emnlp-writing-style
Installation
SKILL.md
EMNLP Writing Style
Use this during revision. EMNLP's house voice is that of a careful experimentalist: the paper reports what was measured, in prose that never outruns the tables. The most common style failure at this venue is not bad English — it is claim inflation, where the abstract promises "understanding" and the evidence delivers accuracy on four datasets.
The scoping discipline
Every empirical claim carries an implicit quantifier, and EMNLP reviewers expand it:
| Draft sentence | What a reviewer hears | Scoped rewrite |
|---|---|---|
| "LLMs cannot track negation" | All LLMs, all negation, all languages | "The three 7-13B models we test fail on double negation in English and German (Table 4)" |
| "Our method improves summarization" | Everywhere, always | "…improves faithfulness on news summarization in two of three datasets; gains do not transfer to dialogue (§6.3)" |
| "This is a multilingual benchmark" | Typologically broad | "…covering five Indo-European languages; we discuss this limit in §Limitations" |
| "significantly better" | A statistical test exists | "+1.8 F1, paired bootstrap p<0.01 over 5 seeds" — or delete "significantly" |