Applications
• Introduction to Machine Learning for SEO