Papers by David Lee

19 papers
AlignFreeze: Navigating the Impact of Realignment on the Layers of Multilingual Models Across Diverse Languages (2025.naacl-short)

Copied to clipboard

Challenge: Realignment techniques are often employed to enhance cross-lingual transfer in multilingual language models, but can degrade performance in languages that differ significantly from the fine-tuned source language.
Approach: They propose a method that freezes either the lower half or upper half of the layers during realignment to prevent performance degradation.
Outcome: The proposed method improves Part-of-Speech (PoS) tagging performance in languages where realignment fails.
Unlocking Parameter-Efficient Fine-Tuning for Low-Resource Language Translation (2024.findings-naacl)

Copied to clipboard

Challenge: Parameter-efficient fine-tuning (PEFT) methods are important in low-resource language (LRL) Neural Machine Translation (NMT) but their practical effectiveness varies significantly across different languages.
Approach: They evaluated the performance of 8 parameters-efficient fine-tuning methods with 15 architectures using the SacreBLEU score.
Outcome: The Houlsby+Inversion adapter outperforms the baseline architectures in both in-domain and out-domain tests and the Houlson+Inverter achieves the best performance overall.
Practical Correlated Topic Modeling and Analysis via the Rectified Anchor Word Algorithm (D19-1)

Copied to clipboard

Challenge: spectral topic models lack reliability in real data and lack of practical implementations.
Approach: They propose to use a spectral topic inference method to infer correlations between topics in real data and a matrix-based approach to inference.
Outcome: The proposed method outperforms tensor-based methods and probabilistic methods in real data and provides a complete guide to correlated topic modeling.
On the Empirical Complexity of Reasoning and Planning in LLMs (2024.findings-emnlp)

Copied to clipboard

Challenge: Evidence shows that the relative performance of CoT, ToT, and their variants may vary from task to task.
Approach: They propose to use chain-of-thought (CoT), tree-of thought (ToT), and related techniques to solve complex reasoning tasks with Large Language Models.
Outcome: The proposed methods outperform the linear structure of CoT on hard reasoning tasks.
ProxyLM: Predicting Language Model Performance on Multilingual Tasks via Proxy Models (2025.findings-naacl)

Copied to clipboard

Challenge: Performance prediction is a method to estimate the performance of Language Models (LMs) on various Natural Language Processing (NLP) tasks.
Approach: They propose a task- and language-agnostic framework to predict the performance of Language Models (LMs) using proxy models.
Outcome: The proposed framework outperforms the state-of-the-art in root-mean-square error (RMSE) and other robustness tests on multilingual NLP tasks.
Quantifying the Visual Concreteness of Words and Topics in Multimodal Datasets (N18-1)

Copied to clipboard

Challenge: Existing work suggests that concepts with concrete visual manifestations are easier to learn than abstract ones.
Approach: They propose an algorithm for automatically computing the visual concreteness of words and topics within multimodal datasets.
Outcome: The proposed algorithm predicts the capacity of machine learning algorithms to learn textual/visual relationships.
Unsupervised Discovery of Multimodal Links in Multi-image, Multi-sentence Documents (D19-1)

Copied to clipboard

Challenge: a structured training objective based on identifying whether collections of images and sentences co-occur in documents can suffice to predict links between specific images and specific sentences.
Approach: They propose algorithms that discover image-sentence relationships without explicit annotation . they experiment on seven datasets of varying difficulty .
Outcome: The proposed algorithms can predict links between images and sentences without explicit multimodal annotation.
You don’t need a personality test to know these models are unreliable: Assessing the Reliability of Large Language Models on Psychometric Instruments (2024.naacl-long)

Copied to clipboard

Challenge: Large Language Models (LLMs) are popular for research in social sciences . currently, prompting LLMs is insufficient to accurately and reliably capture model perceptions, and we discuss potential alternatives to improve this.
Approach: They construct a dataset that contains 693 questions encompassing 39 different instruments of persona measurement on 115 persona axes and a set of questions containing minor variations.
Outcome: The proposed model can generate answers and negate statements in a consistent and robust manner.
Pre-Trained Multilingual Sequence-to-Sequence Models: A Hope for Low-Resource Language Translation? (2022.findings-acl)

Copied to clipboard

Challenge: Pre-trained multilingual sequence-to-sequence models like mBART and mT5 can be used to translate low-resource languages, but their practical application is unclear.
Approach: They conduct an empirical experiment in 10 languages to determine what can pre-trained multilingual sequence-to-sequence models like mBART do to translate low-resource languages?
Outcome: The proposed models are robust to domain differences, but translations for unseen and typologically distant languages remain below 3.0 BLEU.
CReSE: Benchmark Data and Automatic Evaluation Framework for Recommending Eligibility Criteria from Clinical Trial Information (2024.findings-eacl)

Copied to clipboard

Challenge: Eligibility criteria (EC) are defined as a set of conditions an individual must meet to participate in a clinical trial.
Approach: They propose to recommend EC based on clinical trial information, including trial titles, and introduce an automatic evaluation framework to assess clinical validity of the EC recommendation model.
Outcome: The proposed model outperforms existing language models pre-trained on the biomedical domain in EC clustering.
WorldCuisines: A Massive-Scale Benchmark for Multilingual and Multicultural Visual Question Answering on Global Cuisines (2025.naacl-long)

Copied to clipboard

Challenge: Vision Language Models struggle with cultural-specific knowledge, especially in languages other than English and in underrepresented cultural contexts.
Approach: They propose a visual question answering (VQA) dataset with text-image pairs across 30 languages and dialects and a training dataset.
Outcome: The proposed model performs better with correct location context, but struggles with adversarial contexts and predicting specific regional cuisines and languages.
Predicting Machine Translation Performance on Low-Resource Languages: The Role of Domain Similarity (2024.findings-eacl)

Copied to clipboard

Challenge: Existing approaches for predicting the performance of NLP models for low-resource languages (LRLs) focus on high-resourced languages, overlooking LRLs and domain shifts.
Approach: They investigate the impact of domain similarity on predicting performance of machine translation models in low-resource languages.
Outcome: The results show that domain similarity has the most important impact on predicting the performance of Machine Translation models.
A Pretrainer’s Guide to Training Data: Measuring the Effects of Data Age, Domain Coverage, Quality, & Toxicity (2024.naacl-long)

Copied to clipboard

Challenge: a large number of pretraining data design practices are under-documented, authors say . authors: strong performance of modern language models depends on selfsupervised pretraining .
Approach: They propose to pretrain models on data curated at different collection times . they find temporal shift between evaluation data and pretraining data leads to performance degradation .
Outcome: The results validate, quantify, and expose many undocumented intuitions about text pretraining . authors say this practice has outperformed other models in the field .
URIEL+: Enhancing Linguistic Inclusion and Usability in a Typological and Multilingual Knowledge Base (2025.coling-main)

Copied to clipboard

Challenge: URIEL is limited in terms of linguistic inclusion and overall usability . URIel+ provides robust, customizable distance calculations to better suit the needs of users.
Approach: They propose a new version of URIEL and a query tool that provides a standardized approach to representing languages as geographical, phylogenetic, and typological vectors.
Outcome: URIEL+ expands the user experience with robust, customizable distance calculations to better suit the needs of users.
When ”A Helpful Assistant” Is Not Really Helpful: Personas in System Prompts Do Not Improve Performances of Large Language Models (2024.findings-emnlp)

Copied to clipboard

Challenge: Commercial AI systems often define the role of the LLM in system prompts.
Approach: They conduct a systematic evaluation of personas in system prompts by adding 162 roles covering 6 types of interpersonal relationships and 8 domains of expertise.
Outcome: The proposed model does not improve performance in the system prompt setting where no persona is added.
Fˆ2-Softmax: Diversifying Neural Text Generation via Frequency Factorized Softmax (2020.emnlp-main)

Copied to clipboard

Challenge: Existing methods for text generation do not fully reflect the rich diversity of human language.
Approach: They propose to use F2-Softmax and MefMax to train a balanced frequency distribution using a frequency class-based method.
Outcome: The proposed methods improve the diversity and quality of generated texts.
MERLIN: Multi-Stage Curriculum Alignment for Multilingual Encoder-LLM Integration in Cross-Lingual Reasoning (2026.eacl-long)

Copied to clipboard

Challenge: Existing methods to align large language models with multilingual encoders raise accuracy for low-resource languages (LRLs) but performance of LLMs in low- and high-resourced languages remains a problem.
Approach: They propose a model-stacking framework that iteratively refines in 2-stages based on a curriculum strategy and adapts only a small set of DoRA weights.
Outcome: The proposed framework improves exact-match accuracy by +12.9 pp over MindMerger and outperforms GPT-4o-mini by 15.2 pp on the AfriMGSM benchmark.
SIB-200: A Simple, Inclusive, and Big Evaluation Dataset for Topic Classification in 200+ Languages and Dialects (2024.eacl-long)

Copied to clipboard

Challenge: despite progress in building multilingual language models evaluation is limited to a few languages with available datasets . despite this, we create a large-scale open-sourced benchmark dataset for topic classification in 205 languages and dialects to address the lack of evaluation dataset for Natural Language Understanding (NLU).
Approach: They create a large-scale open-sourced benchmark dataset for topic classification in 205 languages and dialects to address the lack of evaluation dataset for Natural Language Understanding (NLU).
Outcome: The proposed dataset addresses the lack of evaluation dataset for Natural Language Understanding (NLU) for many languages, it is the first publicly available evaluation dataset.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations