Mark Simmons

PhD Student in Linguistics

Computational Speech Processing | Language documentation | Phonology

About Me

Portrait

I am a PhD student in Linguistics specializing in computational speech processing. I am interested in how Automatic Speech Recognition (ASR), spoken Keyword Search (KWS), and related speech and language processing technologies can benefit linguists engaged in fieldwork and documentation of low-resource languages.

Research Interests

My research focuses on computational methods in language documentation, particularly focusing on adapting foundational ASR architectures for processing fieldwork data. I work with Tira, a Kordofanian language spoken in Sudan. I am interested in how to improve automatic speech recognition on low-resourced languages, especially when code-mixed with high-resource languages, and how ASR methods can aid language documentation. In particular, I am interested in applying ASR to bilingual audio. Most fieldworkers interact with the language community they work with using a meta-language, such as English, French or Spanish, which results in a lot of fieldwork audio being bilingual or code-switched. It's difficult enough to apply ASR to monolingual fieldwork data, bilingual fieldwork data is an extra challenge! Another major hurdle for implementing ASR (or any NLP tool) into linguistic fieldwork is the availability of consistent data. Fieldworkers necessarily work on languages that do not have pre-existing standards for lexicography, orthography, or even phonetic transcription. Furthermore, morphologically rich languages (like Tira!) face a sparsity problem since many inflected forms of a word may be missing from a given dataset. To overcome these challenges, I am currently looking into how spoken keyword search (KWS) can be applied to speed up annotation linguistic fieldwork audio without requiring any training or fine-tuning. I am also interested in exploring how morphological parsers can boostrap high-quality datasets for training NLP algorithms on a fieldwork language with minimal manual annotation, a strategy which takes advantage of the grammatical insight a linguist can provide.

Publications

2025 2022

Presentations and talks

Teaching experience

Languages and skills

Awards and grants

Work experience

Data augmentation for low-resource bilingual ASR from Tira linguistic elicitation using Whisper

Python demo of Forced Align with HMM-GMMs (WIP), pdf

Contact

Email: mjsimmons at ucsd dot edu

GitHub: github.com/markjosims

LinkedIn: www.linkedin.com/in/mark-simmons-794020175/