Papers by Hassan Sajjad
Analyzing Redundancy in Pretrained Transformer Models (2020.emnlp-main)
Copied to clipboard
| Challenge: | Recent work shows that transformer-based deep NLP models are over-parameterized and do not require all the representational power lent by the rich architectural choices during inference. |
| Approach: | They define a notion of Redundancy and propose a feature-based transfer learning procedure which maintains 97% performance while using at-most 10% of the original neurons. |
| Outcome: | The proposed model maintains 97% performance while using 10% of the original neurons. |
LLMs Underperform Graph-Based Parsers on Supervised Relation Extraction for Complex Graphs (2026.acl-short)
Copied to clipboard
| Challenge: | Relation extraction is a core NLP task which involves extracting [head, relation, dependent] RDF triples from text. |
| Approach: | They evaluate four large language models against a graph-based parser on six relation extraction datasets with sentence graphs of varying sizes and complexities. |
| Outcome: | The graph-based parser outperforms the LLMs on six relation extraction datasets with sentence graphs of varying sizes and complexities. |
Compressing Large-Scale Transformer-Based Models: A Case Study on BERT (2021.tacl-1)
Copied to clipboard
Prakhar Ganesh, Yao Chen, Xin Lou, Mohammad Ali Khan, Yin Yang, Hassan Sajjad, Preslav Nakov, Deming Chen, Marianne Winslett
| Challenge: | Popular pre-trained Transformers have improved performance for various NLP tasks by sizable margins, but are too resource-hungry and computation-intensive to suit low-capacity devices or applications with strict latency requirements. |
| Approach: | They present a literature review of the compression of Transformers, focusing on the popular BERT model, which has attracted considerable research attention. |
| Outcome: | The proposed models improve Sentiment analysis, paraphrase detection, machine reading comprehension, question answering, text summarization, and other tasks by sizable margins. |
How transfer learning impacts linguistic knowledge in deep NLP models? (2021.findings-acl)
Copied to clipboard
| Challenge: | Several researchers have shown that deep NLP models learn non-trivial amount of linguistic knowledge, captured at different layers of the model. |
| Approach: | They propose to fine-tune pre-trained models towards downstream NLP tasks to capture linguistic knowledge. |
| Outcome: | The proposed model is adapted to GLUE tasks and retains linguistic information in the network while forgetting it. |
Impact of Adversarial Training on Robustness and Generalizability of Language Models (2023.findings-acl)
Copied to clipboard
| Challenge: | Adversarial training is widely acknowledged as the most effective defense against adversarial attacks, but achieving both robustness and generalization requires a trade-off. |
| Approach: | They propose to compare pre-training data augmentation and training time input perturbations with embedding space perturbations to find out whether they improve generalization. |
| Outcome: | The proposed methods improve generalization and robustness of the trained models. |
One Size Does Not Fit All: Comparing NMT Representations of Different Granularities (N19-1)
Copied to clipboard
| Challenge: | Recent work has shown that contextualized word representations are a viable alternative to simple word prediction tasks. |
| Approach: | They propose to use subword units and characters to model morphology, syntax, and semantics instead of word embeddings. |
| Outcome: | The proposed representations are better for modeling syntax and more robust to noisy input. |
Multilingual Nonce Dependency Treebanks: Understanding how Language Models Represent and Process Syntactic Structure (2024.naacl-long)
Copied to clipboard
| Challenge: | a number of studies have focused on making explicit the linguistic information encoded in language models (LMs) however, this method has been criticized for various reasons. |
| Approach: | They introduce a framework for creating nonce treebanks for multilingual UD corpora . they investigate word co-occurrence statistics and show how nonce data affects the performance of syntactic dependency probes. |
| Outcome: | The proposed framework satisfies syntactic argument structure and ensures grammaticality via language-specific rules. |
Analyzing Individual Neurons in Pre-trained Language Models (2020.emnlp-main)
Copied to clipboard
| Challenge: | Recent work shows that deep NLP models capture linguistic knowledge but little attention is paid to individual neurons. |
| Approach: | They conduct a neuron-level analysis of pre-trained neural language models to determine linguistic properties. |
| Outcome: | The proposed model is more localized and disjoint when predicting properties than BERT and others. |
SAGE: A Search-AuGmented Evaluation of Large Language Models on Free-Form QA (2026.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are prone to hallucination and rely on static, pre-annotated references for evaluation. |
| Approach: | They propose a framework to assess large language models without fixed ground-truth answers by iteratively generating web queries and synthesizing external evidence. |
| Outcome: | The proposed framework achieves substantial to perfect agreement with human evaluations on multiple free-form QA benchmarks. |
Are We Ready for this Disaster? Towards Location Mention Recognition from Crisis Tweets (2020.coling-main)
Copied to clipboard
| Challenge: | Despite the widespread use of Twitter during emergencies, the majority of tweets do not have geoinformation. |
| Approach: | They propose to use Twitter to train location mention recognition models using different training settings. |
| Outcome: | The results show that training on near or far-away events boosts the performance compared to training on distant events. |
A System for Diacritizing Four Varieties of Arabic (D19-3)
Copied to clipboard
| Challenge: | Short vowels, aka diacritics, are omitted when writing different varieties of Arabic . diacritization is essential for language learning and text-to-speech applications . |
| Approach: | They propose a system for recovering diacritics in Arabic without short vowels . they use a character-based sequence-to-sequence deep learning model . |
| Outcome: | The proposed system beats all previous SOTA systems for Arabic varieties . it uses a character-based sequence-to-sequence deep learning model . |
Incremental Decoding and Training Methods for Simultaneous Translation in Neural Machine Translation (N18-2)
Copied to clipboard
| Challenge: | a tunable agent decides the best segmentation strategy for a user-defined BLEU loss and Average Proportion (AP) constraint. |
| Approach: | They propose a tunable agent which decides the best segmentation strategy for a user-defined BLEU loss and average proportion (AP) constraint. |
| Outcome: | The proposed agent outperforms existing Wait-if-diff and Wait-If-worse agents on BLEU with a lower latency. |
Effect of Post-processing on Contextualized Word Representations (2022.coling-1)
Copied to clipboard
| Challenge: | Post-processing of static embeddings has been shown to improve their performance on both lexical and sequence-level tasks. |
| Approach: | They standardize individual neuron activations using z-score, min-max normalization, and remove top principal components using the all-but-the-top method. |
| Outcome: | The proposed method unwraps vital information present in the representations for both lexical and sequence classification tasks. |
UrBLiMP: A Benchmark for Evaluating the Linguistic Competence of Large Language Models in Urdu (2026.findings-acl)
Copied to clipboard
| Challenge: | Evaluating how large language models capture grammatical structure of low-resource languages remains underexplored. |
| Approach: | They evaluate a set of 5,696 minimal pairs that contrast grammatical acceptability across ten core syntactic and morpho-syntactical phenomena in Urdu. |
| Outcome: | The proposed framework compares multilingual models with the proprietary model . the proposed framework achieves the highest average accuracy on regular phenomena . |
Probing for Constituency Structure in Neural Language Models (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Using standard probing techniques, we examine whether contextual neural language models implicitly learn syntactic structure. |
| Approach: | They investigate to which extent contextual neural language models implicitly learn syntactic structure. |
| Outcome: | The proposed model is able to represent constituents of different categories within the neuron activations of a LM such as RoBERTa with high performance even on manipulated data. |
Similarity Analysis of Contextual Word Representation Models (2020.acl-main)
Copied to clipboard
| Challenge: | Existing and novel similarity measures are used to analyze contextual word representations . different architectures have rather similar representations, but different individual neurons. |
| Approach: | They propose a method to analyze contextual word representation models using similarity analysis. |
| Outcome: | The proposed approach can be used to analyze model similarity without external annotations. |
NxPlain: A Web-based Tool for Discovery of Latent Concepts (2023.eacl-demo)
Copied to clipboard
| Challenge: | Interpretability of deep neural networks has gained a lot of attention in recent years, especially in NLP, where state-of-the-art models are being widely deployed and used in practice. |
| Approach: | They propose to analyze what linguistic and non-linguistic knowledge is learned within deep neural networks and highlight the salient parts of the input. |
| Outcome: | The proposed tool is useful for debugging, unraveling model bias, and for highlighting spurious correlations in a model. |
Analyzing Encoded Concepts in Transformer Language Models (2022.naacl-main)
Copied to clipboard
| Challenge: | a new framework to analyze how latent concepts are encoded in representations learned in pre-trained lan-guage models is proposed . conceptX uses clustering to discover the encoded concepts and align them with a large set of human-defined concepts. |
| Approach: | They propose a framework to analyze how latent concepts are encoded in representations learned within pre-trained lan-guage models. |
| Outcome: | The proposed framework explains encoded concepts by aligning with human-defined concepts. |
On the Transformation of Latent Space in Fine-Tuned NLP Models (2022.emnlp-main)
Copied to clipboard
| Challenge: | a large body of work analyzed the knowledge learned within representations of pre-trained models. |
| Approach: | They use hierarchical clustering to discover latent concepts in representational space . they compare pre-trained and fine-tuned models and perform a thorough analysis . |
| Outcome: | The results show that the model space evolves towards task-specific concepts whereas the lower layers retain generic concepts acquired in the pre-trained model. |
NeuroX Library for Neuron Analysis of Deep NLP Models (2023.acl-demo)
Copied to clipboard
| Challenge: | NeuroX is an open-source toolkit to conduct neuron analysis of natural language processing models. |
| Approach: | They propose a Python toolkit to conduct neuron analysis of natural language processing models. |
| Outcome: | a new open-source toolkit enables neuron analysis of natural language processing models . the framework provides a framework for data processing and evaluation, making it easier for researchers and practitioners to perform neuron analyses. |
AraBench: Benchmarking Dialectal Arabic-English Machine Translation (2020.coling-main)
Copied to clipboard
| Challenge: | Existing efforts to translate Arabic dialects to English are limited due to the lack of evaluation benchmarks. |
| Approach: | They propose an evaluation suite for Arabic to English machine translation using existing Arabic resources. |
| Outcome: | The evaluation suite for Arabic to English machine translation is based on existing evaluation benchmarks. |
Highly Effective Arabic Diacritization using Sequence to Sequence Modeling (N19-1)
Copied to clipboard
| Challenge: | Arabic text is written without short vowels (or diacritics) their presence is essential for properly verbalizing Arabic . |
| Approach: | They propose a character-level sequence-to-sequence deep learning model that recovers both types of diacritics without the use of explicit feature engineering. |
| Outcome: | The proposed model outperforms all previous state-of-the-art models on overlapping windows of words . it achieves a word error rate (WER) of 4.49% compared to the state- of-the art systems . |
Neuron-level Interpretation of Deep NLP Models: A Survey (2022.tacl-1)
Copied to clipboard
| Challenge: | Existing work on deep neural networks has focused on representation analysis, but recent work focused on analyzing neurons within these models. |
| Approach: | They propose to analyze neural networks to uncover linguistic concepts captured by the network . they propose to use a granular approach to analyze neurons within these models . |
| Outcome: | The proposed method combines methods to discover and understand neurons in a network with evaluation methods. |
Not Lost After All: How Cross-Encoder Attribution Challenges Position Bias Assumptions in LLM Summarization (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Position bias is a key limitation in automatic summarization. |
| Approach: | They propose a cross-encoder-based alignment method that processes summary-source sentence pairs . |
| Outcome: | The proposed method allows better identification of semantic correspondences even when summaries substantially rewrite the source. |
Fine-grained Interpretation and Causation Analysis in Deep NLP Models (2021.naacl-tutorials)
Copied to clipboard
| Challenge: | Despite the proven efficacy of deep neural networks at-large, their opaqueness is a major cause of concern. |
| Approach: | They will present research work on interpreting fine-grained components of a neural network model from two perspectives, i) fine-grain interpretation, and ii) causation analysis. |
| Outcome: | This paper presents work on interpreting fine-grained components of a neural network model from two perspectives, i) fine-grain interpretation, and ii) causation analysis. |
Latent Concept-based Explanation of NLP Models (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing attempts to explain deep learning models rely on input features, such as the words . however, such explanations are often less informative due to the discrete nature of words and lack of contextual verbosity. |
| Approach: | They propose a method that generates explanations for predictions based on latent concepts . they map the representations of salient input words into the training latent space . |
| Outcome: | The proposed method generates explanations for predictions based on latent concepts . it maps representations of salient input words into training latent space . |
Fighting the COVID-19 Infodemic: Modeling the Perspective of Journalists, Fact-Checkers, Social Media Platforms, Policy Makers, and the Society (2021.findings-emnlp)
Copied to clipboard
Firoj Alam, Shaden Shaar, Fahim Dalvi, Hassan Sajjad, Alex Nikolov, Hamdy Mubarak, Giovanni Da San Martino, Ahmed Abdelali, Nadir Durrani, Kareem Darwish, Abdulaziz Al-Homaid, Wajdi Zaghouani, Tommaso Caselli, Gijs Danoe, Friso Stolk, Britt Bruntink, Preslav Nakov
| Challenge: | a dataset of 16K manually annotated tweets is used to analyze disinformation . the democratic nature of social media has raised questions about the quality and the factuality of the information that is shared on these platforms. |
| Approach: | They use a dataset of manually annotated tweets to analyze COVID-19 disinformation . they show that tweets contain fake cures, rumors, conspiracy theories and xenophobia . |
| Outcome: | The proposed dataset shows that it is useful in monolingual vs. multilingual settings. |
Immunization against harmful fine-tuning attacks (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Large Language Models are often trained with safety guards to prevent harmful text generation. |
| Approach: | They propose a formal framework based on the training budget of an attacker to validate defenses against harmful fine-tuning attacks. |
| Outcome: | The proposed framework validates whether a model has been fine-tuned against harmful fine-uning attacks on harmful datasets. |
Long-form evaluation of model editing (2024.naacl-long)
Copied to clipboard
| Challenge: | Existing evaluations of model editing only use the ‘next few tokens’ completions after a prompt. |
| Approach: | They propose a new evaluation protocol that measures the efficacy and impact of model editing in long-form generative settings by using a machine-rated survey and a classifier which correlates well with human ratings. |
| Outcome: | The proposed evaluation protocol has little relationship with short-form metrics despite being designed to extend efficacy, generalization, locality, and portability into a long-form setting. |