Upload folder using huggingface_hub

README.md

@@ -15,9 +15,11 @@ size_categories:
 
 # Pragmatic Blind Spots Under Roman Urdu Framing
 
-## Research question
+## Question 1: the blind spot
 
-Does a small frontier language model lose pragmatic and social meaning when a situation is framed in Roman Urdu rather than English, and does supplying the missing social context explicitly recover it?
+The central blind spot in multilingual evaluation is that models frequently lose pragmatic and social meaning under Roman Urdu framing, even when they understand the individual words. I write in Roman Urdu myself, and I regularly see language models misunderstand the intended meaning in everyday exchanges. While standard multilingual benchmarks evaluate formal Perso-Arabic Urdu script or sanitized translations, millions of speakers use Latin-script Roman Urdu for daily communication. In this informal setting, models struggle with subtle social cues, such as softening a disagreement with an elder or recognizing when a polite offer is not meant to be taken literally. Natural language processing research has largely overlooked these dynamics by treating code-switching and non-standard spelling as machine translation problems rather than challenges in social reasoning.
+
+An early diagnostic evaluation I ran on MiniCPM5-2B across 12 hand-constructed Pakistani scenarios (48 prompts) highlights this gap. In that test, framing situations in Roman Urdu rather than English dropped the model's score by an average of roughly two points on an eight-point rubric. Error analysis pointed to different kinds of breakdowns. In scenarios like respect_register and indirect_no, which fell from 8/8 in English to 1/8 under Roman Urdu framing, the difficulty was pragmatic: the model processed the vocabulary but failed to grasp the social context. In other cases, such as the *khala* failure where the model confused a maternal aunt with a Middle Eastern dish, a basic vocabulary gap caused the breakdown. When follow-up matched reruns showed that an earlier moderator effect I had observed between isolated code-switches and full narrative framing was actually an artifact of uneven English baselines, I publicly withdrew the claim. On the four cleanly rewritten scenarios, the language gap averaged 2.75 points across two coders. Given the small sample and wide uncertainty interval, this result is an early diagnostic signal rather than a definitive baseline, but it illustrates how models can produce fluent text while still misinterpreting everyday social dynamics in informal registers.
 
 ## Design
 
@@ -144,15 +146,13 @@ The second coder scored more leniently and every effect is smaller, but the dire
 
 The largest drops under Roman Urdu framing are in unsupported assumptions, from 1.33 to 0.42, and semantic preservation, from 0.92 to 0.33. Intent inference holds up comparatively well. The model often identifies roughly what the speaker wants and then invents surrounding detail or loses the practical implication.
 
-## Path forward
-
-A contrastive pragmatic alignment corpus built around communicative acts rather than isolated translations. Each item would contain the same act in explicit English, indirect English, Roman Urdu framing, and code-switched form, annotated for literal content, intended speech act, relationship constraints, uncertainty, and acceptable responses.
+## Question 3: path forward
 
-Supervised fine-tuning on those contrasts would teach the mapping from surface wording to pragmatic representation. A preference stage would then reward preserving intended meaning and calibrating uncertainty, with hard negatives that are fluent and literally correct while missing the refusal, request, or obligation the context encodes.
+Addressing these failures requires evaluating and training models on communicative intent rather than surface word overlap. My proposed approach starts with a contrastive pragmatic alignment corpus built around specific speech acts. Each scenario will contain four parallel representations: explicit English, indirect English, Roman Urdu framing, and natural code-switched text. For every item, structured pragmatic annotations will define the literal content, the speaker's illocutionary intent, interpersonal constraints such as age or kinship hierarchy, calibrated uncertainty, and acceptable responses.
 
-The kinship failure suggests a cheaper first step. A vocabulary and kinship-term audit would establish how much of the gap is lexical before any alignment work begins, since a lexical gap and a pragmatic gap call for different interventions.
+Training with this resource will follow two stages. First, supervised fine-tuning across the four parallel conditions will teach models to connect different surface realizations to the same underlying pragmatic meaning. Second, preference optimization will reward answers that preserve intent and reflect the appropriate social context. The preference dataset will include hard negatives: responses that are grammatically fluent and literally correct, but nevertheless miss the intended politeness or indirect refusal.
 
-Evaluation after training should use held-out expressions, relationships, and scenarios, so that gains cannot come from memorising the specific phrasings used in training.
+Before large-scale alignment, a practical and low-cost first step is a targeted vocabulary and kinship-term audit. As the *khala* example showed, a missing word can look like a failure in social reasoning, and each problem requires a different solution. Finally, evaluation must test generalization on held-out expressions and social scenarios to ensure models are not simply memorizing specific phrasing. This diagnostic approach helps determine where multilingual understanding breaks down across scripts and everyday registers, providing a clearer view of model performance.
 
 ## Limitations