Open Source Code
Machine Translation
Each source tarball contains a README with compilation instructions. Code I wrote is under the LGPL and all dependencies are open source.- KenLM: Language model inference
- Does one thing well: load a language model and return probabilities. Uses less RAM and CPU than other language model packages.
- Language model filter
- Fast filtering of language models to multiple vocabularies. Yields a 92% reduction in model size for system combination and 36% for phrase-based translation.
- MEMT: System combination
- My research multi-engine machine translation system. Known in various evaluations as cmu-combo, cmu-heafield-combo, and cmu-statxfer.
- Scoring
- Script that makes it easy to score machine translation output using NIST's BLEU and NIST, TER, and METEOR. Use plaintext with one segment per line instead of three different formats. Puts all the scores on a single line ready for inclusion in a table. Updated November 30, 2011.
Fun with C++
In addition to this code, I have quiz on C++ corner cases.- producer consumer.h
- Exception safe producer consumer class supporting multiple readers and writers. Uses Boost.
- underhanded.c
- For the 2008 underhanded C competition. The goal is to appear to properly redact a PPM file while leaking part of it.
- prime time.c
- Looking for a prime time? Call 5373737.
- text twist cheat.tar.gz
- Lists words for use in text twist.