Skip to content

Latest commit

 

History

History
53 lines (44 loc) · 2.53 KB

README.md

File metadata and controls

53 lines (44 loc) · 2.53 KB
Raw

Segment-Based IMT

Segment-based Interactive Machine Translation implementation based on Moses XML scheme.

Requirements

Moses and mgiza are needed in order to run this software. Alternatively, you can run it through Docker.

Usage

Alignments generation

This software requires of an alignment model. Therefore, prior to its use, alignments must be trained by doing:

tools/alignments.sh -s src_file -t tgt_file -o output_file -m mgiza_bin {options}

options:
     -b: use IBM Model 1. (Default: Use HMM.)

where src_file and tgt_file are the source and target of the training dataset; output_file is the file in which to store the alignments (which will be required for using the segment-based software); mgiza_bin is the path to mgiza's bin folder (e.g., /opt/moses/mgiza/mgizapp/bin); and the -b flags switches from using Hidden Markov Models to using IBM Model 1.

User simulation

You can simulate a user working on a segment-based IMT framework by doing:

simulation.py [-h] -s source_file -r reference_file -c moses_ini -a
                     alignments_file [-v] [-x] [-p threshold] [-m moses_bin]
                     -l log_file

optional arguments:
  -h, --help            show this help message and exit
  -s source_file, --sources source_file
                        file containing the source segments.
  -r reference_file, --references reference_file
                        file containing the reference segments.
  -c moses_ini, --config moses_ini
                        file containing moses configuration.
  -a alignments_file, --alignments alignments_file
                        file containing the alignments generated by alignments.sh
  -v, --verbose         Activate verbose mode.
  -x, --xml             Show XML markup.
  -p threshold, --probability threshold
                        probability threshold. (Default 0.)
  -m moses_bin, --moses moses_bin
                        Path to moses bin. (Default: /opt/moses/bin/moses.)
  -l log_file, --log log_file
                        File to store the log.

Run through Docker

If you want to run this software through Docker you can have a look at this repo.

Citation

On using this software, please cite the following paper:

Miguel Domingo and Álvaro Peris and Francisco Casacuberta. Segment-Based Interactive-Predictive Machine Translation. Machine Translation Journal, 31:163–185, 2017