← Subtext · evaluation · film-sampled group · 102 lines · run 2026-09-09

The 102 lines we chose ourselves

The literature chose the film; we chose the line, by one loose criterion: that it carry profanity. That makes this the weakest of the four groups and it is labelled as such — there is no scholar behind each line, only behind each film. What every line does have is its official subtitle and its pair_id in the corpus. None of the 102 is in the corpus the system retrieves from. Under each source line, two controls: the same Gemini with no corpus, and Google Translate, which has no Mexican Spanish to ask for.