Papers by Marek Rei
Grammatical Error Correction in Low Error Density Domains: A New Benchmark and Analyses (2020.emnlp-main)
Copied to clipboard
| Challenge: | CWEB is a new benchmark for grammatical error correction (GEC) systems . website data contains far fewer grammamatical errors than learner essays . |
| Approach: | They propose to broaden the target domain of grammatical error correction (GEC) systems . website data contains far fewer grammamatical errors than learner essays . |
| Outcome: | The proposed model can't rely on a strong internal language model in low error density domains. |
No Need for Explanations: LLMs can implicitly learn from mistakes in-context (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing literature assumes that correct answers to large language models must be accompanied by comprehensive rationales to be helpful. |
| Approach: | They propose to show incorrect answers to Large Language Models (LLMs) as a popular strategy to improve their performance in reasoning-intensive tasks. |
| Outcome: | The proposed approach outperforms chain-of-thought prompting in math reasoning tasks. |
Verbal Multiword Expressions for Identification of Metaphor (2020.acl-main)
Copied to clipboard
| Challenge: | Metaphor is a linguistic device in which a concept is expressed by mentioning another . Verbal MWEs are examples of non-literal language in which multiple words form a single unit of meaning. |
| Approach: | They propose to analyze the interplay between metaphor and multiword expressions processing by informing the model of the presence of MWEs. |
| Outcome: | The proposed architecture reach state-of-the-art on two established metaphor datasets. |
Distilling Robustness into Natural Language Inference Models with Domain-Targeted Augmentation (2024.findings-acl)
Copied to clipboard
| Challenge: | Knowledge distillation optimises a smaller student model to behave similarly to a larger teacher model, retaining some performance benefits. |
| Approach: | They propose to augment the distillation with generated unlabelled examples that match the target distribution and upsamples data points among the training set that are similar to the target. |
| Outcome: | The proposed method outperforms previous robustness solutions on the task of natural language inference (NLI) it also improves performance on OOD domains even beyond the target domain. |
Modelling the interplay of metaphor and emotion through multitask learning (D19-1)
Copied to clipboard
| Challenge: | Existing research suggests that metaphorical phrases are more emotionally evocative than their literal counterparts. |
| Approach: | They propose a joint model of the relationship between metaphor and emotion within a computational framework by using hard and soft parameter sharing. |
| Outcome: | The proposed model advances the state of the art in both of these tasks. |
An Extended Sequence Tagging Vocabulary for Grammatical Error Correction (2023.findings-eacl)
Copied to clipboard
| Challenge: | Current sequence-to-sequence and sequence-tagging approaches treat GEC as a machine-translation problem. |
| Approach: | They propose to introduce specialised tags for spelling correction and morphological inflection using the SymSpell and LemmInflect algorithms. |
| Outcome: | The proposed approach outperforms existing methods on the BEA benchmark. |
A Simple and Robust Approach to Detecting Subject-Verb Agreement Errors (N19-1)
Copied to clipboard
| Challenge: | a recent study shows that neural sequential labelers overfit their training data to detect SVA errors. |
| Approach: | They propose a simple protocol that generates a neural sequential labeler from silver standard data and gold standard data. |
| Outcome: | The proposed method leads to more robust detection of SVA errors on silver standard data and gold standard data. |
Logical Reasoning with Span-Level Predictions for Interpretable and Robust NLI Models (2022.emnlp-main)
Copied to clipboard
| Challenge: | Current models learn from annotation artefacts and dataset biases, but it is unclear to what extent they are learning the task of NLI. |
| Approach: | They propose a logical reasoning framework that allows models to learn from annotation artefacts and dataset biases. |
| Outcome: | The proposed model outperforms humans on in-distribution test sets without using span labels . the model is more robust in a reduced data setting, and out-of-disturbance performance is improved . |
Prompting open-source and commercial language models for grammatical error correction of English learner text (2024.findings-acl)
Copied to clipboard
Christopher Davis, Andrew Caines, O Andersen, Shiva Taslimipoor, Helen Yannakoudakis, Zheng Yuan, Christopher Bryant, Marek Rei, Paula Buttery
| Challenge: | Recent advances in generative AI have enabled us to prompt large language models (LLMs) to produce texts which are fluent and grammatical. |
| Approach: | They evaluate model performance by measuring their performance on established benchmarks. |
| Outcome: | The proposed models outperform supervised English GEC models on fluency correction benchmarks and commercial LLMs on edit benchmarks. |
DiffuseDef: Improved Robustness to Adversarial Attacks via Iterative Denoising (2025.acl-long)
Copied to clipboard
| Challenge: | Existing adversarial defense methods for natural language processing still pose challenges to adversarials. |
| Approach: | They propose a novel adversarial defense method that incorporates a diffusion layer as a denoiser between the encoder and the classifier. |
| Outcome: | The proposed method improves over existing adversarial defense methods and achieves state-of-the-art performance against black-box and white-box adversarials. |
Seeing Both the Forest and the Trees: Multi-head Attention for Joint Classification on Different Compositional Levels (2020.coling-main)
Copied to clipboard
| Challenge: | Neural networks can capture expressive language features, but insights into the link between words and sentences are difficult to acquire automatically. |
| Approach: | They propose a deep neural network architecture that explicitly wires lower and higher linguistic components and evaluate its ability to perform the same task at different hierarchical levels. |
| Outcome: | The proposed model outperforms equivalent models that are not incentivized towards compositional representations. |
Scoring Lexical Entailment with a Supervised Directional Similarity Network (P18-2)
Copied to clipboard
| Challenge: | Existing word embeddings that use supervision only improve the embeddable word embeds of words with annotated lexical relations. |
| Approach: | They propose a supervised directional similarity network for learning task-specific transformation functions on top of general-purpose word embeddings. |
| Outcome: | The proposed model outperforms existing models on the HyperLex dataset on a directional graded lexical entailment task by 25%. |
Improving the OOD Performance of Closed-Source LLMs on NLI Through Strategic Data Selection (2026.findings-eacl)
Copied to clipboard
| Challenge: | Existing methods to improve robustness require changing the fine-tuning process or large-scale data augmentation, which are infeasible or cost prohibitive for closed-source models. |
| Approach: | They propose to prioritize more complex examples or replace existing training examples with LLM-generated data to improve performance on OOD NLI datasets. |
| Outcome: | The proposed methods improve performance on difficult OOD datasets while training with synthetic data leads to substantial improvements on easier OOD data. |
Memorisation versus Generalisation in Pre-trained Language Models (2022.acl-long)
Copied to clipboard
| Challenge: | State-of-the-art pre-trained language models have been shown to memorise facts and perform well with limited amounts of training data. |
| Approach: | They propose to extend pre-trained language models to generalise and memorise facts in noisy and low-resource scenarios. |
| Outcome: | The proposed extension improves performance in low-resource named entity recognition tasks. |
Guiding Visual Question Generation (2022.naacl-main)
Copied to clipboard
| Challenge: | Existing approaches to Visual Question Generation (VQG) are trained to mimic an arbitrary choice of concept but only one or a few are captured by the human references. |
| Approach: | They propose a variant of Visual Question Generation which conditions the question generator on categorical information based on expectations on the type of question and the objects it should explore. |
| Outcome: | The proposed model improves on the current state of the art on an answer-category augmented VQA dataset and human evaluation validates that guidance helps the generation of questions that are grammatically coherent and relevant to the given image and objects. |
Zero-Shot Sequence Labeling: Transferring Knowledge from Sentences to Tokens (N18-1)
Copied to clipboard
| Challenge: | Recent work has used attention weights to visualize the focus of neural models in input data. |
| Approach: | They propose to use attention-based visualization techniques to infer token-level labels from a network trained only on sentence-level labeling. |
| Outcome: | The proposed approach outperforms gradient-based methods on four datasets and is expected to outperfect supervised methods. |
Bad Form: Comparing Context-Based and Form-Based Few-Shot Learning in Distributional Semantic Models (D19-61)
Copied to clipboard
| Challenge: | Word embeddings are an essential component of many natural language processing applications. |
| Approach: | They propose 3 new tasks to obtain higher-quality vectors for word embeddings . they use word forms in training data that are related to word forms themselves . |
| Outcome: | The proposed methods improve the performance of both baseline and advanced models on 4 out of 6 tasks. |
GiBERT: Enhancing BERT with Linguistic Information using a Lightweight Gated Injection Method (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Recent pre-trained language models such as BERT have led to noticeable improvements in semantic similarity detection. |
| Approach: | They propose to explicitly inject linguistic information in the form of word embeddings into any layer of a pre-trained BERT. |
| Outcome: | The proposed method improves on multiple semantic similarity datasets and shows that it is beneficial and currently missing from the original model. |
Variable Typing: Assigning Meaning to Variables in Mathematical Text (N18-1)
Copied to clipboard
| Challenge: | Scientific documents rely on mathematics to communicate ideas and results . textual contexts are strong domainspecific conventions governing how content is presented . |
| Approach: | They introduce a task of assigning one mathematical type to each variable in a sentence . they also introduce 'variable typing' task that focuses on assignment of meaning to variables . |
| Outcome: | The proposed model is the best performing model on an extrinsic task, the authors show . their model is compared to a formula index only containing raw symbols . |
Multimodal Conversation Modelling for Topic Derailment Detection (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing work on analysing textual dialogues that derailed into toxic content ignores visual information, such as images and videos. |
| Approach: | They propose a new multimodal conversational architecture that utilises visual and conversational contexts to classify comments for derailment. |
| Outcome: | The proposed approach outperforms existing methods and is more robust to textual noise. |
Visual Cues and Error Correction for Translation Robustness (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Existing robustness techniques fail when faced with unseen types of noise and their performance degrades on clean texts. |
| Approach: | They propose visual context to improve translation robustness for noisy texts . they also propose an error correction training regime that can be used as an auxiliary task . |
| Outcome: | The proposed training regime improves translation robustness on noisy texts while maintaining translation quality on clean texts. |
AfriMTE and AfriCOMET: Enhancing COMET to Embrace Under-resourced African Languages (2024.naacl-long)
Copied to clipboard
Jiayi Wang, David Adelani, Sweta Agrawal, Marek Masiak, Ricardo Rei, Eleftheria Briakou, Marine Carpuat, Xuanli He, Sofia Bourhim, Andiswa Bukula, Muhidin Mohamed, Temitayo Olatoye, Tosin Adewumi, Hamam Mokayed, Christine Mwase, Wangui Kimotho, Foutse Yuehgoh, Anuoluwapo Aremu, Jessica Ojo, Shamsuddeen Muhammad, Salomey Osei, Abdul-Hakeem Omotayo, Chiamaka Chukwuneke, Perez Ogayo, Oumaima Hourrane, Salma El Anigri, Lolwethu Ndolela, Thabiso Mangwana, Shafie Mohamed, Hassan Ayinde, Oluwabusayo Awoyomi, Lama Alkhaled, Sana Al-azzawi, Naome Etori, Millicent Ochieng, Clemencia Siro, Njoroge Kiragu, Eric Muchiri, Wangari Kimotho, Toadoum Sari Sakayo, Lyse Naomi Wamba, Daud Abolade, Simbiat Ajao, Iyanuoluwa Shode, Ricky Macharm, Ruqayya Iro, Saheed Abdullahi, Stephen Moore, Bernard Opoku, Zainab Akinjobi, Abeeb Afolabi, Nnaemeka Obiefuna, Onyekachi Ogbu, Sam Ochieng’, Verrah Otiende, Chinedu Mbonu, Yao Lu, Pontus Stenetorp
| Challenge: | Recent advances in machine translation (MT) have focused on scaling multilingual machine translation models and evaluation data to hundreds of languages, including multiple under-resourced languages. |
| Approach: | They propose to use n-gram matching metrics to measure progress in multilingual machine translation to 13 typologically diverse African languages to create high-quality human evaluation data with simplified MQM guidelines. |
| Outcome: | The proposed metrics have a higher correlation with human judgments than n-gram matching metrics such as BLEU and METEOR. |
When and Why Does Bias Mitigation Work? (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Neural models exploit shallow surface features to perform language understanding tasks, rather than learning the deeper language understanding and reasoning skills that practitioners desire. |
| Approach: | They propose to use model debiasing techniques to pressure models away from spurious features and to use them to learn useful representations instead. |
| Outcome: | The proposed methods increase models' reliance on hidden biases instead of learning robust features that help them solve a task. |
Meta-Reasoning Improves Tool Use in Large Language Models (2025.findings-naacl)
Copied to clipboard
| Challenge: | External tools have proven effective at boosting the performance of large language models in knowledge-intensive tasks such as QA and math problem-solving. |
| Approach: | They propose a tool-based meta-reasONing system that makes a final decision based on a custom language modelling head. |
| Outcome: | The proposed system improves the performance of large language models in QA and math problem-solving tasks. |
Multidirectional Associative Optimization of Function-Specific Word Representations (2020.acl-main)
Copied to clipboard
| Challenge: | Existing approaches to learning word associations rely on distributional hypothesis and learn a single word vector space based on word co-occurrences in large text corpora. |
| Approach: | They propose to induce a joint function-specific word vector space where vectors of plausible SVO compositions lie close together. |
| Outcome: | The proposed model outperforms existing models on selectional preference and event similarity tasks while reducing the number of parameters by up to 95%. |
Atomic Inference for NLI with Generated Facts as Atoms (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing models that can provide accurate explanations are not interpretable, i.e. they do not reflect the inner workings of the model. |
| Approach: | They propose to use LLM-generated facts as atoms to make interpretable models that can be used to make accurate predictions for each component part of an input. |
| Outcome: | The proposed method outperforms existing methods on natural language understanding tasks with a multi-stage fact generation process and a training regime that incorporates the facts. |
Semi-Supervised Bootstrapping of Dialogue State Trackers for Task-Oriented Modelling (D19-1)
Copied to clipboard
| Challenge: | Existing systems rely on modular, domain-focused frameworks for analyzing complex problems. |
| Approach: | They propose semi-supervised learning methods that can reduce the amount of required intermediate labelling by leveraging un-annotated data instead of transcribed utterances. |
| Outcome: | The proposed model reduces the amount of turn-level annotations by 30% while maintaining equivalent system performance. |
Modelling Temporal Document Sequences for Clinical ICD Coding (2023.eacl-main)
Copied to clipboard
| Challenge: | Existing studies on the ICD coding task focus on extracting codes from the discharge summary, but there is potential to automate the task by identifying relevant information from clinical notes. |
| Approach: | They propose a hierarchical transformer architecture that uses text across the entire sequence of clinical notes in each hospital stay for ICD coding. |
| Outcome: | The proposed model exceeds the state-of-the-art when using only discharge summaries as input and achieves performance improvements when all clinical notes are used as input. |
AgentCoMa: A Compositional Benchmark Mixing Commonsense and Mathematical Reasoning in Real-World Scenarios (2026.acl-long)
Copied to clipboard
| Challenge: | brittleness of Large Language Models in reasoningintensive tasks is a problem . current compositional benchmarks focus on *either* commonsense or math reasoning . |
| Approach: | They propose a "Co**mmonsense and Ma**th" benchmark where each compositional task requires a commonsense reasoning step *and* a math reasoning step. |
| Outcome: | The proposed benchmarks show that LLMs can solve both steps in isolation, but their accuracy drops by nearly 30% when the two steps are combined. |
Grammatical error detection in transcriptions of spoken English (2020.coling-main)
Copied to clipboard
| Challenge: | CrowdED corpus of spoken English monologues on business topics was crowdsourced from native speakers of English and learners of English with German as their first language. |
| Approach: | They propose to use the corpus recordings to correct existing speech transcriptions and edit them to make them more fluent. |
| Outcome: | The proposed transcription corrections and annotations can be used for automatic transcription post-editing and grammatical error correction for spoken English. |