difficulty 3/5
Named entity recognition
A label for every token, read back as spans — people, places, organisations. The output is structured, so one wrong token invalidates a whole entity and accuracy stops being the right metric.
Intuition
Most of the tasks on this site return one thing per input: a class, a number, a sentence. Named entity recognition returns one label per token, and then those labels are read back as spans — Angela Merkel is one person, not two tokens that happen to be people.
That small change of shape has consequences out of proportion to it. The output is now structured, so the labels are not independent of each other: some sequences of them are not merely unlikely but meaningless. And the unit you care about is the span, not the token, so a model can be right about 95% of tokens and wrong about a third of the entities.
Lample et al. state the underlying difficulty plainly: "in most languages and domains, there is only a very small amount of supervised training data available. On the other, there are few constraints on the kinds of words that can be names, so generalizing from this small sample of data is difficult" [1, §1]. Anything can be a name. The model has to learn that from very little.
Formal definition
How it works
Where it's used
Why & when
Worked example
Pitfalls
Check yourself
3 questions
The questions on this page come with access. They are graded on the server, so the answers stay there until you have given yours.
On the cheat sheet: NLP tasks →
Sources
- Neural Architectures for Named Entity Recognition (2016)paperhttps://ar5iv.labs.arxiv.org/html/1603.01360
- BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding (2019)paperhttps://ar5iv.labs.arxiv.org/html/1810.04805
- SentencePiece: A simple and language independent subword tokenizer and detokenizer (2018)paperhttps://ar5iv.labs.arxiv.org/html/1808.06226