Research
My field is in natural language processing and machine learning. I study edge language models, or small models that can run on commodity devices or under limited connectivity, with the motivation of enabling the development and deployment of language technologies for the Global South.
I want to find out how we can pack strong capabilities that the Global South cares about into a small model footprint, given that most of these communities don’t have the resources to run models at frontier scale. I’m a strong proponent of data-centric approaches, with a belief that data quality is a compute multiplier: high-quality data leads to more signal per token, letting us build smaller, more capable models.
I’m interested in techniques such as knowledge distillation, synthetic data generation, and building faithful benchmarks. Currently, I’m thinking about how we can start with the actual needs of Global South communities and work our way backwards to right-size our data and compute interventions.
Selected Publications
Below is a selection of work that reflects my current interests. My work has been published in top NLP conferences such as ACL, NAACL, and EMNLP. I’m always excited for potential internships or research visits, so just reach out if you find me a good match!
Keywords: data-centric NLP, multilinguality, resources & evaluation
Multilinguality at the Edge: Developing Language Models for the Global South
Preprint ‘26
Lester James V. Miranda, Songbo Hu, Roi Reichart, and Anna Korhonen
We survey 232 papers across the multilingual and edge NLP literatures and provide a research agenda for bringing language models to the Global South.
Polyglot Teachers: Evaluating Models for Multilingual Synthetic Data Generation
EMNLP ‘26
Lester James V. Miranda, Ivan Vulić, and Anna Korhonen
We systematically analyze how to select teacher models and offer a recipe for multilingual synthetic data generation.
Hybrid Preferences: Learning to Route Instances for Human vs. AI Feedback
ACL ‘25 (work done as pre-doc at Ai2)
Lester James V. Miranda★, Yizhong Wang★, Yanai Elazar, Sachin Kumar, Valentina Pyatkin, Faeze Brahman, Noah A. Smith, Hannaneh Hajishirzi, and Pradeep Dasigi
We build a router that decides whether each preference instance for RLHF should be annotated by a human or a language model.
M-RewardBench: Evaluating Reward Models in Multilingual Settings
ACL ‘25 (work done as part of Expedition Aya w/ Cohere Labs)
Srishti Gureja★, Lester James V. Miranda★, Shayekh bin Islam★, Rishabh Maheshwary★, Drishti Sharma, Gusti Winata, Nathan Lambert, Sebastian Ruder, Sara Hooker, and Marzieh Fadaee
We build the first multilingual reward-model benchmark, covering 23 typologically diverse languages, and show that reward models carry a substantial gap from English into every other language.
I have been involved in several open-model efforts such as Tülu 3, OLMo 2 (synthetic preferences for RLHF) and OLMo 3 (synthetic tool-use trajectories for SFT) during my days as a pre-doc in Ai2. I also contributed datasets in the SEACrowd project for Southeast Asian languages.
Finally, I care a lot about advancing Filipino NLP and representing my native language. This involves:
- developing benchmarks such as FilBench-Eval (EMNLP ‘25),
- building resources such as UD-NewsCrawl (ACL ‘25) and TLUnified-NER (SEALP ‘23), and
- creating open-source tools like calamanCy (NLP-OSS ‘23).
I write a lot about Filipino NLP in this blog and organize researchers on collaborative projects through the FilBench collective.
Invited Talks
- Gumagana ba talaga yan sa Filipino? Evaluating and Building Language Models for the Philippines. Analytics & AI Association of the Philippines. Hosted by Michelle Alarcon. 2026.
- NLP Seminar for the Philippine Team at the International Olympiad in AI. Philippine AI Olympiad Team. Hosted by Martin Gomez. 2026.
- Artisanal Filipino Resources in the Age of LLMs. De La Salle University - Manila (Graduate Seminar). Hosted by Dr. Charibeth Cheng. 2024.
- Labeling with LLMs. University of North Carolina - Charlotte (Course Seminar). Hosted by Dr. Ryan Wesslen and Chang Hsin Lee. 2024.
Research Posters
I enjoy making posters for my research projects, and still strive to improve this craft. Here are some that I presented in various conferences and workshops!
