This paper investigates whether the power of the models pre-trained on text data, such as BERT, can be transferred to general token sequence classification applications. To verify pre-trained models' transferability, we test the pre-trained models on
text classification tasks with meanings of tokens mismatches, and real-world non-text token sequence classification data, including amino acid, DNA, and music. We find that even on non-text data, the models pre-trained on text converge faster, perform better than the randomly initialized models, and only slightly worse than the models using task-specific knowledge. We also find that the representations of the text and non-text pre-trained models share non-trivial similarities.
The need to deploy large-scale pre-trained models on edge devices under limited computational resources has led to substantial research to compress these large models. However, less attention has been given to compress the task-specific models. In th
is work, we investigate the different methods of unstructured pruning on task-specific models for Aspect-based Sentiment Analysis (ABSA) tasks. Specifically, we analyze differences in the learning dynamics of pruned models by using the standard pruning techniques to achieve high-performing sparse networks. We develop a hypothesis to demonstrate the effectiveness of local pruning over global pruning considering a simple CNN model. Later, we utilize the hypothesis to demonstrate the efficacy of the pruned state-of-the-art model compared to the over-parameterized state-of-the-art model under two settings, the first considering the baselines for the same task used for generating the hypothesis, i.e., aspect extraction and the second considering a different task, i.e., sentiment analysis. We also provide discussion related to the generalization of the pruning hypothesis.
This paper introduces our neural machine translation systems' participation in the WAT 2021 shared translation tasks (team ID: sakura). We participated in the (i) NICT-SAP, (ii) Japanese-English multimodal translation, (iii) Multilingual Indic, and (
iv) Myanmar-English translation tasks. Multilingual approaches such as mBART (Liu et al., 2020) are capable of pre-training a complete, multilingual sequence-to-sequence model through denoising objectives, making it a great starting point for building multilingual translation systems. Our main focus in this work is to investigate the effectiveness of multilingual finetuning on such a multilingual language model on various translation tasks, including low-resource, multimodal, and mixed-domain translation. We further explore a multimodal approach based on universal visual representation (Zhang et al., 2019) and compare its performance against a unimodal approach based on mBART alone.
With the growing popularity of smart speakers, such as Amazon Alexa, speech is becoming one of the most important modes of human-computer interaction. Automatic speech recognition (ASR) is arguably the most critical component of such systems, as erro
rs in speech recognition propagate to the downstream components and drastically degrade the user experience. A simple and effective way to improve the speech recognition accuracy is to apply automatic post-processor to the recognition result. However, training a post-processor requires parallel corpora created by human annotators, which are expensive and not scalable. To alleviate this problem, we propose Back TranScription (BTS), a denoising-based method that can create such corpora without human labor. Using a raw corpus, BTS corrupts the text using Text-to-Speech (TTS) and Speech-to-Text (STT) systems. Then, a post-processing model can be trained to reconstruct the original text given the corrupted input. Quantitative and qualitative evaluations show that a post-processor trained using our approach is highly effective in fixing non-trivial speech recognition errors such as mishandling foreign words. We present the generated parallel corpus and post-processing platform to make our results publicly available.
