India

Large Language Models (LLMs) have advanced the automated generation of code from natural language prompts. However, low-resource languages (LRLs) like Bangla remain underrepresented due to the limited availability of instruction-to-code datasets and evaluation benchmarks. To address this, the BLP Workshop at IJCNLP-AACL 2025 introduced a shared task on &quot;Code Generation in Bangla&quot;. In this work, we propose a method that combines instruction prompting with a test-driven, feedback-guided iterative refinement process using a fine-tuned Qwen2.5-14B model. The model generates code from Bangla instructions, tests it against unit tests, and iteratively refines any failing outputs through three evaluation passes, using test feedback to guide each step. This approach helped our team &quot;Retriv&quot; to secure 2nd place in the shared task with a Pass@1 score of 0.934. The analysis highlights challenges in Bangla instruction understanding and Python code generation, emphasizing the need for targeted methods in LRLs. We made experimental scripts publicly available for the community.

IJCNLP-AACL 2025

TeamB2B at BLP-2025 Task 2: BanglaForge: LLM Collaboration with Self-Refinement for Bangla Code Generation

llm collaboration

bangla

code generation

machine translation

### Welcome to IJCNLP-AACL 2025! 
 It is a great honor to host this joint conference in Mumbai, India, from December 20 to 24, 2025. The joint conferences of IJCNLP and AACL are organized with alternating leadership in the Asia-Pacific region. The event is run by the Asian Federation of Natural Language Processing (AFNLP) in odd years, and by AACL in even years, while it is organized solely by ACL when the annual ACL meeting is held in the region. This year, the conference is primarily organized by AFNLP. 
*Kentaro Inui
MBZUAI, UAE
General Chair, IJCNLP-AACL 2025* 
Read full message and download the Conference Handbook [**here**](https://drive.google.com/file/d/1UTwxkAqSqI-GAoJC3wE1zZt5VP1Y8GX0/view?usp=sharing).

The 14th IJCNLP & 4th AACL will be held in Mumbai, India from December 20th to December 24th, 2025.

Large Language Models (LLMs) have advanced the automated generation of code from natural language prompts. However, low-resource languages (LRLs) like Bangla remain underrepresented due to the limited availability of instruction-to-code datasets and evaluation benchmarks. To address this, the BLP Workshop at IJCNLP-AACL 2025 introduced a shared task on "Code Generation in Bangla". In this work, we propose a method that combines instruction prompting with a test-driven, feedback-guided iterative refinement process using a fine-tuned Qwen2.5-14B model. The model generates code from Bangla instructions, tests it against unit tests, and iteratively refines any failing outputs through three evaluation passes, using test feedback to guide each step. This approach helped our team "Retriv" to secure 2nd place in the shared task with a Pass@1 score of 0.934. The analysis highlights challenges in Bangla instruction understanding and Python code generation, emphasizing the need for targeted methods in LRLs. We made experimental scripts publicly available for the community.

workshop paper

Sign language is a vital means of communication for the deaf and hard-of-hearing community. However, translating Indian Sign Language (ISL) into regional languages like Odia remains a significant technological challenge due to the language’s rich morphology, agglutinative grammar, and complex script. This work presents a real-time ISL recognition and translation system that converts hand gestures into Odia text, enhancing accessibility and promoting inclusive communication. The system leverages MediaPipe for real-time key-point detection and uses a custom-built dataset of 1,200 samples across 12 ISL gesture classes, captured under diverse Indian backgrounds and lighting conditions to ensure robustness. Both 2D and 3D Convolutional Neural Networks (CNNs) were explored, with the 2D CNN achieving superior performance 98.33% test accuracy compared to the 3D CNN’s 78.33%. 
Recognized gestures are translated into Odia using a curated gesture-to-text mapping dictionary, seamlessly integrated into a lightweight Tkinter-based GUI. Unlike other resource-heavy systems, this model is optimized for deployment on low-resource devices, making it suitable for rural and educational contexts. Beyond translation, the system can function as an assistive learning tool for students and educators of ISL. This work demonstrates that combining culturally curated datasets with efficient AI models can bridge communication gaps and create regionally adapted, accessible technology for the deaf and mute community in India.

Indian Sign Language Recognition and Translation into Odia

Glossing is the task of translating from a written language into a sequence of glosses, i.e., textual representations of signs from some sign language. While glossing is in principle 'just' a machine translation (MT) task, sign languages still lack the large parallel corpora that exist for many written language pairs and underlie the development of dedicated MT systems. 
In this work, we demonstrate that glossing can be significantly improved through data augmentation. We fine-tune a Spanish transformer model both on a small dedicated corpus 3,000 Spanish–Mexican Sign Language (MSL) gloss sentence pairs, and on a corpus augmented with an English–American Sign Language (ASL) gloss corpus. We obtain the best results when we oversample from the ASL corpus by a factor of ~4, achieving a BLEU increase from 62 to 85 and a TER reduction from 44 to 20. This demonstrates the usefulness of combining corpora in low-resource glossing situations.

Low-Resource Sign Language Glossing Profits From Data Augmentation

Bangla is a low-resource language for code generation, lacking large-scale annotated datasets and tools to transform natural language specifications into executable programs. This makes Bangla-to-code generation a challenging task requiring innovative solutions. To address this, we introduce BanglaForge, a novel framework for generating code from Bangla function descriptions. BanglaForge leverages a retrieval-augmented dual-model collaboration paradigm with self-refinement, combining in-context learning, llm-based translation, systematic prompt engineering, and iterative self-refinement based on execution feedback, where a coder generates initial solutions and a reviewer enhances them for robustness. On the BLP-2025 Bangla Code Generation benchmark, BanglaForge achieves a competitive Pass@1 accuracy of 84.00%, demonstrating the effectiveness of retrieval, model collaboration, and self-refinement for low-resource Bangla code generation.

Musafir at BLP_2025 Task 2: Generating Python Code from Bangla Prompts using a Multi Model Cascade and Unit Test Validation

This paper presents our approach for the BLP25 Task 2: Code Generation in Bangla. To address the scarcity of Bangla–code training data, we adopt a two-stage pipeline. First, Bangla problem statements are translated into English using a neural translation model optimized for preserving technical semantics. Then, the translated text is passed to a Qwen-based code generation model to produce executable solutions. This translation–generation strategy leverages the strengths of English-centric code models while ensuring fidelity to the original Bangla instructions. Our system achieved competitive performance on the leaderboard, achieving the 3rd place with score of 91.8% while demonstrating that translation-augmented pipelines are effective for low-resource code generation tasks.

PyBangla at BLP-2025 Task 2: Enhancing Bangla-to-Python Code Generation with Iterative Self-Correction and Multilingual Agents

Code generation from natural language instructions in Bangla is a fundamental task in programming automation, as explored in BLP-2025 Shared Task 2: Code Generation in Bangla. Current code generation models are designed primarily for high-resource languages such as English, which creates major limitations when applied to Bangla. The key challenges are limited training data and difficulty in correctly interpreting Bangla programming instructions. In this paper, to accommodate Bangla instructions, we present a chain of thought (CoT) based prompting approach with Qwen2.5-Coder-14B model. We further introduce few-shot example in the prompt template to improve the accuracy. During competition, an accuracy of 93\% is achieved in the shared test set (test\_v1.csv) and 82.6\% is achieved on the public and private test sets (hidden). After the competition is closed, we implement a debugger prompt technique which refines answers with 3 iterative fixing attempts. Applying this new technique on the public shared test set, our system outperforms by 7\% and achieves 100\% accuracy on the public test set, highlighting the effectiveness of combining CoT prompting with iterative debugging.

Overview of BLP-2025 Task 2: Code Generation in Bangla

We present a formally grounded description of a reward-selective fine-tuning (RSFT) pipeline for code generation from Bangla natural-language prompts. The implemented system mines candidate programs via temperature and nucleus sampling, executes candidates in a sandbox and retains programs that pass all unit tests, performs supervised fine-tuning (SFT) on winners using parameter-efficient Low rank adaptation (LoRA) adapters, and augments robustness through fuzzed asserts. We specify the exact objectives and estimators used, provide a Bangla-aware preprocessing recipe, prove simple properties of the sampling budget, and report an ablation showing the effect of inference sample budget $K$ on accuracy. We also include a threat model for safe execution. Our codes are available on GitHub.

Troopers at BLP-2025 Task 2: Reward-Selective Fine-Tuning based Code Generation Approach for Bangla Prompts

This paper explores various prompting strategies in the BLP-2025 Shared Task 2, utilizing a pipeline that first translates Bangla problem descriptions into English with GPT-4o,
then applies techniques like zero-shot, few-shot,
chain of thought, synthetic test case integration, and a self-repair loop. We evaluated four
LLMs (GPT-4o, Grok-3, Claude 3.7 Sonnet,
and Qwen2.5-Coder 14B). Our findings reveal
that while traditional methods like few-shot
and chain-of-thought prompting provided inconsistent gains, the integration of explicit unit
tests delivered a substantial performance boost
across all models. The most effective strategy
combined zero-shot prompting with these synthetic tests and a self-repair loop, leading GPT4o to achieve a top Pass@1 score of 72.2%.
These results represent the value of using explicit constraints and iterative feedback in code
generation, offering a solid framework that improves the model’s code generation capabilities.

AlphaBorno at BLP-2025 Task 2: Code Generation with Structured Prompts and Execution Feedback

This paper presents our submission for Task 2 of the Bangla Language Processing (BLP) Workshop, which focuses on generating Python code from Bangla programming prompts in a low-resource setting. We address this challenge by fine-tuning the gemma-2-9b instruction-tuned model using parameter-efficient fine-tuning (PEFT) with QLoRA. We propose an iterative self-improvement strategy that augments the extremely limited training data (74 examples) by reusing verified correct predictions from the development set, alongside LoRA rank experiments (8, 16, 32), observing a clear correlation between rank and accuracy, with rank 32 delivering the best results. Compared to translation-based and retrieval-augmented baselines, our approach achieves significantly higher accuracy, with a pass rate of 47% on the development set and 37% on the hidden test set. These results highlight the effectiveness of combining iterative data augmentation with rank optimisation for specialised, low-resource code generation tasks.

CodeAnubad at BLP-2025 Task 2: Efficient Bangla-to-Python Code Generation via Iterative LoRA Fine-Tuning of Gemma-2

This study proposes a hybrid framework for Bangla-to-Python code generation, emphasizing improved code accuracy through a two-phase pipeline: generation and debugging. During development, standalone models such as TigerLLM and StarCoder achieved modest accuracies of 27% and 24%, respectively, while more advanced models, Gemini-1.5-flash and Gemma, reached 60% and 64%. Integrating Gemma with the gpt-oss debugger substantially increased accuracy to 99.75%, highlighting the critical role of a dedicated debugging stage. In testing on unseen data, gpt-oss alone achieved 67%, which improved to 71% with self-debugging. The highest performance, 84%, was obtained by pairing Gemini-2.5-flash as the generator with gpt-oss for debugging. These findings demonstrate that combining a strong generative model with an effective debugging component yields superior and robust code generation results, outperforming existing approaches such as TigerLLM. The full implementation of the framework is publicly available at https://github.com/fahmid-juboraj/Code_generation.

BRACU_CL at BLP-2025 Task 2: CodeMist: A Transformer-Based Framework for Bangla Instruction-to-Code Generation

Code synthesis from natural language problem statements has recently gained popularity with the use of large language models (LLMs). Most of the available systems and benchmarks, however, are developed for English or other high-resource languages, and a gap exists for low-resource languages such as Bangla. Addressing the gap, the Bangla Language Processing (BLP) Workshop at AACL-IJCNLP 2025 featured a shared task on Bangla-to-Python code generation. Participants were asked to design systems that consume Bangla problem statements and generate executable Python programs. A benchmark data set of training, development, and test splits was provided, and evaluation utilized the Pass@1 metric through hidden test cases. We present here a system we developed, using the state-of-the-art LLMs through a zero-shot prompting setup. We report outcomes on several models, including variants of GPT-4 and Llama-4, and specify their relative strengths and weaknesses. Our best-performing system, based on GPT-4.1, achieved a Pass@1 score of 78.6% over the test dataset. We address the challenges of Bangla code generation, morphological richness, cross-lingual understanding, and functional correctness, and outline the potential for future work in multilingual program synthesis.

Downloads

Next from IJCNLP-AACL 2025

Indian Sign Language Recognition and Translation into Odia

Similar lecture

Musafir at BLP_2025 Task 2: Generating Python Code from Bangla Prompts using a Multi Model Cascade and Unit Test Validation