language technology product teams

Abstract: Despite recent advancements in ASR, identifying effective architectures, fine-tuning approaches, and data adaptation strategies for low-resource languages remains challenging. We present a systematic comparison across five low-resource African languages and Swahili as a mid-resource contrast, fine-tuning three models from distinct architectures: Phi-4-multimodal-instruct, Whisper-large-v3-turbo, and MMS-1B-all. We analyze data sufficiency, monolingual versus multilingual adaptation, domain fine-tuning, and language-adapter transfer. Phi-4 achieves the best overall accuracy, MMS shows comparable CER, while Whisper exhibits high CER, consistent with known hallucinations and repetition loops. Performance saturates at different data scales, offering practical data-efficiency targets, while typological complexity consistently shapes error rates. Domain fine-tuning benefits Phi-4 and Whisper but degrades MMS, revealing limitations of adapter-based adaptation. Proxy language adapters outperform training MMS adapters from scratch.

Abstract: LLM-as-a-Judge evaluation is increasingly applied to multilingual tasks, yet prompting strategies such as Chain-of-Thought and Role Conditioning are recommended based solely on English benchmarks. Whether their effectiveness transfers across judge models and languages remains unknown: prompt-sensitivity studies are largely English-centric, while existing multilingual judge studies fix the prompt template. In this work, we address this gap by measuring this interaction through a benchmark of 9 prompting strategies, 4 open-weight judges, 27 languages, and 4 tasks across 13 datasets, yielding over 1M pointwise scores on gold-reference outputs, validated against human judgments. We find that prompting strategies are not transferable: the choice of the judge model dominates the choice of prompting strategy, strategy effects reverse direction across judges, and stability degrades in low-resource languages, where these effects are concealed by aggregate reporting. These findings show that prompting effects in multilingual evaluation are conditioned on the judge, task, and language. Consequently, prompting strategies should be validated for the intended evaluation setting, and cross-study comparisons should account for both judge and prompt configuration.

Abstract: Autoregressive encoder-decoder speech recognition models trained with teacher forcing can suffer from exposure bias, limiting their ability to recover from their own prediction errors at inference time. Recent work has shown that reinforcement learning with Group Relative Policy Optimisation (GRPO) improves speech recognition in major languages with LLM-based models. We extend this line of work to smaller, non-LLM encoder-decoder ASR models, using Whisper-large-v3 as a case study across 51 African languages. The benefits of this approach are most pronounced when training data is scarce or when input audio is noisy or degraded. Our resulting model, SunflowerASR, achieves the best published macro-averaged WER on the AfriVox-v2 (0.227) and SimbaBench (0.372) benchmarks, and is publicly available.

Abstract: Automatic Speech Recognition (ASR) system evaluations have consistently revealed performance disparities along lines of race, gender, socioeconomic status, and language variety. While these disparities are well-documented, they are often (implicitly or explicitly) treated as issues that subsequent, larger, and more universal models will correct. We challenge this assumption of steady progress in reducing accent-related performance disparities, conducting a longitudinal evaluation of 16 prominent systems released between 2021-2025 across 8 datasets representing diverse English accent varieties. We find that the performance disparity between Mainstream US English (MUSE) and minority accents is stubborn, not reliably declining as a function of time. Specifically, while systems are reliably performant on MUSE, minority accents are subject to risks of large performance degradations between generations of systems, sometimes worsening by 65% absolute Word Error Rate. We argue that standard ASR benchmarks enact an implicit language policy privileging Mainstream US English varieties. As such, these performance disparities reflect systemic failures of language technology development infrastructure that disproportionately affect underrepresented speaker populations.

Abstract: Tokenizers are commonly optimized for compression, but a compact vocabulary does not necessarily distribute its capacity evenly across languages. We introduce the Latent Core Tokenizer (LCT), a language-agnostic approach that separates structural discovery from vocabulary construction. LCT uses Minimum Description Length, entropy-based boundary signals, and morphotactic constraints to identify reusable linguistic units before constructing a shared vocabulary. Across 104 languages with a 200K-token vocabulary, LCT achieves lower fertility and higher MorphScore than BPE, Unigram, and parity-aware BPE, while maintaining comparable cross-lingual disparity in tokenization cost. Across four multilingual downstream benchmarks, LCT improves aggregate score by 1.48, 1.83, and 2.00 points over BPE, Unigram, and parity-aware BPE, respectively. Our findings show that compression alone does not predict representation quality and highlight the importance of morphology-driven structural discovery and how frequency is used to allocate the final vocabulary across languages.

Abstract: We understand little about how capabilities acquired in one language carry over to another, or what governs this transfer: evaluations rely on incomparable, saturation-prone datasets and rarely examine its determinants jointly. Identifying what predicts transfer would let us avoid exhaustive evaluation across all language pairs and let developers target the factors that limit performance in low-resource languages. To evaluate cross-lingual capability transfer, we introduce Multilingual GSM-Symbolic, an extensible multilingual mathematical dataset covering 30,000 item-matched question-answer pairs and spanning 15 languages. It utilises symbolic templates to prevent overfitting and ensure generalisation by allowing generation of millions of high-quality variations from a single sample. Using Multilingual GSM-Symbolic, we quantify the largest determinants of capability as model size ($β= 1.77$), language resource level ($β= 0.77$), reasoning ($β= 0.67$) and typological distance ($β= -0.25$). This joint estimation allows these determinants to be expressed in terms of one another: a 32B model evaluated in Marathi performs like a 10B model in English. Our findings have important implications for model developers, showing that model size and reasoning narrow the performance gap between low- and high-resource languages ($β= -0.27$ and $β= -0.20$, respectively), while similar levers have little or no effect on typologically distant languages. Overall, our analysis framework explains 92% of between-language variation, but only 23% of the model-by-language variation, and predicts a model's performance on an unseen language within 6.0pp (r=.96). Incorporating measurements from just 10 templates in the target language reduces this to 4.19pp, enabling reasonable estimates of performance with little or no downstream dataset.

Abstract: Computational methods have consistently struggled to determine the dominant temporal orientation of a sentence. This difficulty is especially pronounced when a sentence contains multiple embedded clauses with competing tense and aspectual information. To address this difficulty, we propose an alternative approach for identifying a sentence's past, present, or future global reference interval. Our method does not use any form of pretrained embeddings. We rather encode sentences using fixed-length inventory vectors that are comprised of part-of-speech counts, dependency relation counts, and explicit futurate pattern counts. We term this inventory vector of a sentence a "Morphosyntacton". The method does not use any padding, sequence models, or large language models. Evaluation on 1,799 syntactically complex English sentences, annotated as past, present or future, shows balanced and high accuracy multiclass classification, achieving an overall multi-class accuracy of 92%.

Abstract: Machine translation (MT) for Indian languages remains constrained by the limited availability of high-quality, Indic-centric parallel corpora and evaluation benchmarks. Existing multilingual resources are largely constructed from English-pivot content and often fail to capture the linguistic diversity, cultural complexity, and domain-specific characteristics of Indian languages. We present COILD, an Indic-centric parallel corpus comprising over 1.16 million human-translated and human-verified sentence pairs, covering 20 Indian language pairs across the Indo-Aryan, Dravidian, Tibeto-Burman, and Austro-Asiatic language families. The corpus is built entirely from original Indian language sources collected from licensed repositories spanning eight domains with direct real-world applicability. Furthermore, we introduce a domain-centric benchmark comprising 2,000 expert-verified sentences to enable consistent multilingual and cross-lingual evaluation across Indian language pairs. To validate the effectiveness of COILD, we fine-tune two representative multilingual neural machine translation models, IndicTrans2-Distilled and NLLB-200. Experimental results demonstrate consistent improvements across language pairs, domains, automatic evaluation metrics, and human evaluation, highlighting the effectiveness of high-quality Indic-centric supervision. COILD provides a valuable training and evaluation resource for advancing multilingual machine translation and future multilingual language models for Indian languages.

Abstract: We introduce QuranicMMLU, a benchmark for evaluating generative AI on Quranic Arabic across multiple dimensions of linguistic complexity. Existing Quranic benchmarks center on general question answering and semantic retrieval, without probing specific linguistic competencies or stratifying by cognitive demand and verse difficulty. We construct a five-pillar Quranic taxonomy spanning Phonology, Morphology, Syntax, Semantics, and Pragmatics, with 31 leaves covering phenomena from tajwīd and root-and-pattern morphology to occasions of revelation and inter-surah coherence. For each leaf we generate questions stratified by Bloom's cognitive level and verse perplexity, then have LLM as a judge to independently answer and score every item and route the annotations to manual review. The resulting dataset comprises 980 human-reviewed questions, each issued in both open-ended and multiple-choice form. We benchmark 12 systems on these items and find that the Islamic-specialized model leads, yet every system scores higher on multiple-choice accuracy (average 84%) than open-ended answer quality (average 60%): the two rankings agree closely (Kendall's τ=0.73), but multiple-choice scoring hides failures that surface only once answer choices are removed. QuranicMMLU thus offers a rigorous, linguistically grounded framework for evaluating Arabic NLP in the Quranic domain.

Abstract: Learning from limited text requires models to use context, generalize to new inputs, and retain useful capabilities. Qiushi Engine conducted a long-horizon, end-to-end autonomous research program on BabyLM 2026 Strict-Small, within 10 million corpus words and 100 million cumulative word presentations. Three stages connected frontier advancement, principle discovery, and principle-guided model improvement. Stage I combined compact restatements, budget reinvestment, and residual incremental learning to build a frontier model. Stage II found that exact repetition and aligned restatement produce different patterns of context use, depending on target relations and prediction windows. In controlled tasks, recovering familiar performance did not ensure that unseen inputs could still use learned computations. These findings support a testable data-efficient learning principle: organize experience around the contextual dependencies needed for prediction; separately design visible information, supervision, and preservation; test learning, generalization, and retention. Stage III retained source text, masked more local clues, supervised selected targets, and preserved predictions on ordinarily masked inputs. Two continuation seeds from the same parent outperformed ordinary continuation on the complete nine-metric aggregate. Overall rose from 42.02 to 42.25 across two generations; the second achieved the highest Overall in the public Strict-Small snapshot of 8 September 2026. Further studies addressed compression, relational anchors, shared representations, and measurement. Models are available on Hugging Face; code and research records accompany the GitHub repository. Together, these stages illustrate Research RSI: recursive self-improvement of the research process. Scientific understanding and method innovations change subsequent questions and designs; new experiments test and refine them.