Last updated: November 2, 2026
A new meta-analysis pooled the evidence on chatbots learners can speak with and found moderate gains, Cornell University will give staff and students accounts with four AI vendors instead of one, and a Shanghai company’s AI language teaching system won a conference award without publishing any outcome data. On the tool side, Google’s Live Translate now works offline on recent Pixel phones, and Pearson has an official AI practice app for the PTE that earlier roundups missed. Coverage of DeepL, translation earbuds and ImpactBench appeared in the October 17 roundup.
Industry News
Cornell Gives Staff and Students a Choice of AI Vendors. Cornell University announced on October 26 that faculty, staff and doctoral students will get free base-level accounts for ChatGPT, Claude, Microsoft Copilot Chat and, tentatively, Google Gemini starting in November, with master’s and undergraduate students following in early 2027, according to the Cornell Chronicle. Faculty and staff must finish a required orientation on responsible AI use first, and the university says data entered through these accounts will not be used to train AI models. The arrangement is a two-year pilot, after which Cornell plans to review how the tools were used. The multi-vendor approach is the notable part: most universities that license AI pick a single provider, which ties every course to one company’s tools and pricing. For language departments, access to several systems also makes side-by-side comparison of AI feedback on student writing a realistic classroom activity rather than a paid extra.
Shanghai AI Language Teaching System Wins Award, Shares No Results. NetClass Technology, a Nasdaq-listed education software company with offices in Shanghai, Hong Kong, Singapore and Tokyo, says its multi-agent system for assessing foreign language learners received an Outstanding Case Award at the 2026 Digital Lifelong Learning International Conference, held October 16 and 17 at Shanghai Open University with UNESCO institutes as co-organizers, per the company’s press release. The system splits assessment among separate AI agents for pronunciation, vocabulary, grammar, discourse and pragmatics, and sends open-ended speaking and writing to a human instructor for review when the automated scores disagree. The company reports use at nearly 100 institutions and by more than 40,000 students. The release contains no data on whether learners improve, so the award recognizes the design rather than demonstrated results. The human-review step is the design choice worth noting, since it addresses the reliability problem that makes fully automated speaking scores hard to trust.
Tool Updates
Google Live Translate Works Offline on Recent Pixel Phones. Live Translate conversations on Pixel 9, 10 and 11 phones now run without an internet connection between English and ten languages, including Japanese, Hindi, Indonesian and Russian, as reported by BGR in mid-September. Language packs have to be downloaded in advance, and offline mode does not translate directly between two non-English languages, so a Spanish speaker talking with a Japanese speaker still needs a connection. The online version supports more than 70 languages, which shows how much smaller on-device models still are. Korean and Chinese are not among the ten offline languages, a gap for learners and travelers in East Asia. For hardware alternatives, the October 17 roundup covered Timekettle’s new open-ear earbuds.
Pearson Launches an Official AI Practice App for the PTE. Pearson released Official PTE AI Practice on September 1, offering full mock tests, section tests and individual questions for PTE Academic and PTE Core, each with instant AI scoring, feedback and an AI tutor, per Pearson’s announcement. The first two diagnostic tests are free, and further practice requires a paid plan whose prices the release does not list. The launch follows ETS’s TOEIC Pal app, covered in the September 16 roundup, so two of the major test owners now sell AI-scored practice for their own exams. Because PTE is itself scored by machine, practice scores from the test owner may track the real test more closely than third-party apps can, though Pearson publishes no data comparing them. For free practice options, see this site’s TOEIC and IELTS prompts.
Research
Meta-Analysis Finds Speaking Chatbots Help L2 Learners, With Caveats. A meta-analysis in the International Review of Applied Linguistics in Language Teaching pooled 19 studies with 2,643 participants that used LLM-based chatbots learners could speak with, and found a moderate-to-large average effect on second language learning outcomes (Hedges’ g = 0.68) and a similar effect on anxiety, enjoyment, motivation and willingness to communicate (g = 0.64), per the published article by Eunjin Lee, Hansol Lee and Jang Ho Lee. Learner proficiency, educational level, interaction format and context of use all shaped how well the chatbots worked. The authors list real limits: the sample of studies is small, some subgroups rest on only a few studies, and effect sizes were calculated from post-test scores rather than gains. Nine of the individual samples showed effects that could not be distinguished from zero. The review only covers peer-reviewed English-language studies from 2022 onward, so the field is young and the average hides wide variation. For more research on technology in language learning, see this site’s CALL reading list.
Duolingo reports its third-quarter results on November 3, which the next roundup will cover.
AI-assisted, human-edited