With the permission of the authors, this code is an adaptation of code from https://github.com/TurkuNLP/smt-pronouns, for their submission to the 2016 shared cross-lingual pronoun prediction shared task at WMT

• Original authors: Juhani Luotolahti, Jenna Kanerva and Filip Ginter
• Date: May 2016
• Date of copy: 28/03/2017

Cf. the article presenting their system description: Juhani Luotolahti, Jenna Kanerva and Filip Ginter. 2016. Cross-Lingual Pronoun Prediction with Deep Recurrent Neural Networks. In Proceedings of the First Conference on Machine Translation. pp. 596–601. Berlin, Germany

• HM: Train 2 dep parsing models for each language out of EN, FR, DE and ES, namely one form model and one lemma model. Note these models also require a POS model.

• HM: Note that the TARGET side has REPLACE_n tokens that are always subject pronouns. One must give them an automatic PRONOUN tag before parsing, as well as a special replace_label label to keep track. Normally they can be treated as subjects.

• HM: Generate a file, aligned to the input. It will contain four columms: SOURCE_form_depidx SOURCE_form_label TARGET_lemma_depidx TARGET_lemma_label

• HM: Note that the TARGET side has REPLACE_n tokens that are always subject pronouns. One must give them an automatic PRONOUN tag before parsing, as well as a special replace_label label to keep track. Normally they can be treated as subjects.

• RB: put data on cluster

• RB: test keras on cluster

• RB: get morphological information from lexica for both source and target sentences. Getting the morph info for target sentences requires mapping the PoS provided to the PoS in the lexicon.

• RB: finish adapting code to take generic features