AlignMan

This is a tool for manual word alignment of parallel sentences.

Requirements

SpaCy
tokenizer

Usage

In order to do the word alignments a database, containing the sentences to align, has to be created. Then a graphical tool can be used to do the alignments, and finally the alignments can be exported from the database in various formats.

Create new database

python3 primeDB.py --input-file examples/sentences.txt

The input file should be a list of aligned sentences, separated by a tab. The sentences should ideally be tokenized.:

What can I order for you ?    Hvað get ég pantað handa ykkur ?

I 'll have one of those .    Ég þigg einn svoleiðis .

What are you worrying about ?    Hvaða áhyggjur eru þetta ?

An example file is included, examples/sentences.txt.

It is also possible to use untokenized sentences by adding the flag --tokenize. Currently it only supports English and Icelandic. It assumes the first sentence is in English and uses SpaCy to tokenize it, and that the latter sentence is in Icelandic and uses tokenizer from Miðeind to tokenize that.

--db-name dbname.db changes the name of the output database from the default alignments.db.

Manual Word Alignment

python3 align.py

The tool can be run for one user by running the file without any parameters. By using the --user parameter a second user can be selected. The users then align the sentences separately and when both have finished a sentence the alignments are rated as sure or possible. All alignments that both evaluators set as 1-to-1 are tagged sure. Other alignments, the ones that only one evaluator creates or if one or both create 1-to-many, many-to-1 or many-to-many, they are tagged as possible.

Export manual alignments

python3 export_alignments.py --alignments

It is possible to export the alignments in two different formats by adding the parameter --alignment-format:

Examples of both export formats are available in the examples folder.

This script can also export the sentences that were aligned by using the --sentences flag. For other options, run python3 export_alignments.py -h.

Citation

If you use AlignMan for published research, please cite the paper:

@inproceedings{combalign-nodalida2021,
  author    = {Steingrímsson, Steinþór  and  Loftsson, Hrafn  and  Way, Andy},
  title     = {CombAlign: a Tool for Obtaining High-Quality Word Alignments},
  booktitle = {Proceedings of the 23rd Nordic Conference on Computational Linguistics},
  month     = {June},
  year      = {2021},
  address   = {Online},
  publisher = {Link{\"o}ping University Electronic Press},
}

License

Licensed under the terms of the Apache License, version 2.0. A full copy of the license can be found in LICENSE.

lizezhonglaile/AlignMan