Machines & Meaning · Capstone brief

HUM-09 · Multilingual center brief

Study model performance across languages relevant to your program; refuse English-only conclusions.

The question

Study model performance across languages relevant to your program; refuse English-only conclusions.

Texts / materials

At least two languages tied to the center’s actual program. Parallel task (translation, comprehension questions, or summarization) with human reference where possible. Results table with language labels. Explicit refusal of English-only generalization.

Expected failure modes

Testing only English plus one token phrase in another language. Declaring a language “low resource” without a source. Inventing evaluation papers.

Done looks like

Multilingual results table; method note; center-facing recommendation; non-claim list (including what was not tested); English-only conclusions crossed out in the draft and absent in the final.

Five C's

CT: performance ≠ prestige. CR: program-relevant languages. CO: bilingual peer checks one item. CM: center brief. CZ: speakers of the non-English language(s).

Mentor role

Language teacher reviews task fairness across languages. School-supervised.

Rubric calibration

R1: named languages, one task. R2: items documented. R3: human reference if used. R4: coverage limits. R5: brief actionable. R6: refuses English-only conclusions.

Two ways this goes wrong

(a) English results with a multilingual title. (b) Fake citation for “low-resource.”

Credit lane fit

Lane A immediately. No verified credit claim.