OS environment

Ubuntu 16.04.4 LTS (GNU/Linux 4.4.0-130-generic x86_64)

Prerequisites

Download Multinli 0.9

wget https://www.nyu.edu/projects/bowman/multinli/multinli_0.9.zip

Download Stanford Corenlp

wget http://nlp.stanford.edu/software/stanford-corenlp-full-2018-02-27.zip

Download Java 1.8+

(Check with command: java -version) (Download Page)

Requirement

tensorflow == 1.10
tqdm
stanfordcorenlp
nltk
jsonlines

Execute

modify an absolute path for stanford corenlp in ./src/DependencyParsing.py

        self.nlp = StanfordCoreNLP('/home/edlin0249/iis_summer_intern/stanford-corenlp-full-2018-02-27') # modify your absolute path for stanford corenlp

run

python3 train.py -batch_size 50 -display_interval 1000

Experiment

baseline

train

Acc = None , Cross Entropy Loss = None

dev

Acc = None , Cross Entropy Loss = None

test

Acc = None , Cross Entropy Loss = None

w/ dependency parsing

train

Acc = None , Cross Entropy Loss = None

dev

Acc = None , Cross Entropy Loss = None

test

Acc = None , Cross Entropy Loss = None

w/ attention

train

Acc = None , Cross Entropy Loss = None

dev

Acc = None , Cross Entropy Loss = None

test

Acc = None , Cross Entropy Loss = None

References

• Improved Neural Machine Translation with Source Syntax [paper]
• Enhanced LSTM for Natural Language Inference [paper]