ner-capitalization-unreliable-signal

IN premise — entries/2026/06/21/wiki-Natural_language_processing-chunk-2.md

Created 2026-06-21T09:50:10+00:00

Capitalization is an unreliable signal for NER: German capitalizes all nouns, Chinese and Arabic have no capitalization, and sentence-initial words are always capitalized

Summary

You can't use capitalization as a shortcut to spot names and entities in text, because German writes every noun with a capital letter, Chinese and Arabic don't use capitalization at all, and the first word of any sentence is capitalized whether or not it names something. Any entity-detection system that treats "starts with a capital" as a useful cue will break or misfire across multiple languages.

Dependents

These beliefs depend on this one:

IN cross-linguistic-assumptions-create-generalization-barrier — NLP techniques that appear universally applicable often embed language-specific assumptions (space delimitation for tokenization, capitalization for named entity recognition) that fail across languages, revealing a persistent cross-linguistic generalization barrier beneath apparent task simplicity.