• Estonian resource grammar (Inari, Kaarel)
• Hands-free "always on" offline HTML5-based multilingual kitchen timer
• PGF-based keyboard for Android
• Linking CMU Pocketsphinx with the GF C runtime
• Collection of GF error messages

• Estonian
  • spoken in Estonia by ~1 million speakers
  • similar to Finnish
• work started at GF Summer School 2011
  • modifying the Finnish RG
  • progress has been slow
• now
  • started implementing the noun morphology based on "Heiki-Jaan Kaalep. Eesti käänamissüsteemi seaduspärasused"
• project pages:
  • Estonian RG: https://github.com/GF-Estonian/GF-Estonian
  • PhrasebookEst: https://github.com/Kaljurand/PhrasebookEst

Hands-free "always on" offline HTML5-based multilingual kitchen timer

• speech recognition: pocketsphinx.js
  • CMU Pocketsphinx automatically converted from C to Javascript
  • comes with English and French acoustic models, also works with Estonian models
  • uses finite-state language models
  • modern web browser (currently Google Chrome only)
  • demo
• GF grammar
  • time expressions, e.g. Alarm (currently in Estonian and English)
  • dialogue/commands: start, stop, pause, etc.
  • converted to finite state automata
• GF translator to map the raw recognition result to timer commands
  • ideally Javascript-only
  • possibly/temporarily Javascript to GF server on localhost
• some jQuery timer plugin to execute the timer commands

• current (Android) keyboards offer prediction, flow/slide, etc.
  • using an n-gram language model extracted from your email / Facebook posts
  • ignoring the application model (e.g. list view on place names)
  • e.g. SwiftKey, Minuum Keyboard
• project: implement a PGF-based keyboard app
  • implements the Android IME API, i.e. it can be used in all Android apps
  • completes based on a given PGF
  • novel UI, e.g. disable keys that do not lead to completion
  • based on JPGF, or the C runtime of GF (+ Java bindings)
• possible application: offline place names lookup+geocoding, like in car GPSs

• Pocketsphinx (speech recognition engine) currently supports n-gram, JSGF, and FSG language models
• project: add support to GF grammars by linking in the GF C-based runtime
• benefit: context-sensitive speech recognition

• list of all possible error messages
  • expected: X, inferred: Y
  • cannot infer type of constant ...
  • circular definitions
  • ...
• documentation of each
• most likely causes of each
• code that results in the error message
• community edited