JointUD

This is the codebase used in our submission to CoNLL Shared Task 2018 (codename ArmParser). It is a multitask neural model for part-of-speech tagging, morphological parsing and lemmatization․ The network is implemented in Keras 2.1.5.

This network is based on a sequence tagging network developed by UKPLab. Our main contribution is an additional character level lemma generator module which is attached to the sequence tagging network.

Dependency parsing is not yet implemented.

System Description Paper

The paper describing our system is published in Proceedings of the CoNLL 2018 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies. Here is a direct link to the PDF file.

@InProceedings{arakelyan-hambardzumyan-khachatrian:2018:K18-2,
  author    = {Arakelyan, Gor  and  Hambardzumyan, Karen  and  Khachatrian, Hrant},
  title     = {Towards {JointUD}: Part-of-speech Tagging and Lemmatization using Recurrent Neural Networks},
  booktitle = {Proceedings of the {CoNLL} 2018 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies},
  month     = {October},
  year      = {2018},
  address   = {Brussels, Belgium},
  publisher = {Association for Computational Linguistics},
  pages     = {180--186},
  url       = {http://www.aclweb.org/anthology/K18-2018}
}

Installation

Dependencies

• Keras==2.1.5
• Theano==0.9.0
• tensorflow==1.0.0
• nltk==3.2.2
• numpy==1.14.2
• sklearn

Keras backend

• Use Theano backend

Python version

• python3

Prerequisite

• Download pre-trained word vectors to wordvec/ directory.
• Put train.txt, test.txt, dev.txt files into data/{dataset_name}/ directory.
• Write the language alphabet to {dataset_name}.chars file and save it in characters/ directory.

Usage

Training

usage: train.py [-h] [--model_name MODEL_NAME] [--nb_epochs NB_EPOCHS]
                [--early_stopping EARLY_STOPPING] [--batch_size BATCH_SIZE]
                [--optimizer OPTIMIZER] [--classifier CLASSIFIER]
                [--loss LOSS] [--dump_prefix DUMP_PREFIX] [--lstm LSTM]
                [--lstm_size LSTM_SIZE] [--dropout DROPOUT]

optional arguments:
  -h, --help            show this help message and exit
  --model_name MODEL_NAME
                        Model name
  --nb_epochs NB_EPOCHS
                        Number of Epochs
  --early_stopping EARLY_STOPPING
                        Early Stopping
  --batch_size BATCH_SIZE
                        Batch Size
  --optimizer OPTIMIZER
                        Optimizer
  --classifier CLASSIFIER
                        Classifier
  --loss LOSS           Loss
  --dump_prefix DUMP_PREFIX
                        Model dump name prefix
  --lstm LSTM
  --lstm_size LSTM_SIZE
  --dropout DROPOUT

Parsing

usage: parse.py [-h] [--model_path MODEL_PATH] [--mapping_path MAPPING_PATH]
                [--input_path INPUT_PATH] [--output_path OUTPUT_PATH]

optional arguments:
  -h, --help            show this help message and exit
  --model_path MODEL_PATH
                        Model path
  --mapping_path MAPPING_PATH
                        Mapping path
  --input_path INPUT_PATH
                        Input path
  --output_path OUTPUT_PATH
                        Output path

Example

• Input

I have been going there since I was a little boy and love the friendly and relaxing atmosphere.

• Output

1	I	I	PRON	_	Case=Nom|Number=Sing|Person=1|PronType=Prs	_	_	_	_
2	have	have	VERB	_	Mood=Ind|Tense=Pres|VerbForm=Fin	_	_	_	_
3	been	be	AUX	_	Tense=Past|VerbForm=Part	_	_	_	_
4	going	go	VERB	_	VerbForm=Ger	_	_	_	_
5	there	there	ADV	_	PronType=Dem	_	_	_	_
6	since	since	ADP	_	_	_	_	_	_
7	I	I	PRON	_	Case=Nom|Number=Sing|Person=1|PronType=Prs	_	_	_	_
8	was	be	AUX	_	Mood=Ind|Number=Sing|Person=3|Tense=Past|VerbForm=Fin	_	_	_	_
9	a	a	DET	_	Definite=Ind|PronType=Art	_	_	_	_
10	little	little	ADJ	_	Degree=Pos	_	_	_	_
11	boy	boy	NOUN	_	Number=Sing	_	_	_	_
12	and	and	CCONJ	_	_	_	_	_	_
13	love	love	VERB	_	VerbForm=Fin	_	_	_	_
14	the	the	DET	_	Definite=Def|PronType=Art	_	_	_	_
15	friendly	friendly	ADJ	_	Degree=Pos	_	_	_	_
16	and	and	CCONJ	_	_	_	_	_	_
17	relaxing	relax	VERB	_	VerbForm=Ger	_	_	_	_
18	atmosphere	atmosphere	NOUN	_	Number=Sing	_	_	_	_
19	.	.	PUNCT	_	_	_	_	_	_