Code-switching, dialect and intent

Learning outcome: Avoid treating language variation as an error by default.

Code-switching uses more than one language or variety within a conversation. Dialect and informal phrasing can carry meaning that a system trained or tested narrowly handles poorly. Standardising everything may remove tone, identity or useful context.

Define the task carefully. If the goal is to identify a request, preserve its meaning without grading the speaker's language. Compare matched examples with the same intent expressed differently. Ask speakers or qualified reviewers about ambiguous interpretations rather than inventing a cultural explanation. Record failures as properties of the system and test, not defects in the people whose language it mishandles.

Worked example

A fictional support message mixes English with a familiar phrase from another language and asks for a replacement link. The classifier marks it hostile because the phrase is absent from its examples. A meaning-equivalent formal-English message is labelled routine. This difference warrants investigation of the classifier and reference labels.

Try it yourself

Write two versions of a harmless request using language varieties you genuinely understand. Keep the intent constant and define the expected classification before testing.

You can do this exercise on paper. If using an assistant, keep to the fictional material and an approved tool.

Check your understanding

What should remain fixed in a matched language-variation test?

1. The underlying intent of the request
2. The exact spelling of every word
3. The assumption that formal English is always more polite

Workbook

Download the course workbook to record your answer and compare it after the knowledge check.

Further reading

• W3C: images
• W3C: complex images
• W3C: accessible media