Modern Automatic Speech Recognition (ASR) systems can achieve high performance in terms of recognition accuracy. However, a perfectly accurate transcript still can be challenging to read due to grammatical errors, disfluency, and other noises common in spoken communication. These readable issues introduced by speakers and ASR systems will impair the performance of downstream tasks and the understanding of human readers. In this work, we present a task called ASR post-processing for readability (APR) and formulate it as a sequence-to-sequence text generation problem. The APR task aims to transform the noisy ASR output into a readable text for humans and downstream tasks while maintaining the semantic meaning of speakers. We further study the APR task from the benchmark dataset, evaluation metrics, and baseline models: First, to address the lack of task-specific data, we propose a method to construct a dataset for the APR task by using the data collected for grammatical error correction. Second, we utilize metrics adapted or borrowed from similar tasks to evaluate model performance on the APR task. Lastly, we use several typical or adapted pre-trained models as the baseline models for the APR task. Furthermore, we fine-tune the baseline models on the constructed dataset and compare their performance with a traditional pipeline method in terms of proposed evaluation metrics. Experimental results show that all the fine-tuned baseline models perform better than the traditional pipeline method, and our adapted RoBERTa model outperforms the pipeline method by 4.95 and 6.63 BLEU points on two test sets, respectively. The human evaluation and case study further reveal the ability of the proposed model to improve the readability of ASR transcripts.
Nowadays Automatic Speech Recognition (ASR) systems can accurately recognize which words are said. However, due to the disfluency, grammatical error, and other phenomena in spontaneous speech, the verbatim transcription of ASR impairs its readability, which is crucial for human comprehension and downstream tasks processing that need to understand the meaning and purpose of what is spoken. In this work, we formulate the ASR post-processing for readability (APR) as a sequence-to-sequence text generation problem that aims to transform the incorrect and noisy ASR output into readable text for humans and downstream tasks. We leverage the Metadata Extraction (MDE) corpus to construct a task-specific dataset for our study. To solve the problem of too little training data, we propose a novel data augmentation method that synthesizes large-scale training data from the grammatical error correction dataset. We propose a model based on the pre-trained language model to perform the APR task and train the model with a two-stage training strategy to better exploit the augmented data. On the constructed test set, our approach outperforms the best baseline system by a large margin of 17.53 on BLEU and 13.26 on readability-aware WER (RA-WER). The human evaluation also shows that our model can generate more human-readable transcripts than the baseline method.
19 June 2023 in ACM Transactions on Asian and Low-Resource Language Information Processing
This work presents the task of text polishing, which generates a sentence that is more graceful than the input sentence while retaining its semantic meaning. Text polishing has great value in real usage and is an important component in modern writing assistance systems. However, the task is still not well studied in the literature. Further research in this important direction requires more formal task definitions, benchmark datasets, and powerful baseline models. In this work, we formulate the task as a context-dependent text generation problem and conduct a case study on the text polishing with Chinese idiom. To circumvent the difficulties of task data annotation, we propose a semi-automatic data construction pipeline based on human-machine collaboration, and establish a large-scale text polishing dataset consisting of 1.5 million instances. We propose two types of task-specific pre-training objectives for the text polishing task and implement a series of Transformer-based models pre-trained on a massive Chinese corpus as baselines. We conduct extensive experiments with the baseline models on the constructed text polishing datasets and have some major findings. The human evaluation further reveals the polishing ability of the final system.
We present SkillNet-NLG, a sparsely activated approach that handles many natural language generation tasks with one model. Different from traditional dense models that always activate all the parameters, SkillNet-NLG selectively activates relevant parts of the parameters to accomplish a task, where the relevance is controlled by a set of predefined skills. The strength of such model design is that it provides an opportunity to precisely adapt relevant skills to learn new tasks effectively. We evaluate on Chinese natural language generation tasks. Results show that, with only one model file, SkillNet-NLG outperforms previous best performance methods on four of five tasks. SkillNet-NLG performs better than two multi-task learning baselines (a dense model and a Mixture-of-Expert model) and achieves comparable performance to task-specific models. Lastly, SkillNet-NLG surpasses baseline systems when being adapted to new tasks.
The multilingual neural machine translation (NMT) model can handle translation between more than one language pair. From the perspective of industrial applications, the modularized multilingual NMT model (M2 model) that only shares modules between the same languages is a practical alternative to the model that shares one encoder and one decoder (1-1 model). Previous works have proven that the M2 model can benefit from multiway training without suffering from capacity bottlenecks and exhibits better performance than the 1-1 model. However, the M2 model trained on English-centric data is incapable of zero-shot translation due to the ill-formed interlingual space. In this study, we propose a framework to help the M2 model form an interlingual space for zero-shot translation. Using this framework, we devise an approach that combines multiway training with a denoising autoencoder task and incorporates a Transformer attention bridge module based on the attention mechanism. We experimentally show that the proposed method can form an improved interlingual space in two zero-shot experiments. Our findings further extend the use of the M2 model for multilingual translation in industrial applications.