Skip to content
AI Knowledge MapSearch / Text directory

When should I explain Chinese, English, dialect, slang or mixed-language expressions?

A016 · A Expressing intent · Aligning words and meanings

Key answer

Chinese, English and mixed language are usable, but do not assume equal familiarity with every dialect, community expression or combination. Explanation matters when a term determines the task's meaning.

Full explanation

An EMNLP 2024 study tested GPT-3.5 and GPT-4 on ten English varieties, with native-speaker evaluations, and found poorer understanding and response quality for some nonstandard varieties. Language varieties can affect service quality; the same decline cannot be assumed for a Chinese dialect. Paper

A LREC-COLING 2024 study compared six models on code-switching translation across seven datasets, where GPT-4 and GPT-3.5 showed strong capabilities. This also contradicts the blanket claim that mixed Chinese and English necessarily confuse AI. Translation performance is not performance on every reasoning task, however. Paper

There is usually no need to translate an accurate English product name, code identifier or technical term embedded in Chinese instructions. Explain ambiguous abbreviations, new slang, sarcasm, local expressions and words AI has clearly misunderstood. Retain the term and add its meaning here:

In this text, “zhongcao” means making readers interested in trying a product, not gardening; “agent” means an AI program that can call tools to perform tasks. Answer in English and retain code and product names unchanged.

If the task analyzes dialect, irony or style, the original wording is important evidence. Keep it alongside an explanation; standardizing everything can erase the feature under analysis.

References

1. Fleisig et al. (2024), Linguistic Bias in ChatGPT: Language Models Reinforce Dialect Discrimination, EMNLP

2. Huzaifah et al. (2024), Evaluating Code-Switching Translation with Large Language Models, LREC-COLING