# UF Data Studio > The UF Data Studio is the research lab of Dr. Christan Grant, Associate Professor in the Department of Computer & Information Science & Engineering (CISE) at the University of Florida. The lab studies the full data pipeline, from sourcing and labeling raw data through modeling to visualization and human interaction, and asks how people can work with their data more effectively and more fairly. The lab is active and growing, and its research has earned awards and recognition. Recent honors include Best Paper at the IEEE ASRU 2025 workshop on AI for Children's Speech and Language and Overall Winner of the AmericasNLP 2026 shared task, with work published at KDD, ICDE, ACL, NAACL, EMNLP, and NeurIPS venues. Graduates have moved on to Amazon Research, Meta, and university faculty. The lab organizes its work around the data pipeline. Projects span data acquisition and labeling, machine learning and natural language processing, multimodal and audio models, and interactive visualization. Most projects share two concerns. They pursue fairness in machine learning, and they build language technology for the accents, languages, and writing systems that mainstream tools underserve. The group sits in Malachowsky Hall for Data Science & Information Technology in Gainesville, Florida, and draws support from the National Science Foundation, Gatorade, Amazon, Colgate-Palmolive, Boeing, and the National GEM Consortium. The research falls into several active threads. **Semasiographic NLP and the Mixtec codices.** The lab adapts core NLP tasks, including part-of-speech tagging, named entity recognition, regular expression search, and entity resolution, to the Mixtec writing system. Mixtec scribes of precolonial and early colonial Mesoamerica recorded history in pictographic codices rather than alphabetic text. Vision models classify codex figures such as gender, name-date, and year glyphs at close to 90% F1 across the Zouche-Nuttall, Selden, and Vindobonensis codices. A parser compiles glyph sequences into abstract syntax trees and renders English interpretations of scenes. The lab's Indigenous-language translation system won the AmericasNLP 2026 shared task. **Audio and speech language models for accented and multilingual speech.** This thread centers African-accented and multilingual speech across speaker diarization, automatic speech recognition, audio quality assessment, and reasoning in Audio Language Models. The Afrispeech Semantics benchmark measures audio-semantic reasoning across domains and accents. A study of whether large audio models understand child stuttering speech won Best Paper at the IEEE ASRU 2025 workshop on AI for Children's Speech and Language. **Algorithmic fairness under domain shift.** The lab builds methods that keep classifiers fair when data distributions move, working with collaborators at UT Dallas and the University of Arkansas. Results cover fairness-aware domain generalization and unsupervised adaptation, fair disentangled online learning for changing environments, and fair demonstration selection for in-context learning, published at KDD, ICDE, and ACL. **Visual analytics for fairness and anomalies.** Ray Chen designs interactive tools that diagnose fairness beyond single-number metrics. RISE sorts signed residuals into interpretable curves and reads localized bias from their inflection points. RISE-NLP compares subgroup residual distributions in text classifiers and links each region of the curve to representative examples. InsightBoard, a TensorBoard plugin, charts fairness metrics during model training, and earlier tools include Wiggum for analyzing mix effects and hybrid tree views for studying gerrymandering. **Vision-language models and accessibility.** The lab benchmarks VLM reasoning and turns it toward access needs. OPTiCAL tests abstract positional reasoning in vision-language models. The comic interpretation project evaluates VLMs that scan and narrate comics and manga for blind and visually impaired readers. **LLM agents and social cognition.** Agents Among Us runs LLMs through multi-agent social deduction games and measures how they collaborate, detect deception, and vote under noise; generative reasoning reaches 50.4% F1 at spotting impostors against an 85% supervised baseline. A companion study evaluates LLM social cognition through strategic games at IVA 2026. **Retrieval, question answering, and human-in-the-loop labeling.** RAMQA performs retrieval-augmented multilingual question answering with language-agnostic reranking, and M3 trains multi-task multi-hop dense retrieval. Speed Labeling designs tactile, wearable, and BCI interfaces with recommender systems to build gold-standard data quickly and fairly while detecting label fatigue, bias, and systematic error. PredictionCertification measures how well LLM forecasts hold across financial, health, policy, weather, and sports domains. ## Research projects - [Semasiographic NLP with Mixtec Codices](https://ufdatastudio.com/projects/mixtec-nlp/): adapting NLP tasks to the pictographic Mixtec writing system. - [Formal Machine Interpretation of Mixtec Codices](https://ufdatastudio.com/projects/formal-machine-interpretation-of-mixtec-codices/): parsing glyph sequences into abstract syntax trees and English scene interpretations. - [Audio Processing of African Accented and Multilingual Speech](https://ufdatastudio.com/projects/audio-processing/): diarization, ASR, and reasoning in Audio Language Models. - [Comic Book Interpretation](https://ufdatastudio.com/projects/comic-interpretation/): vision-language models that read comics aloud for blind and visually impaired users. - [RISE](https://ufdatastudio.com/projects/rise/): interactive visual diagnosis of fairness under domain shift. - [RISE-NLP](https://ufdatastudio.com/projects/risenlp/): residual-distribution fairness auditing for text classifiers. - [Fair AI Plugin](https://ufdatastudio.com/projects/fai-plugin/): a TensorBoard plugin that visualizes fairness during training. - [Agents Among Us](https://ufdatastudio.com/projects/agents-among-us/): a large-scale study of LLMs in multi-agent social deduction games. - [Prediction Certification](https://ufdatastudio.com/projects/predictions-and-certifications/): measuring how correct LLM forecasts are across domains. - [Speed Labeling](https://ufdatastudio.com/projects/speed-labeling/): fast and fair gold-standard labeling with novel interfaces. ## Key pages - [Research overview](https://ufdatastudio.com/research/): the lab's framing of NLP for semasiographic writing. - [Publications](https://ufdatastudio.com/publications/): papers from 2023 onward, filterable by year and type. - [People](https://ufdatastudio.com/people/): current members and alumni. - [Blog](https://ufdatastudio.com/posts/): conference recaps, project write-ups, announcements, talks, and events. (The former /news/ section now redirects here; individual /news/ article links still resolve.) - [Activities and presentations](https://ufdatastudio.com/activities/): lab talks and slides. - [Funding](https://ufdatastudio.com/funding/): sponsors and grants. - [Contact](https://ufdatastudio.com/contact/): how to reach the lab. ## Recent highlights - [Can Large Audio Language Models Understand Child Stuttering Speech?](https://arxiv.org/abs/2510.20850): Best Paper, IEEE ASRU 2025 AI4CSL workshop (Okocha, Bakri, Grant). - [Retrieval-Augmented Long-Context Translation for Cultural Image Captioning](https://arxiv.org/abs/2605.20626): Overall Winner, AmericasNLP 2026 shared task (Dhawan, Driggers-Ellis, Kasinets, Wang, Grant). - [What data should I include in my POS tagging training set?](https://aclanthology.org/2025.findings-emnlp.448): EMNLP 2025 Findings (Liu, Jasbi, Grant, Sagae, Prud'hommeaux). - [Let The Jury Decide: Fair Demonstration Selection for In-Context Learning](https://aclanthology.org/2025.findings-acl.968/): ACL 2025 Findings (Halim et al.). - [RAMQA: Retrieval-Augmented Multilingual Question Answering](https://aclanthology.org/2025.findings-naacl.60/): NAACL 2025 Findings (Bai, Grant, Wang). - [OPTiCAL: An Abstract Positional Reasoning Benchmark for Vision Language Models](https://openreview.net/forum?id=wnOIAbFE6X): NeurIPS 2025 workshop (Driggers-Ellis, Ayoubi, Grant). ## Optional - [Past researchers](https://ufdatastudio.com/people/researchers/): masters, undergraduate, and high school contributors. - [Christan Grant homepage](https://ceg.me): the PI's personal site. - [GitHub organization](https://github.com/ufdatastudio): code and datasets for lab projects. - [LinkedIn](https://www.linkedin.com/company/ufdatastudio): lab updates. - [RSS feed](https://ufdatastudio.com/rss.xml): blog updates.