Fork of code from "Stacking Dependency Parsers", Martins et al., EMNLP 2008, as released as part of SEMAFOR 2.1. Please cite the paper (bib) if you use this software in your research.

I'm using this forked version in SEMAFOR > 3.0. SEMAFOR's accuracy is consistently better (by about 1 pt F1) when using StackedMstParser than when using MaltParser or 3rd-order TurboParser. I didn't write most of this code; I just made it faster, the models smaller, and the output CoNLL compliant. I'm not planning on providing much maintainence or support. Feel free to open tickets, but no promises.

./scripts/compile.sh

./scripts/runServer.sh <model-file> [<port>]

A pretrained model (trained on the Wall Street Journal section of Penn Treebank) is available here.

<port> defaults to 12345

./scripts/runClient.sh <input-file> <output-file> [<server-hostname> [<port>]]

<server-hostname> defaults to localhost

<port> defaults to 12345

I haven't tried to retrain the model. If you want to retrain the model, you're on your own, but this might work (probably not):

./scripts/trainTestEval.sh