Tree-State Based Rule Selection Models for Hierarchical Phrase-Based Machine Translation

Authors: Shujian Huang, Huifeng Sun, Chengqi Zhao, Jinsong Su, Xin-Yu Dai, Jiajun Chen

IJCAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results demonstrate that these simple models could bring significant improvements to the translation quality.
Researcher Affiliation Academia 1State Key Laboratory for Novel Software Technology, Nanjing University, P. R. China 2Xiamen University, Xiamen, P. R. China
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper mentions using public tools like 'Berkeley parser', 'ICTCLAS', 'maximum entropy model' (with a GitHub link to a tool), and 'neural network based models' (with a CodePlex link to a tool), but does not state that the authors' own implementation code for the methodology described is open-source or provide a link to it.
Open Datasets Yes The translation model (TM) of the system is trained on parallel sentences from LDC, including LDC2002E18, LDC2003E14, LDC2004E12, LDC2004T08, LDC2005T10, LDC2007T09, which consists of 8.3 million of sentence pairs. We train a language model (LM) from monolingual data which includes Xinhua portion of Gigaword corpus. ... Penn Treebank1. 1http://www.cis.upenn.edu/ treebank/
Dataset Splits Yes We use multi-reference data MT03 as the development data (dev), MT04 and MT05 as test data. These data are mainly in the same genre, avoiding the extra consideration of domain adaptation. ... Data Usage ... MT03 dev 919
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, or memory amounts) used for running its experiments.
Software Dependencies No The paper mentions using 'Berkeley parser' [Petrov and Klein, 2007] and 'ICTCLAS' (http://ictclas.nlpir.org/), and open-source tools for models: 'maximum entropy model' (https://github.com/lzhang10/maxent) and 'neural network based models' (https://cntk.codeplex.com/). However, it does not specify version numbers for these software components.
Experiment Setup Yes We set the beam size to 20. We train a 5-gram language model on the monolingual data with MKN smoothing. ... For the training of maximum entropy model, we filter features that occur less than 20 times. We set the maximum iteration number of the L-BFGS algorithm to 150. ... We set the number of hidden nodes in first hidden layer to be 512, and the number of hidden nodes in the second hidden layer to be 128. We perform mini-batched back-propagation training for 150 iterations, with a mini-batch size 1024 and a dropout rate 0.4.