Bangla-English tutoring requires more than producing a correct translation: learners also need explanations of grammar differences, awareness of their likely errors, and targeted practice. We present TRACE-BN, a curriculum-guided dataset of structured tutoring traces for Bangla-speaking learners of English at the CEFR A1-A2 level. Each trace combines word-level glosses, literal and natural translations, Bangla grammar explanations, a plausible learner error, and a targeted practice question with its answer. The traces are generated by Gemini 3.5 Flash Lite as the teacher model from NCTB Classes 9-10 English curriculum units, then filtered for structural validity, script integrity, and semantic duplication. We transfer the resulting structured tutoring behavior to Qwen3-0.6B using LoRA with 4-bit quantization for resource-constrained offline deployment. On held-out inputs, schema validity increases from 85.4% to 95.8%, while, against teacher-model references, chrF++ improves from 15.28 to 34.77 and BLEU from 4.52 to 21.03. Field-level evaluation by two independent judges shows improvements across translation, grammar explanation, learner-error diagnosis, and practice alignment, while a human audit supports the quality of the supervision data. The results show that curriculum-guided structured supervision can transfer multi-component tutoring behavior to a sub-1B model under these resource constraints. The dataset, model checkpoints, and code are publicly available at https://huggingface.co/datasets/RaiyanKhaan/Trace-BN
Figures & tables
Fig. 1: TRACE-BN construction and adaptation pipeline. Curriculum-guided Bangla inputs are used to obtain teacher-generated tutoring traces, which are filtered into TRACE-BN and used to adapt Qwen3-0.6B with LoRA.
Fig. 2: Example TRACE-BN tutoring trace. A Bangla learner input is mapped to word-level glosses, literal and natural English, a Bangla contrastive explanation, a common learner error, and a short practice exercise with its answer.
Fig. 4: Qualitative comparison of base vs. TRACE-BN-tuned Qwen3-0.6B outputs on representative held-out Bangla inputs. The base model produces literal word substitutions or degenerative loops, whereas the tuned model outputs fluent target translations alongside contrastive Bangla grammar explanations.
Dataset Attribute
Specification / Count
Corpus Scale & Partitions
Total structured tutoring traces
4,099
Training split ( Dtrain )
3,667 (89.5%)
Held-out evaluation split ( Deval )
432 (10.5%)
Curriculum & Pedagogical Scope
Target learner level
CEFR A1–A2 (Beginner)
TABLE I: Key specifications and composition of the TRACE-BN dataset.
Fig. 5: Training loss progression of Qwen3-0.6B on TRACE-BN across three epochs (621 steps). Loss decreases steadily and stabilizes near training completion.
Fig. 6: Empirical evaluation of TRACE-BN adaptation: (a) Benchmark model comparison against zero-shot baselines on schema validity (%), chrF++, and BLEU; (b) Field-level pedagogical ratings on a 0–4 anchored scale evaluated independently by dual automated judges (GPT-4o-mini and DeepSeek-V4-Flash-0731); (c) Trace component ablation demonstrating the impact of multi-task supervision on translation and schema adherence.
Model
Params
Schema
chrF++ ↑
BLEU ↑
Tok/s
Gemma-4 E2B
2.3B
90.5%
50.91
33.54
45.97
TigerLLM-1B
1.0B
67.1%
22.23
8.21
48.79
Llama-3.2-1B-Instruct
1.0B
12.7%
13.21
1.95
73.11
Qwen3-0.6B (base)
0.6B
85.4%
15.28
4.52
90.89
Qwen3-0.6B (ours)
0.6B
95.8%
34.77
21.03
75.47
TABLE II: Performance of zero-shot baselines and the TRACE-BN-tuned Qwen3-0.6B on the shared held-out set ( n=432 ). Schema validity requires valid JSON with all seven fields present, including a parseable English natural_translation ; translation metrics use the teacher-model output as a silver reference; Tok/s is inference throughput on the evaluation set.
Dimension
Base
Ours
κ
Word gloss accuracy
0.10 (0.0%)
1.48 (16.4%)
0.788
Literal translation
0.21 (0.0%)
1.33 (14.8%)
0.754
Natural translation
0.26 (1.9%)
1.31 (16.2%)
0.803
Grammar explanation
0.27 (0.0%)
1.48 (9.7%)
0.489
Learner-error diagnosis
0.23 (0.0%)
1.47 (11.3%)
0.585
Practice alignment
0.16 (0.2%)
1.17 (9.7%)
0.529
TABLE III: Field-level pedagogical evaluation of the base and TRACE-BN-tuned Qwen3-0.6B on the shared held-out set ( n=432 ). Scores are means on a 0–4 scale; the percentage in parentheses denotes outputs receiving a score of at least 3. Inter-judge agreement is reported using quadratic weighted Cohen’s κ .
Training Configuration
Target Fields
Schema
chrF++ ↑
Translation Only
1 (String)
N/A
31.84
Translation + Explanation
2 (JSON)
91.2%
33.12
Full TRACE-BN (Ours)
7 (JSON)
95.8%
34.77
TABLE IV: Ablation of tutoring trace components on Qwen3-0.6B ( n=432 ).
Corrective feedback is among the best-evidenced drivers of second-language acquisition, yet corrections delivered during lessons rarely accumulate into an actionable view of grammar mastery. Prompted frontier models can provide such a view from learner--tutor lesson transcripts, but they are costly at scale. We close this gap by fine-tuning Qwen3.5 small language models (SLMs) on filtered and rebalanced teacher-generated supervision, then deploying an efficient 0.8B model in an end-to-end grammar mastery tracker for all English learners on our platform. Internalizing the annotation contract into adapter weights enables pairing the 0.8B model with a compact matched prompt rather than verbose instructions. On two human-curated benchmarks, both the deployed 0.8B model and a 4B reference comparator outperform prompted GPT-5.4 and GPT-5.6 Sol in precision and recall under nested matching criteria of increasing strictness: concept, evidence span, and correctness. The deployed 0.8B SLM reduces serving cost by approximately 16×. A feature-level online experiment shows significant gains in learner engagement (+15.8%) and key business metrics, including scheduled hours (+2.1%) and GMV from new lessons (+13.2%).
Marjan Celikik, Ana Peleteiro Ramallo, Javier Morales
Recent advances in Multilingual Large Language Models (MLLMs) have significantly enhanced cross-lingual conversational capabilities, yet modeling culturally nuanced and context-dependent communication remains a critical bottleneck. Specifically, existing state-of-the-art models exhibit a severe pragmatic gap when handling structural variations, regional idioms, and honorific consistencies in low-resource contexts like Bangla. To address this limitation, we introduce a novel, culturally aligned instruction-tuning dataset for BangLa Application and DialoguE generation - BLADE and benchmarking framework comprising 4,196 meticulously curated interaction pairs. We leverage this resource to systematically fine-tune and evaluate leading open-weight architectures, including DeepSeek-8B and LLaMA-3.2-3B, utilizing parameter-efficient fine-tuning via LoRA adapters in a 4-bit NormalFloat (NF4) quantization framework. Our empirical evaluations demonstrate that models fine-tuned on our dataset yield substantial improvements in structural fidelity and honorific alignment, providing a rigorous benchmark for bridging pragmatic disparities in low-resource multilingual text generation. Code and dataset: https://github.com/ashuvo25/Bangla_Application_LLM/tree/main
Task-oriented dialogue understanding remains underexplored for Bangla, particularly in multilingual settings involving English and romanized Bangla (Banglish). We introduce BanglaTOD, a parallel tri-lingual benchmark comprising 209 multi-turn dialogues across nine Bangladesh-specific service domains and 64 intents. Each conversational turn is aligned across Bangla, English, and Banglish, yielding 4,689 language-specific turn records annotated with intents, token-level slots, state updates, and accumulated dialogue states. We establish context-aware baselines for joint intent classification and slot filling using mBERT, BERT-base-uncased, and BanglaBERT. Cross-language analysis reveals substantial performance differences across models and linguistic representations, while removing dialogue history reduces intent accuracy by 19.0--25.7 percentage points. BanglaTOD provides a foundation for multilingual dialogue understanding and future dialogue state tracking research in Bangladesh-specific conversational applications. The dataset is publicly available at Hugging Face.
Afrina Alam Lamia, Sumiaya Sharif, Prottoy Modak +2