11.4.1 Sequence Labeling Roadmap: One Label per Token

Sequence labeling predicts one label for each token. NER, word segmentation, part-of-speech tagging, and slot filling all use this idea.

See the Label Path First

The key output is not one sentence label, but aligned token-level tags such as B-PER, I-PER, and O.

Run a BIO Tag Check

tokens = ["Ada", "Lovelace", "wrote", "notes"]
tags = ["B-PER", "I-PER", "O", "O"]

for token, tag in zip(tokens, tags):
    print(token, tag)

Expected output:

Ada B-PER
Lovelace I-PER
wrote O
notes O

If tokenization changes, labels must stay aligned. Many sequence-labeling bugs are alignment bugs.

Learn in This Order

Step	Read	Practice Output
1	NER and BIO	Create token-level labels and entity spans
2	HMM/CRF history	Understand sequence constraints and label transitions
3	BiLSTM-CRF	Connect contextual features with valid label paths
4	Project practice	Evaluate precision, recall, F1, boundary errors

Pass Check

You pass this chapter when you can inspect token/tag alignment and explain one boundary error or invalid tag transition.