Papers by Kenton Murray
Efficiency through Auto-Sizing: Notre Dame NLP’s Submission to the WNGT 2019 Efficiency Task (D19-56)
Copied to clipboard
| Challenge: | Notre Dame Natural Language Processing group applied auto-sizing to the Transformer network to reduce the number of parameters in the model. |
| Approach: | They investigated the impact of auto-sizing on the Transformer network by applying a method to inducing sparsity in parameters. |
| Outcome: | The proposed method eliminated more than 25% of the model’s parameters while suffering a decrease of only 1.1 BLEU. |
Condensing Multilingual Knowledge with Lightweight Language-Specific Modules (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods to boost performance in multilingual models but scalability is difficult to manage. |
| Approach: | They propose a method that incorporates language-specific (LS) modules to boost model performance. |
| Outcome: | The proposed method outperforms state-of-the-art methods while outperforming existing methods. |
The Importance of Being Parameters: An Intra-Distillation Method for Serious Gains (2022.emnlp-main)
Copied to clipboard
| Challenge: | Recent pruning methods remove redundant parameters according to parameter sensitivity, a gradient-based measure reflecting the contribution of the parameters. |
| Approach: | They propose a general task-agnostic method to balance parameter sensitivity and a novel adaptive learning method to control strength of intra-distillation loss for faster convergence. |
| Outcome: | The proposed method can reduce redundant parameters by over 80% without obvious performance degradation. |
Auto-Sizing the Transformer Network: Improving Speed, Efficiency, and Performance for Low-Resource Machine Translation (D19-56)
Copied to clipboard
| Challenge: | Neural sequence-to-sequence models are sensitive to architecture and hyperparameter settings. |
| Approach: | They incorporate architecture search into a single training run through auto-sizing . they show that auto-size can improve BLEU scores by up to 3.9 points . |
| Outcome: | The proposed algorithm improves BLEU scores on low-resource language pairs while removing one-third of the parameters from the model. |
CommonLID: Re-evaluating State-of-the-Art Language Identification Performance on Web Data (2026.acl-long)
Copied to clipboard
Pedro Ortiz Suarez, Laurie Burchell, Catherine Arnett, Rafael Mosquera, Sara Hincapié Monsalve, Thom Vaughan, Damian Stewart, Malte Ostendorff, Idris Abdulmumin, Vukosi Marivate, Shamsuddeen Hassan Muhammad, Atnafu Lambebo Tonja, Hend Al-Khalifa, Nadia Ghezaiel Hammouda, Verrah Akinyi Otiende, Tack Hwa Wong, Jakhongir Saydaliev, Melika Nobakhtian, Muhammad Ravi Shulthan Habibi, Chalamalasetti Kranti, Carol Muchemi, Khang Nguyen, Faisal Muhammad Adam, Luis Frentzen Salim, Reem Alqifari, Cynthia Jayne Amol, Joseph Marvin Imperial, Ilker Kesen, Ahmad Mustafid, Pavel Stepachev, Leshem Choshen, David Anugraha, Hamada Nayel, Seid Muhie Yimam, Vallerie Alexandra Putra, My Chiffon Nguyen, Azmine Toushik Wasi, Gouthami Vadithya, Rob Van Der Goot, Lanwenn ar C’horr, Karan Dua, Andrew Yates, Mithil Bangera, Yeshil Bangera, Hitesh Laxmichand Patel, Shu Okabe, Fenal Ashokbhai Ilasariya, Dmitry Gaynullin, Genta Indra Winata, Yiyuan Li, Juan Pablo Martínez, Amit Agarwal, Ikhlasul Akmal Hanif, Raia Abu Ahmad, Esther Adenuga, Filbert Aurelian Tjiaranata, Weerayut Buaphet, Michael Anugraha, Sowmya Vajjala, Benjamin L Rice, Azril Hafizi Amirudin, Jesujoba Oluwadara Alabi, Srikant Panda, Yassine Toughrai, Bruhan Kyomuhendo, Daniel Ruffinelli, null Akshata, Manuel Goulão, Ej Zhou, Ingrid Gabriela Franco Ramirez, Cristina Aggazzotti, Konstantin Dobler, Jun Kevin, Quentin Pagès, Nicholas Andrews, Nuhu Ibrahim, Mattes Ruckdeschel, Amr Keleg, Mike Zhang, Casper Rufaro Muziri, Saron Samuel, Sotaro Takeshita, Kun Kerdthaisong, Luca Foppiano, Rasul Dent, Tommaso Green, Ahmad Mustapha Wali, Kamohelo Makaaka, Vicky Feliren, Inshirah Idris, Hande Celikkanat, Abdulhamid Abubakar, Jean Maillard, Benoît Sagot, Thibault Clérice, Kenton Murray, Sarah K. K. Luger
| Challenge: | Language identification (LID) is a fundamental step in curating multilingual corpora. |
| Approach: | They introduce CommonLID, a community-driven, human-annotated LID benchmark for the web domain, covering 109 languages. |
| Outcome: | The proposed benchmark covers 109 languages and shows that existing evaluations overestimate accuracy for many languages in the web domain. |
Where are you from? Geolocating Speech and Applications to Language Identification (2024.naacl-long)
Copied to clipboard
Patrick Foley, Matthew Wiesner, Bismarck Odoom, Leibny Paola Garcia Perera, Kenton Murray, Philipp Koehn
| Challenge: | Language identification (LID) is a critical component in many modern multilingual speech technologies. |
| Approach: | They propose to use radio broadcasts with known origin to train regression models . they also propose to explore using geolocation as a proxy task for LID . |
| Outcome: | The proposed model outperforms pretrained models on the FLEURS benchmark and on the VoxLingua benchmark. |
Joint Universal Syntactic and Semantic Parsing (2021.tacl-1)
Copied to clipboard
| Challenge: | Several attempts have been made to jointly parse syntax and semantics, but this trade-off is not well understood. |
| Approach: | They propose multiple model architectures that exploit the rich syntactic and semantic annotations contained in the Universal Decompositional Semantics dataset to obtain state-of-the-art results. |
| Outcome: | The proposed model outperforms existing models in 8 languages and their results are consistent across languages. |
Why Does Zero-Shot Cross-Lingual Generation Fail? An Explanation and a Solution (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing studies on cross-lingual transferability of multilingual LMs show that they can perform tasks in low-resource languages. |
| Approach: | They propose a method to regularize the model from learning language invariant representations and a way to select model checkpoints without a development set in the target language. |
| Outcome: | The proposed method reduces the accidental translation problem by 68% and improves the ROUGE-L score by 1.5 on average. |
Query Decomposition for RAG: Balancing Exploration-Exploitation (2026.eacl-long)
Copied to clipboard
Roxana Petcu, Kenton Murray, Daniel Khashabi, Evangelos Kanoulas, Maarten de Rijke, Dawn Lawrie, Kevin Duh
| Challenge: | Complex user queries often involve the exclusion of information, negation, or missing entities. |
| Approach: | They propose to decompose user requests into subqueries, retrieve potentially relevant documents for each and then aggregate them to generate an answer. |
| Outcome: | The proposed method achieves 35% gain in document-level precision and 15% increase in -nDCG . it also improves the downstream task of long-form generation. |
Exploring Geometric Representational Disparities between Multilingual and Bilingual Translation Models (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing work shows that limited modeling capacity is a major contributor to reduced performance in multilingual models. |
| Approach: | They investigate the isotropy of multilingual model decoder representations using intrinsic dimensionality and IsoScore to measure how they utilize the dimensions in their underlying vector space. |
| Outcome: | The proposed model decoder representations are less isotropic and occupy fewer dimensions than bilingual models. |
Evaluating Large Language Models along Dimensions of Language Variation: A Systematik Invesdigatiom uv Cross-lingual Generalization (2024.emnlp-main)
Copied to clipboard
| Challenge: | Xue et al., 2021) show that large language models suffer from performance degradation on unseen closely-related languages and dialects relative to their high-resource language neighbour (HRLN). |
| Approach: | They propose to model phonological, morphological, and lexical distance as Bayesian noise processes to synthesize artificial languages that are controllably distant from the HRLN. |
| Outcome: | The proposed model offers insights on model robustness to isolated and composed linguistic phenomena and the impact of task and HRL characteristics on PD. |
BERT, mBERT, or BiBERT? A Study on Contextualized Embeddings for Neural Machine Translation (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for incorporating pre-trained models into NMT systems are non-trivial and lack a comparison of the impact that other pre-trainers may have on translation performance. |
| Approach: | They propose to use the input of a bilingual pre-trained language model as the input for NMT encoders and a stochastic layer selection approach to ensure sufficient utilization of contextualized embeddings. |
| Outcome: | The proposed bilingual pre-trained language model outperforms all other pre-train models on the IWSLT’14 dataset and the proposed dual-directional translation model. |
Towards Being Parameter-Efficient: A Stratified Sparsely Activated Transformer with Dynamic Capacity (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Recent studies have established that Mixture-of-experts models are parameter-inefficient as the improvement in performance diminishes with an increasing number of experts. |
| Approach: | They propose a mix-of-experts model that uses sparse activation to increase the number of parameters while maintaining low computational requirements per token. |
| Outcome: | The proposed models outperform state-of-the-art models on three multilingual machine translation benchmarks with 4, 15, and 94 language pairs. |
Faux Polyglot: A Study on Information Disparity in Multilingual Large Language Models (2025.naacl-long)
Copied to clipboard
| Challenge: | Recent surge in multilingual large language models (LLMs) and Retrieval Augmented Generation (RAG) has significantly expanded conversational search across varied linguistic and cultural demographics. |
| Approach: | They found that LLMs displayed systemic bias towards information in the same language as query language in document retrieval and answer generation. |
| Outcome: | The results highlight the linguistic divide within multilingual LLMs in information search systems. |
Kreyòl-MT: Building MT for Latin American, Caribbean and Colonial African Creole Languages (2024.naacl-long)
Copied to clipboard
Nathaniel Robinson, Raj Dabre, Ammon Shurtz, Rasul Dent, Onenamiyi Onesi, Claire Monroc, Loïc Grobol, Hasan Muhammad, Ashi Garg, Naome Etori, Vijay Murari Tiyyala, Olanrewaju Samuel, Matthew Stutzman, Bismarck Odoom, Sanjeev Khudanpur, Stephen Richardson, Kenton Murray
| Challenge: | Creole languages are used in much of Latin America, Africa and the Caribbean . a large multilingual bitext like ours has potential to build the best yet or first ever MT models for many languages . |
| Approach: | They present the largest cumulative dataset to date for Creole language MT . they provide MT models supporting all 41 Creoles in 172 translation directions . |
| Outcome: | The proposed model outperforms a genre-specific Creole MT model on its own benchmark for 23 of 34 translation directions. |
DialUp! Modeling the Language Continuum by Adapting Models to Dialects and Dialects to Models (2025.acl-long)
Copied to clipboard
Niyati Bafna, Emily Chang, Nathaniel Romney Robinson, David R. Mortensen, Kenton Murray, David Yarowsky, Hale Sirin
| Challenge: | Recent advances in MT quality and language coverage have shown that language varieties with low baseline performance are more likely to benefit from these approaches. |
| Approach: | They propose a training-time technique for adapting a pretrained model to dialectal data and an inference-time intervention adapting dialectal datasets to the model expertise. |
| Outcome: | The proposed model shows significant performance gains for several dialects from four language families, and modest gains for two other language families. |
Por Qué Não Utiliser Alla Språk? Mixed Training with Gradient Optimization in Few-Shot Cross-Lingual Transfer (2022.findings-naacl)
Copied to clipboard
| Challenge: | a lack of labeled data for low-resource languages leads to the need for effective cross-lingual transfer learning. |
| Approach: | They propose a mixed training method that trains on both source and target data with stochastic gradient surgery, a novel gradient-level optimization. |
| Outcome: | The proposed method outperforms current methods on all tasks and escapes overfitting issues. |
Narrowing the Gap between Zero- and Few-shot Machine Translation by Matching Styles (2024.findings-naacl)
Copied to clipboard
Weiting Tan, Haoran Xu, Lingfeng Shen, Shuyue Stella Li, Kenton Murray, Philipp Koehn, Benjamin Van Durme, Yunmo Chen
| Challenge: | Recent work shows that large language models can generalize to machine translation using zero-shot examples with in-context learning. |
| Approach: | They investigate the factors contributing to this gap by matching the writing styles of the target corpus. |
| Outcome: | The proposed methods can be enhanced without the need for parallel demonstration examples. |
Whisper-UT: A Unified Translation Framework for Speech and Text (2025.emnlp-main)
Copied to clipboard
Cihan Xiao, Matthew Wiesner, Debashish Chakraborty, Reno Kriz, Keith Cunningham, Kenton Murray, Kevin Duh, Luis Tavarez-Arce, Paul McNamee, Sanjeev Khudanpur
| Challenge: | Encoder-decoder models have achieved remarkable success in speech and text tasks, but efficiently adapting them to diverse uni/multimodal scenarios remains a challenge. |
| Approach: | They propose a framework that leverages lightweight adapters to enable seamless adaptation across tasks. |
| Outcome: | The proposed framework improves speech translation performance through a 2-stage decoding strategy without requiring 3-way parallel data. |
Upsample or Upweight? Balanced Training on Heavily Imbalanced Datasets (2025.naacl-long)
Copied to clipboard
| Challenge: | a lack of data across domains creates significant imbalances in training data sizes . a recent study shows that temperature sampling and scaling are equivalent but differ under stochastic gradient descent due to differences in gradient variance. |
| Approach: | They propose a method that upsamples low-resource languages and upweights their loss functions to address this disparity. |
| Outcome: | The proposed method competes effectively with existing data re-weighting techniques while offering computational efficiency. |
Everything Is All It Takes: A Multipronged Strategy for Zero-Shot Cross-Lingual Information Extraction (2021.emnlp-main)
Copied to clipboard
Mahsa Yarmohammadi, Shijie Wu, Marc Marone, Haoran Xu, Seth Ebner, Guanghui Qin, Yunmo Chen, Jialiang Guo, Craig Harman, Kenton Murray, Aaron Steven White, Mark Dredze, Benjamin Van Durme
| Challenge: | Zero-shot cross-lingual information extraction (IE) is a technique for training data in a source language but not in . |
| Approach: | They explore techniques including data projection and self-training to improve zero-shot cross-lingual information extraction (IE) IE is a construction of an IE model for some target language given existing annotations exclusively in English. |
| Outcome: | The proposed techniques show that they perform better than any single strategy. |