Sha Liu (@ShaLiu2010) header
Sha Liu (@ShaLiu2010) profile picture

PhD in Language Assessment @BristolUni|Automated speaking/writing assessment, feedback; feedback literacy; eye-tracking

Bristol, England
Joined November 2017
Photo by Sha Liu (@ShaLiu2010)Photo by Sha Liu (@ShaLiu2010)Photo by Sha Liu (@ShaLiu2010)Photo by Sha Liu (@ShaLiu2010)Photo by Sha Liu (@ShaLiu2010)Photo by Sha Liu (@ShaLiu2010)Photo by Sha Liu (@ShaLiu2010)Photo by Sha Liu (@ShaLiu2010)Photo by Sha Liu (@ShaLiu2010)Photo by Sha Liu (@ShaLiu2010)
Submissions for the second annual Yale AI Social Science Methods conference (to be held April 28–29) close Monday November 30, 2026. All are welcome to apply: web.cvent.com/event/3942a257… I'm on the organizing committee as the political science rep. We look forward to seeing your work!
7
37
155
12,739
What is your impression of AI chatbots in speaking assessment? "(Too) robotic, flat, lack of emotion…"? In late September, ElevenLabs brought Eleven v4 Turbo, which it calls its "most emotive" text-to-speech model, to live conversations. I tried it in a speaking task.
8
1
53
3) A real complaint call often carries frustration. Should a speaking task bring it out in the learner, or keep it out of what is assessed? 7/9 --- 4) What, if anything, should automatic sentiment and frustration scores on a learner's turns be used for?
7
1) The model asks the voice to sound "regretful" or "warmly". How well does the voice carry this out, and how could we judge that? 6/9 --- 2) Does an expressive AI interlocutor bring the task closer to real life, or add something it was not designed to elicit?
3
Although these are observations from a few calls on one task, I believe that Eleven v4 Turbo and similar models give us a chance to explore how to make AI chatbots more authentic. My trial left me with four questions:
4
Were these emotive features inherent to to the synthetic voice? My own earlier call with the same lines got almost the same scores (6 of 9 identical, the rest within 0.2).
4
The platform also scored the guest's turns for sentiment and frustration. Early on, when she says the honeymoon mood has been destroyed: "Very negative" sentiment, "Moderate" frustration. About five minutes later, after she accepted the compensation offered: "Very positive".
5
The language model writing the manager's words added directions for the voice, such as "regretful", "sincere", "measured" and "warmly" (in square brackets in the video). Not too surprisingly, I think I did hear them in his voice.
6
The video shows two clips of a hotel complaint role-play speaking task I designed. A learner plays a guest whose honeymoon evening has gone wrong, and an AI interlocutor, as the general manager, calls to make amends. Both voices run on Eleven v4 Turbo.
9
🔷Pleased to share our new conference paper "Developing a multimodal chatbot for speaking assessment: Integrating LLMs, eye tracking, and neuroimaging." We integrate GPT-Realtime, eye tracking, and fNIRS to examine speaking & neurocognitive processes. 📄 researchgate.net/publication…
2
8
294
🔷Journal of Second Language Writing Special Issue: GenAI-Assisted L2 Writing as a Process Editors: Vahid Aryadoust & Andrea Revesz 🔷Focus: empirical studies on the cognitive processes underlying human-AI interaction. 🔷Abstract deadline: 30Nov 2026 forms.cloud.microsoft/pages/…
5
12
753
Sha Liu retweeted
"LLM-assisted screening can substantially reduce the time and cost of review preparation" Henninger et al. (in press). Automating Abstract Screening in Research Synthesis Using Large Language Models: A Tutorial and Proof-of-Concept Study doi.org/10.1177/251524592614…
2
9
511
Sha Liu retweeted
New article: Gayed, J. M. (2026). Evaluating open-weight LLMs for scoring L2 essays: Accuracy, prompt transfer, and deployment of AiAWE. Research Methods in Applied Linguistics, 5(3), 100390: sciencedirect.com/science/ar…
2
4
357
Sha Liu retweeted
📣 Call for Proposals | #LTRC2027 Ghent, Belgium, June 26–30, 2027! 🌐 Theme: Language Assessment in Interdisciplinary Research We welcome papers, demos, posters, WIPs & symposia. 📅 Deadlines: Symposia: Oct. 7, 2026 All other formats: Oct. 23, 2026 11:59 p.m. (AoE)
2
1
3
158
In the AI age, citation checking is Pandora's box containing fabricated references, Frankenstein citations, retractions, superseded preprints... So I built an open-source check to help: github.com/Shalice-in-AIland…
4
112
Sha Liu retweeted
"The purpose of this brief editorial is to announce the launch of two new article types . . .: Data Reports and Registered Reports." Plonsky & McManus (in press). SSLA expands its commitment to open science: Introducing Data Reports and Registered Reports doi.org/10.1017/S02722631261…
1
4
12
853
Our Cambridge Element is now published & available Open Access 🎉 Generative Artificial Intelligence and English-Medium Education examines what GenAI means for teaching, assessment, learner autonomy and agency in EME. #AIinEducation #EMI #HE #AI #GenAI tinyurl.com/AIEMIElement
7
14
1,297