Lattice-Based Recurrent Neural Network Encoders for Neural Machine Translation

Authors: Jinsong Su, Zhixing Tan, Deyi Xiong, Rongrong Ji, Xiaodong Shi, Yang Liu

AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiment results on Chinese-English translation demonstrate the superiorities of the proposed encoders over the conventional encoder.
Researcher Affiliation Academia Xiamen University, Xiamen, China1 Soochow University, Suzhou, China2 Tsinghua University, Beijing, China3
Pseudocode No The paper presents mathematical equations for the GRU and its variants, along with architectural diagrams, but does not include any formal pseudocode or algorithm blocks.
Open Source Code No The paper mentions 'the toolkit2 released by Stanford to train word segmenters' and provides a URL (http://nlp.stanford.edu/software/segmenter.html#Download). However, this refers to a third-party tool used by the authors, not the open-sourcing of their own proposed methodology's code.
Open Datasets Yes Our training data consists of 1.25M sentence pairs extracted from LDC2002E18, LDC2003E07, LDC2003E14, Hansards portion of LDC2004T07, LDC2004T08 and LDC2005T06, with 27.9M Chinese words and 34.5M English words.
Dataset Splits Yes We chosed the NIST 2005 dataset as the validation set and the NIST 2002, 2003, 2004, 2006, and 2008 datasets as test sets.
Hardware Specification Yes We used a single GPU device Titan X to train models.
Software Dependencies No The paper mentions 'Rmsprop (Graves 2013)' and 'multi-bleu.perl script', and 'the toolkit2 released by Stanford' but does not specify version numbers for any software components, libraries, or dependencies used for the experiments.
Experiment Setup Yes During this procedure, we set the following hyper-parameters: word embedding dimension as 320, hidden layer size as 512, learning rate as 5 10 4, batch size as 80, gradient norm as 1.0.