Toyota Technological Institute at Chicago researcher in spoken language modeling and efficient Transformer LMs
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
47
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-15) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
Output is not yet scored: this snapshot provides raw annual counts but no normalized production composite, so this axis would measure publication volume relative to field baselines; no estimate is made.
Impact—
Impact is not scored here: the record lists h-index and i10-index but no citation-normalized or venue-weighted composite, so this axis would measure per-paper citation influence against field benchmarks; no score is estimated.
Focus0/100
The focus dimension scores 0, reflecting an even field split: 50% speech & natural language processing (spoken language modeling) and 50% natural language processing. Topic counts reinforce this breadth — Speech Recognition and Synthesis (164), Speech and Audio Processing (137), Topic Modeling (69), Phonetics and Phonology Research (68), and Natural Language Processing Techniques (65) — pointing to a deliberately two-field, bridge-building agenda rather than concentration in a single niche.
Collaboration39/100
The collaboration dimension scores 39. The raw signal is 6 distinct coauthors in the snapshot window, with a stable recurring group — Ju-Chieh Chou, Jiawei Zhou, Zixuan Lan, Jessica Yang, and Yanhong Li — suggesting a lab-centered pattern where a small set of collaborators recurs across papers. This is consistent with the two 2026 reviewed papers, one with a middle-author slot and one as last author.
Activity100/100
The activity dimension scores 100, the maximum. The counts-by-year series shows uninterrupted output from 1999 through 2026 with no blank years, and a marked acceleration this decade: 24 works in 2022, a peak of 32 in 2024, and 12 already logged in 2026. The recent three-year window holds 57 works, which directly supports the top activity rating.
Reproducibility78/100
The reproducibility dimension scores 78, the mean readiness across the 2 Evolve-reviewed papers. The token-merging distillation study is rated high, while the speech-text modality gap study is rated medium — its headline comparisons depend on an ASR-mediated evaluation metric, which Evolve flags as the key caveat. With only 2 scored papers, this figure should be read as indicative rather than definitive.
Research summary
Karen Livescu is a researcher at Toyota Technological Institute at Chicago (US), with a bibliometric record resolved from OpenAlex and ORCID 0000-0003-4962-946X. The snapshot reports an h-index of 47 and an i10-index of 112, with works logged in every year from 1999 through 2026 and 57 works in the most recent three-year window. Annual counts show sustained growth: single digits through most of the 2000s, 17 works in 2016, 24 in 2022, and a peak of 32 in 2024. The topical profile is dominated by Speech Recognition and Synthesis (164 works) and Speech and Audio Processing (137), with further clusters in Topic Modeling (69), Phonetics and Phonology Research (68), and Natural Language Processing Techniques (65); the field distribution splits evenly between spoken language modeling and natural language processing. Evolve's catalog includes two 2026 arXiv papers: a middle-author study distilling sequential computation in Transformer language models via token merging, rated high for reproducibility readiness, and a last-author matched-data study quantifying the generation modality gap in speech-text language models, reporting generation perplexity of 50-90 with text conditioning versus 190-290 without, with the ASR-mediated metric flagged as the key caveat.
Collaboration pattern
The snapshot lists 6 distinct coauthors with a collaboration score of 39. The recurring core is compact — Ju-Chieh Chou, Jiawei Zhou, Zixuan Lan, Jessica Yang, and Yanhong Li (Jiawei Zhou appears twice in the source list, likely a duplicate name form) — consistent with a lab-centered pattern in which a small, stable group recurs across papers rather than a wide rotating network. The two 2026 reviewed papers fit this: one middle-author, one last-author.
Research keywords
Highlights
Token merging distills sequential computation in Transformer LMs
Matched-data quantification of the speech-text generation modality gap
A two-field bridge: spoken language modeling meets efficient LMs
The 2026 pair spans both halves of the 50/50 field split — one paper compresses Transformer language-model sequences via token merging, the other evaluates speech-text joint generation — and both Evolve verdicts center on evaluation controls, a shared methodological thread.
Timeline
- 1999Earliest works appear in the OpenAlex record, with 1 work logged in this year.
- 2016Annual output reaches 17 works, the highest yearly count in the record up to that point.
- 2022Output climbs to 24 works in a single year, continuing the decade's acceleration.
- 2024Peak annual output: 32 works logged in the counts-by-year series.
- 2026Two arXiv papers enter the Evolve catalog — token-merging distillation for Transformer LMs (middle author) and a matched-data speech-text modality gap study (last author); 12 works logged so far this year.
Papers in the Evolve catalog
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Bibliometric figures (h-index 47, i10-index 112, counts by year, topic clusters) come from an OpenAlex-resolved identity, so name ambiguity is possible even though resolution uses public bibliographic data.
- The output and impact radar dimensions are unscored in this snapshot; only collaboration (39), activity (100), focus (0), and reproducibility (78) have computed values.
- The reproducibility mean of 78 is based on only 2 Evolve-reviewed papers and may not represent the broader catalog.
- Counts-by-year coverage runs 1999-2026; recent years may reflect indexing lag (2026 shows 12 works as of the snapshot).