- POS tagging = identifying the part of speech of each word in a sentence
- Used to be important for a wide range of routine NLP tasks
- Named entity recognition (NER) is easier once you know what the nouns are
- Parsing a sentence for, eg, a knowledge-based translator
- Today it’s more niche, because LLMs can provide zero-shot inferences for most of these tasks
- Still useful, eg, when training those LLMs
- Extensive research applications
- Modern POS tagging is usually accomplished using an LLM (typically BERT) that has been fine-tuned for the purpose
- Commonly accessed via spaCy
- Older approaches:
- Earliest approaches involved manual tagging
- Hidden Markov models (HMMs) became prevalent in the beginning of the computational linguistics era (1980s)
- RNNs largely supplanted HMMs because they essentially provide a much more feature rich version of the same approach
- Then, as noted, came transformers