NLP researcher at NAIST working on machine translation decoding, text-generation evaluation, and LLM text watermarking
4
papers here
1
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
13
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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—
Not yet scored. This axis would measure publication volume and productivity over time from indexed records; raw counts exist (125 works in the last three years) but no composite output score is available yet.
Impact—
Not yet scored. This axis would measure citation influence — how often the work is cited relative to field norms. The snapshot lists h-index 13 and i10-index 17, but no normalized impact score has been computed.
Focus100/100
Scores 100/100, reflecting a highly concentrated topic profile. Topic Modeling (168 topic-assigned works) and Natural Language Processing Techniques (136) dominate the record, with smaller clusters in Speech Recognition and Synthesis (38) and Multimodal Machine Learning Applications (35); Advanced Text Analysis adds 20. The Evolve field distribution lists a single field — natural language processing (LLM text watermarking) — underscoring a tight, NLP-centered research identity.
Collaboration56/100
Scores 56/100. The recent window records 16 distinct coauthors, and all four Evolve-catalog papers are middle-authored rather than solo or corresponding — a team-based pattern. The frequent-coauthor list is anchored by Manabu Okumura, Taro Watanabe, Katsuhiko Hayashi, Kotaro Funakoshi, Boxuan Lyu, and Guangyu Chen, with further collaborators including Yusuke Sakai, Sangjun Moon, Dasom Choi, and Jingun Kwon. The mid-range score suggests steady but not sprawling collaborative breadth relative to peers.
Activity100/100
Scores 100/100, the maximum. OpenAlex counts 125 works in the most recent three years, and annual counts climb from single digits (1 in 2012, 5 in 2019, 8 in 2020) to 28 in 2021, 49 in 2024, 43 in 2025, and 33 already indexed for 2026. This steep, sustained acceleration places the researcher in the top activity band of the snapshot.
Reproducibility100/100
Scores 100/100 (reproMean 100), but on a thin base: only 1 paper has been formally assessed so far. Supporting signals come from Evolve's editorial repro-readiness ratings on the four catalog papers — TripPattern and the SVD/MBR paper are rated high, QUIT and ExpArt-KG medium — with Evolve noting, for example, underspecified statistical methodology in the MBR work. The axis looks strong but should firm up as more papers are assessed.
Research summary
Hidetaka Kamigaito (Nara Institute of Science and Technology, Japan; profile name-resolved from OpenAlex bibliographic data, ORCID 0000-0002-5249-5813) works in natural language processing, concentrating on machine translation decoding, text-generation evaluation, and LLM text watermarking. The bibliometric record shows steep acceleration: annual indexed works rise from single digits through the 2010s to 28 in 2021, 49 in 2024, and 43 in 2025, with 125 works in the most recent three years — the basis for a maximum activity score. Topic allocation is equally concentrated (focus 100): Topic Modeling (168) and Natural Language Processing Techniques (136) dominate, with smaller clusters in speech and multimodal machine learning. The h-index is 13 and the i10-index is 17; output and impact composite scores are not yet computed. Four 2026 arXiv papers sit in the Evolve catalog — TripPattern watermarking, SVD-based overfitting mitigation in MBR decoding, ExpArt-KG artwork description via knowledge-graph RAG, and QUIT early stopping for MT reranking — showing a coherent program on the decoding, evaluation, and efficiency of text generation, typically as a middle author within a team whose recurring collaborators include Manabu Okumura, Taro Watanabe, and Katsuhiko Hayashi. One formally assessed paper carries the maximum reproducibility score (reproMean 100).
Collaboration pattern
The collaboration score is 56/100, with 16 distinct coauthors in the recent window. All four Evolve-catalog papers are middle-authored, consistent with contribution inside a larger team rather than solo work. The frequent-coauthor list is anchored by Manabu Okumura, Taro Watanabe, Katsuhiko Hayashi, Kotaro Funakoshi, Boxuan Lyu, and Guangyu Chen, with additional collaborators including Yusuke Sakai, Sangjun Moon, Dasom Choi, Jingun Kwon, Riza Setiawan Soetedjo, and Yuta Kato.
Research keywords
Highlights
TripPattern: pattern-based LLM watermarking with text quality preserved
QUIT: early stopping cuts MBR reranking cost by up to ~4×
SVD truncation curbs metric overfitting in MBR decoding
ExpArt-KG: knowledge-graph RAG for artwork image description
Timeline
- 2012First indexed works appear in the OpenAlex record (1 work in 2012, with single works again in 2013–2015).
- 2020Indexed output reaches 8 works, the last year before a sharp acceleration.
- 2021Annual works jump to 28, more than triple the 2020 count.
- 2024Single-year indexed output peaks at 49 works.
- 202633 works are already indexed for the year, and four 2026 arXiv papers enter the Evolve catalog spanning LLM watermarking, MBR decoding, knowledge-graph RAG, and MT reranking efficiency.
Papers in the Evolve catalog
TripPattern: A Pattern-based Text Watermarking Method for Large Language Models
Overfitting Mitigation via Singular Value Decomposition in Minimum Bayes Risk Decoding
ExpArt-KG: Artwork Image Description Generation through Iterative Exploration of Knowledge Graphs
Quit While You're Ahead: Quit for Efficient Candidate Generation in Machine Translation Reranking
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 indicators (h-index 13, i10-index 17, yearly counts, topics) come from a single OpenAlex snapshot and may lag or miss some venues.
- Output and impact radar axes are unscored in this snapshot; no citation-normalized impact comparison is available.
- All four catalog papers are 2026 arXiv preprints with middle authorship; peer-reviewed versions and author-contribution details are not captured.
- The reproducibility score rests on one formally assessed paper (reproMean 100); the other repro-readiness ratings are Evolve editorial judgments.