Software

Please feel free to contact me about the code/data used in our publications. Some may not be publicly available (sorry!) but we can provide them for the purpose of scientific reproduction.

Fine-tuned Models

🤔 Japanese NLI model
- a fine-tuned masked language model using JGLUE-JNLI and JSICK

Python Scripts

🧋 Sonish: Python-ish syntax sugar added to C++
- generates readable C++ snippet from restricted Python code
🐘 Pic2Braille: a picture to text converter written in Python
- simulates a roughly colored 160x100 dot display using braille characters and ANSI escape code

⚖️ MW-Classify: a discriminative preordering implementation
- official implementation of [Hoshino et al. ACL-IJCNLP2015]
An implementation of head finalization [Isozaki et al. WMT2010; Isozaki et al. TALLIP2012]
- very old script used in our replication studies
An implementation of Kendall’s tau [Kendall 1938]
- used in the replication study of [Isozaki et al. WMT2010] but modified a bit from their paper
An implementation of BLEU [Papineni et al. 2002]
- reproduces the single-reference mode of multi-bleu.perl
An implementation of IBM model 1 [Brown et al. 1993]
- naive (no semi-rings) and unfinished
An implementation of IBM model 2 [Brown et al. 1993]
- somewhat incorrect and unfinished

Links to My Favorites

An implementation of [Hoshino et al. ACL-IJCNLP2015] by Yusuke Oda
- Please note that his feature function is slightly modified from our paper
Ckylark: a Japanese PCFG-LA parser by Yusuke Oda
Haruniwa: a Japanese PCFG-LA model for Berkeley parser by Alastair Butler
Jigg: a converter that can handle KNP and SynCha outputs in XML by Hiroshi Noji
Modifications to BerkeleyParser (missing URL) by NICT
An implementation of Nakagawa ACL-IJCNLP2015 by Tetsuji Nakagawa
Travatar: a forest-to-string machine translation system by Graham Neubig
- accompanied by user friendly automatic evaluation interface