Theresa Breiner

Theresa Breiner

Theresa Breiner is a Software Engineer on Google's Speech team working on language support and quality, often for languages spoken in regions that have not previously had much access to technology. She specifically works on research and infrastructure scaling efforts to internationalize the language models that support systems for mobile input and speech recognition, and are used in cool Google products like Gboard and the Assistant. Challenges include finding data to train new models and scaling pipelines to seamlessly work for all kinds of languages despite their differences. Theresa earned her Bachelor of Arts in Linguistics and a Master's of Computer and Information Technology (MCIT), both at the University of Pennsylvania in her hometown of Philadelphia.
Authored Publications
Google Publications
Other Publications
Sort By
  • Title
  • Title, descending
  • Year
  • Year, descending
    Preview abstract In this paper we share findings from our effort towards building practical machine translation (MT) systems capable of translating across over one thousand languages. We describe results across three research domains: (i) Building clean, web-mined datasets by leveraging semi-supervised pre-training for language-id and developing data-driven filtering techniques; (ii) Leveraging massively multilingual MT models trained with supervised parallel data for over $100$ languages and small monolingual datasets for over 1000 languages to enable translation for several previously under-studied languages; and (iii) Studying the limitations of evaluation metrics for long tail languages and conducting qualitative analysis of the outputs from our MT models. We hope that our work provides useful insights to practitioners working towards building MT systems for long tail languages, and highlights research directions that can complement the weaknesses of massively multilingual pre-trained models in data-sparse settings. View details
    Preview abstract Personalization of speech models on mobile devices (on-device personalization) is an active area of research, but more often than not, mobile devices have more text-only data than paired audio-text data. We explore training a personalized language model on text-only data, used during inference to improve speech recognition performance for that user. We experiment on a user-clustered LibriSpeech corpus, supplemented with personalized text-only data for each user from Project Gutenberg. We release this User-Specific LibriSpeech (UserLibri) dataset to aid future personalization research. LibriSpeech audio-transcript pairs are grouped into 55 users from the test-clean dataset and 52 users from test-other. We are able to lower the average word error rate per user across both sets in streaming and nonstreaming models, including an improvement of 2.5 for the harder set of test-other users when streaming. View details
    Preview abstract Most studies in cross-device federated learning focus on small models, due to the server-client communication and on-device computation bottlenecks. In this work, we leverage various techniques for mitigating these bottlenecks to train larger language models in cross-device federated learning. With systematic applications of partial model training, quantization, efficient transfer learning, and communication-efficient optimizers, we are able to train a 21M parameter Transformer that achieves the same perplexity as that of a similarly sized LSTM with ~10x smaller client-to-server communication cost and 11% lower perplexity than smaller LSTMs commonly studied in literature. View details
    Preview abstract Pronunciation modeling is a key task for building speech technology in new languages, and while solid grapheme-to-phoneme (G2P) mapping systems exist, language coverage can stand to be improved. The information needed to build G2P models for many more languages can easily be found on Wikipedia, but unfortunately, it is stored in disparate formats. We report on a system we built to mine a pronunciation data set in 819 languages from loosely structured tables within Wikipedia. The data includes phoneme inventories, and for 63 low-resource languages, also includes the grapheme-to-phoneme (G2P) mapping. 54 of these languages do not have easily findable G2P mappings online otherwise. We turned the information from Wikipedia into a structured, machine-readable TSV format, and make the resulting data set publicly available so it can be improved further and used in a variety of applications involving low-resource languages. View details
    Preview abstract Large text corpora are increasingly important for a wide variety of Natural Language Processing (NLP) tasks, and automatic language identification (LangID) is a core technology needed to collect such datasets in a multilingual context. LangID is largely treated as solved in the literature, with models reported that achieve over 90% average F1 on as many as 1,366 languages. We train LangID models on up to 1,629 languages with comparable quality on held-out test sets, but find that human-judged LangID accuracy for web-crawl text corpora created using these models is only around 5% for many lower-resource languages, suggesting a need for more robust evaluation. Further analysis revealed a variety of error modes, arising from domain mismatch, class imbalance, language similarity, and insufficiently expressive models. We propose two classes of techniques to mitigate these errors: wordlist-based tunable-precision filters (for which we release curated lists in about 500 languages) and transformer-based semi-supervised LangID models, which increase median dataset precision from 5.5% to 71.2%. These techniques enable us to create an initial data set covering 100K or more relatively clean sentences in each of 500+ languages, paving the way towards a 1,000-language web text corpus. View details
    Preview abstract We present our approach to automatically generating keyboard layouts on mobile devices for typing low-resource languages written in the Latin script. For many speakers, one of the barriers in accessing and creating text content on the web is the absence of input tools for their language. Ease in typing in these languages would lower technological barriers to online communication and collaboration, likely leading to the creation of more web content. Unfortunately, it can be time-consuming to develop layouts manually even for language communities that use a keyboard layout very similar to English: starting from scratch requires many configuration files to describe multiple possible behaviors for each key. With our approach, we only need a small amount of data in each language to generate keyboard layouts with very little human effort. This process can help serve speakers of low-resource languages in a scalable way, allowing us to develop input tools for more languages. Having input tools that reflect the linguistic diversity of the world will let as many people as possible use technology to learn, communicate, and share their thoughts in their own native languages. View details
    Preview abstract This technical report describes our deep internationalization program for Gboard, the Google Keyboard. Today, Gboard supports 900+ language varieties across 70+ writing systems, and this report describes how and why we added support for these language varieties from around the globe. Many languages of the world are increasingly used in writing on an everyday basis, and we describe the trends we see. We cover technological and logistical challenges in scaling up a language technology product like Gboard to hundreds of language varieties, and describe how we built systems and processes to operate at scale. Finally, we summarize the key take-aways from user studies we ran with speakers of hundreds of languages from around the world. View details
    Mining Training Data for Language Modeling across the World’s Languages
    Proceedings of the 6th International Workshop on Spoken Language Technologies for Under-resourced Languages (SLTU 2018)
    Preview abstract Building smart keyboards and speech recognition systems for new languages requires a large, clean text corpus to train n-gram language models on. We report our findings on how much text data can realistically be found on the web across thousands of languages. In addition, we describe an innovative, scalable approach to normalizing this data: all data sources are noisy to some extent, but this situation is even more severe for low-resource languages. To help clean the data we find across all languages in a scalable way, we built a pipeline to automatically derive the configuration for language-specific text normalization systems, which we describe here as well. View details
    No Results Found