Graph Based Translation Memory for Neural Machine Translation

Authors: Mengzhou Xia, Guoping Huang, Lemao Liu, Shuming Shi7297-7304

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on three language pairs show that the proposed approach is efficient in terms of running time and space occupation, and particularly it outperforms multiple strong baselines in terms of BLEU scores.
Researcher Affiliation Collaboration Mengzhou Xia Carnegie Mellon University mengzhox@andrew.cmu.edu Guoping Huang Tencent AI Lab donkeyhuang@tencent.com Lemao Liu Tencent AI Lab redmondliu@tencent.com Shuming Shi Tencent AI Lab shumingshi@tencent.com
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any statement or link for open-source code for the described methodology.
Open Datasets Yes Following the previous works investigating on incorporating TM into NMT models, we use the JRC-Acquis corpus for training and evaluating our proposed model. We manage to obtain preprocessed datasets from Gu et al. (2018).
Dataset Splits Yes For each language pair, we randomly select 3000 samples to form a development and a test set respectively. The rest of the pairs are used as the training set.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory) used for running its experiments.
Software Dependencies No The paper mentions 'Tensor2Tensor(Vaswani et al. 2018) package' but does not provide specific version numbers for software dependencies.
Experiment Setup Yes For training all systems, we maintain the same hyperparameters as shown in Table 1 for comparison. We set the warm-up step to be 5 epochs and we early stop the model after training 20 epochs. Table 1: Word embedding 512 Layers 6 TM dropout 0.6 Other dropout 0.1 Beam size 5 Label smoothing 0.1 Batch size (tokens) 8192