In this paper, we propose a unified pre-training approach called UniSpeech to learn speech representations with both unlabeled and labeled data, in which supervised phonetic CTC learning and phonetically-aware contrastive self-supervised learning are conducted in a multi-task learning manner. The resultant representations can capture information more correlated with phonetic structures and improve the generalization across languages and domains. We evaluate the effectiveness of UniSpeech for cross-lingual representation learning on public CommonVoice corpus. The results show that UniSpeech outperforms self-supervised pretraining and supervised transfer learning for speech recognition by a maximum of 13.4% and 17.8% relative phone error rate reductions respectively (averaged over all testing languages). The transferability of UniSpeech is also demonstrated on a domain-shift speech recognition task, i.e., a relative word error rate reduction of 6% against the previous approach.
Commonsense reasoning requires a model to make presumptions about world events via language understanding. Many methods couple pre-trained language models with knowledge graphs in order to combine the merits in language modeling and entity-based relational learning. However, although a knowledge graph contains rich structural information, it lacks the context to provide a more precise understanding of the concepts and relations. This creates a gap when fusing knowledge graphs into language modeling, especially in the scenario of insufficient paired text-knowledge data. In this paper, we propose to utilize external entity description to provide contextual information for graph entities. For the CommonsenseQA task, our model first extracts concepts from the question and choice, and then finds a related triple between these concepts. Next, it retrieves the descriptions of these concepts from Wiktionary and feed them as additional input to a pre-trained language model, together with the triple. The resulting model can attain much more effective commonsense reasoning capability, achieving state-of-the-art results in the CommonsenseQA dataset with an accuracy of 80.7% (single model) and 83.3% (ensemble model) on the official leaderboard.
Cross-lingual Summarization (CLS) aims at producing a summary in the target language for an article in the source language. Traditional solutions employ a two-step approach, i.e. translate then summarize or summarize then translate. Recently, end-to-end models have achieved better results, but these approaches are mostly limited by their dependence on large-scale labeled data. We propose a solution based on mixed-lingual pre-training that leverages both cross-lingual tasks such as translation and monolingual tasks like masked language models. Thus, our model can leverage the massive monolingual data to enhance its modeling of language. Moreover, the architecture has no task-specific components, which saves memory and increases optimization efficiency. We show in experiments that this pre-training scheme can effectively boost the performance of cross-lingual summarization. In Neural Cross-Lingual Summarization (NCLS) dataset, our model achieves an improvement of 2.82 (English to Chinese) and 1.15 (Chinese to English) ROUGE-1 scores over state-of-the-art results.
Neural models have become successful at producing abstractive summaries that are human-readable and fluent. However, these models have two critical shortcomings: they often don't respect the facts that are either included in the source article or are known to humans as commonsense knowledge, and they don't produce coherent summaries when the source article is long. In this work, we propose a novel architecture that extends Transformer encoder-decoder architecture in order to improve on these shortcomings. First, we incorporate entity-level knowledge from the Wikidata knowledge graph into the encoder-decoder architecture. Injecting structural world knowledge from Wikidata helps our abstractive summarization model to be more fact-aware. Second, we utilize the ideas used in Transformer-XL language model in our proposed encoder-decoder architecture. This helps our model with producing coherent summaries even when the source article is long. We test our model on CNN/Daily Mail summarization dataset and show improvements on ROUGE scores over the baseline Transformer model. We also include model predictions for which our model accurately conveys the facts, while the baseline Transformer model doesn't.
With the abundance of automatic meeting transcripts, meeting summarization is of great interest to both participants and other parties. Traditional methods of summarizing meetings depend on complex multi-step pipelines that make joint optimization intractable. Meanwhile, there are a handful of deep neural models for text summarization and dialogue systems. However, the semantic structure and styles of meeting transcripts are quite different from articles and conversations. In this paper, we propose a novel end-to-end abstractive summary network that adapts to the meeting scenario. We design a role vector to depict the difference among speakers and a hierarchical structure to accommodate long meeting transcripts. Empirical results show that our model considerably outperforms previous approaches in both automatic metrics and human evaluation. For example, in the ICSI dataset, the ROUGE-1 score increases from 32.00% to 39.51%.
A commonly observed problem with abstractive summarization is the distortion or fabrication of factual information in the article. This inconsistency between summary and original text has led to various concerns over its applicability. In this paper, we firstly propose a Fact-Aware Summarization model, FASum, which extracts factual relations from the article and integrates this knowledge into the decoding process via neural graph computation. Then, we propose a Factual Corrector model, FC, that can modify abstractive summaries generated by any model to improve factual correctness. Empirical results show that FASum generates summaries with significantly higher factual correctness compared with state-of-the-art abstractive summarization systems, both under an independently trained factual correctness evaluator and human evaluation. And FC improves the factual correctness of summaries generated by various models via only modifying several entity tokens.
A commonly observed problem with abstractive summarization is the distortion or fabrication of factual information in the article. This inconsistency between summary and original text has led to various concerns over its applicability. In this paper, we propose to boost factual correctness of summaries via the fusion of knowledge, i.e. extracted factual relations from the article. We present a Fact-Aware Summarization model, FASum. In this model, the knowledge information can be organically integrated into the summary generation process via neural graph computation and effectively improves the factual correctness. Empirical results show that FASum generates summaries with significantly higher factual correctness compared with state-of-the-art abstractive summarization systems, both under an independently trained factual correctness evaluator and human evaluation. For example, in CNN/DailyMail dataset, FASum obtains 1.2% higher fact correctness scores than UniLM and 4.5% higher than BottomUp.
Lead bias is a common phenomenon in news summarization, where early parts of an article often contain the most salient information. While many algorithms exploit this fact in summary generation, it has a detrimental effect on teaching the model to discriminate and extract important information. We propose that the lead bias can be leveraged in a simple and effective way in our favor to pretrain abstractive news summarization models on large-scale unlabeled corpus: predicting the leading sentences using the rest of an article. Via careful data cleaning and filtering, our transformer-based pretrained model without any finetuning achieves remarkable results over various news summarization tasks. With further finetuning, our model outperforms many competitive baseline models. Human evaluations further show the effectiveness of our method.