Language & learning.
Building with language, learning through systems.
Researcher & product builder / 2024 — ongoing
Language learners need usable resources, context and opportunities to practise.
Translation-platform development, corpus-platform work, and backend/product direction for Yakky.
A first-author paper, a second-author corpus paper, registered software and a speaking-product prototype.
Build a platform. Study how it is used.
Sinolingo brings prompt templates, terminology and language resources into translation practice. I am first author of the published study. The work joins software development with a question about autonomous learning.
30 translation undergraduates completed three stages: conventional tools, Sinolingo, then a new task after one month of autonomous use. Outputs were compared with BLEU, chrF++ and BERTScore.
The paper reports BLEU 21.4 → 28.9, chrF++ 48.7 → 56.2 and BERTScore 0.812 → 0.864 from the first to second task.
Read the published study ↗A small, single-institution study with sequential tasks and a short intervention. These are task-level findings, not proof of lasting learning for all students.
Useful data needs a usable way in.
I contributed as a team member to the political-news corpus and its platform, and am second author of the DEIT 2026 paper. The work connects data preparation with retrieval and classroom use.
Reported alignment accuracy: 96.5%. Mean post-test translation scores were 89.1 with corpus-supported practice and 82.4 in the comparison group.
The corresponding university innovation project completed with an Excellent result. The paper is published in ACM ICPS and indexed by EI Compendex.
Read the corpus paper ↗- 01
Collect
Political news from three sources
- 02
Translate & review
Machine translation + human editing
- 03
Align
Automatic matching + manual review
- 04
Retrieve & study
11,000 sentence pairs · teaching study
One cohort, a short intervention and political-news texts. The result does not establish gains across other domains or long-term retention.
More opportunities to speak.
As co-founder of Guangzhou Yanyun, I lead backend development and product direction for Yakky. We started from a familiar gap: exam-focused study offers limited speaking practice, while human tutoring can be expensive.
The app organises practice around scenarios and conversation partners. The early phone captures also show rough edges: repeated replies and a progress marker appearing inside the conversation. A connected speech pipeline alone was not enough to make practice feel natural.
Speech-model capability at the time was a constraint. This experience made capability, cost and the quality of practice part of the same product decision, before our work expanded into e-commerce with CreaTrades.
- 01
Choose a scenario
A context and a speaking partner
- 02
Speak
Microphone audio over WebSocket
- 03
Listen & respond
Speech recognition, dialogue, voice
- 04
Continue
Session context & conversation history
Original phone screenshots provided by me, November 2025. Process summarised from the existing application and speech-demo code; these document a prototype, not current service availability.
Work that became software.
I also contributed to the national DeepSeek-based Chinese traditional-culture translation project, completed with an Excellent result, and developed corpus/terminology resources for a GDUFS teaching-reform project.
Vietnamese–Chinese news corpus platform
Developer · joint rights holder
2025SR2489410Huaxia YiDianTong translation platform
Developer · joint rights holder
2025SR2526074Guyan Zhixue classical Chinese learning platform
Developer · joint rights holder
2026SR0548123Echoo simultaneous interpreting system
Developer · registered to GDUFS
2025SR2445119