I am a NLP Master's student of the Erasmus-Mundus Dual Degree Research Program, affiliated with both University of Groningen and Saarland University. Currently, I am empirically validating Simon Ostermann's positional paper for my Master's Thesis (supervised by him and Prof. Arianna Bisazza), as well as safety evaluating LLMs as a Student Research Assistant at DFKI (High-level Project Report here). Separately, I showed transformer models' under-trained behaviour on noisy subjective tasks during my Bachelor's Thesis (appeared @ Findings of EMNLP and AACL Main Conference).
Generally, exploring LLMs' limitations excites me. Across my Master's study, I showcased Character Diffusion to improve on OOV-prevalent tasks [1], showed the long-sequence deficiency of language models also applies to SOTA Protein LMs [2], and found VLMs' gains on complex multimodal persuasion tasks are uneven across sub-tasks [3]. During my collaboration with @GroNLP, I benchmarked cross-lingual multi-step reasoning in LLMs [4]. Moreover, I contributed to OCR on ancient documents [5] and found context enrichment can dilute LLM reasoning [6]. I was also a Software Engineer at Reve Systems for around 3 years.
University of Groningen, Saarland UniversitySeptember 2024 - now
Masters in NLP
Shahjalal University of Science and TechnologyJan. 2017 - Jan. 2021
B.Sc. (Engg.) in Computer Science & Engineering (advisor: Md Saiful Islam)
Reveal-Bangla: A Dataset for Cross-Lingual Multi-Step Reasoning Evaluation
Khondoker Ittehadul Islam, Gabriele Sarti
BLP Workshop @ IJCNLP-AACL 2025
EmoNoBa: A Dataset for Analyzing Fine-Grained Emotions on Noisy Bangla Texts
Khondoker Ittehadul Islam*, Tanvir Hossain Yuvraz*, Md Saiful Islam, Enamul Hassan
AACL-IJCNLP 2022
SentNoB: A dataset for analysing sentiment on noisy Bangla texts
Khondoker Ittehadul Islam, Sudipta Kar, Md Saiful Islam, Mohammad Ruhul Amin
Findings of EMNLP 2021
Sentiment analysis in Bengali via transfer learning using multi-lingual BERT
Khondoker Ittehadul Islam, Md Saiful Islam, Mohammad Ruhul Amin
ICCIT 2020
Evaluating VLMs on Multimodal Aristotelian Persuasion Tasks
Khondoker Ittehadul Islam
arxiv 2026
Reveal-Bangla: A Dataset for Cross-Lingual Multi-Step Reasoning Evaluation
Khondoker Ittehadul Islam, Gabriele Sarti
BLP Workshop @ IJCNLP-AACL 2025
Improving OCR for Historical Texts of Multiple Languages
Hylke Westerdijk, Ben Blankenborg, Khondoker Ittehadul Islam
arxiv 2025
Joint Effects of Argumentation Theory, Audio Modality and Data Enrichment on LLM-Based Fallacy Classification
Hongxu Zhou, Hylke Westerdijk, Khondoker Ittehadul Islam
arxiv 2025
What Really Counts? Examining Step and Token Level Attribution in Multilingual CoT Reasoning
Jeremias Ferrao, Ezgi Basar, Khondoker Ittehadul Islam, Mahrokh Hassani
arxiv 2025
Leveraging sentiment for offensive text classification
Khondoker Ittehadul Islam
arxiv 2024
EmoNoBa: A Dataset for Analyzing Fine-Grained Emotions on Noisy Bangla Texts
Khondoker Ittehadul Islam*, Tanvir Hossain Yuvraz*, Md Saiful Islam, Enamul Hassan
AACL-IJCNLP 2022
SentNoB: A dataset for analysing sentiment on noisy Bangla texts
Khondoker Ittehadul Islam, Sudipta Kar, Md Saiful Islam, Mohammad Ruhul Amin
Findings of EMNLP 2021
Sentiment analysis in Bengali via transfer learning using multi-lingual BERT
Khondoker Ittehadul Islam, Md Saiful Islam, Mohammad Ruhul Amin
ICCIT 2020