Molecular graph generation is a fundamental but challenging task in various applications such as drug discovery and material science, which requires generating valid molecules with desired properties. Auto-regressive models, which usually construct g
raphs following sequential actions of adding nodes and edges at the atom-level, have made rapid progress in recent years. However, these atom-level models ignore high-frequency subgraphs that not only capture the regularities of atomic combination in molecules but also are often related to desired chemical properties. In this paper, we propose a method to automatically discover such common substructures, which we call {em graph pieces}, from given molecular graphs. Based on graph pieces, we leverage a variational autoencoder to generate molecules in two phases: piece-level graph generation followed by bond completion. Experiments show that our graph piece variational autoencoder achieves better performance over state-of-the-art baselines on property optimization and constrained property optimization tasks with higher computational efficiency.
Recent years have witnessed the rapid advance in neural machine translation (NMT), the core of which lies in the encoder-decoder architecture. Inspired by the recent progress of large-scale pre-trained language models on machine translation in a limi
ted scenario, we firstly demonstrate that a single language model (LM4MT) can achieve comparable performance with strong encoder-decoder NMT models on standard machine translation benchmarks, using the same training data and similar amount of model parameters. LM4MT can also easily utilize source-side texts as additional supervision. Though modeling the source- and target-language texts with the same mechanism, LM4MT can provide unified representations for both source and target sentences, which can better transfer knowledge across languages. Extensive experiments on pivot-based and zero-shot translation tasks show that LM4MT can outperform the encoder-decoder NMT model by a large margin.
Language coverage bias, which indicates the content-dependent differences between sentence pairs originating from the source and target languages, is important for neural machine translation (NMT) because the target-original training data is not well
exploited in current practice. By carefully designing experiments, we provide comprehensive analyses of the language coverage bias in the training data, and find that using only the source-original data achieves comparable performance with using full training data. Based on these observations, we further propose two simple and effective approaches to alleviate the language coverage bias problem through explicitly distinguishing between the source- and target-original training data, which consistently improve the performance over strong baselines on six WMT20 translation tasks. Complementary to the translationese effect, language coverage bias provides another explanation for the performance drop caused by back-translation. We also apply our approach to both back- and forward-translation and find that mitigating the language coverage bias can improve the performance of both the two representative data augmentation methods and their tagged variants.
System combination is an important technique for combining the hypotheses of different machine translation systems to improve translation performance. Although early statistical approaches to system combination have been proven effective in analyzing
the consensus between hypotheses, they suffer from the error propagation problem due to the use of pipelines. While this problem has been alleviated by end-to-end training of multi-source sequence-to-sequence models recently, these neural models do not explicitly analyze the relations between hypotheses and fail to capture their agreement because the attention to a word in a hypothesis is calculated independently, ignoring the fact that the word might occur in multiple hypotheses. In this work, we propose an approach to modeling voting for system combination in machine translation. The basic idea is to enable words in hypotheses from different systems to vote on words that are representative and should get involved in the generation process. This can be done by quantifying the influence of each voter and its preference for each candidate. Our approach combines the advantages of statistical and neural methods since it can not only analyze the relations between hypotheses but also allow for end-to-end training. Experiments show that our approach is capable of better taking advantage of the consensus between hypotheses and achieves significant improvements over state-of-the-art baselines on Chinese-English and English-German machine translation tasks.
