Machine Translation into Learning
summary
Funded two‑year project builds a proof‑of‑concept iCALL system that combines semantic-based machine translation with deep handcrafted syntactic grammars (English Resource Grammar, Chinese grammar) and minimal recursion semantics to diagnose Mandarin L2 errors. System adds controlled “mildly ungrammatical” rules, narrows lexicons to curriculum vocab, and when parses are ambiguous it prompts the learner in their native language instead of guessing (current automatic guesses are ~60–70% accurate). Output is web‑based, open source, tied to NTU course materials and a growing learner corpus.
AI-written summary · about
chapters (18)
- 0:00Start
- 0:51Self Introduction
- 1:42What about Language Learning and Technology?
- 2:42Background: Grammars in On-line Education
- 3:42An example from Grade 5
- 4:22There are a few possible answers
- 4:56Because language is infinite... remember?
- 5:16How to deal with this?
- 7:12Sample Errors
- 7:27HPSG: Head-driven Phrase Structure Grammar
- 8:40Changes to the general-purpose grammar
- 9:43Does this really work?
- 13:59Grammar Feedback Through Translation
- 15:46The Power of Semantic Disambiguation
- 16:53An Intelligent Language Tutor
- 17:34Our Focus: Chinese Learners (English natives)
- 18:32Future Work
- 20:16Summary
related talks
33mTranslation Augmented Generation - Breaking Language Barriers in RAG...
01h 00mHuman and machine inductive biases for compositional linguistic generalization
18mAI translation + AI agents = i18n made easy
21mThe indigenous languages technology (ILT) project at NRC in 2021
32mSetting up a machine translation service for Timor-Leste
32mLinguistic, domain and task adaptation of LLMs using the PLLuM models as an example
44mSome tentative first steps towards a Star Trek universal communicator
12mLanguage-Model Probabilistic Programming for Improving Autoformalization via Cycle(…)