Machine translation models struggle when translating out-of-domain text, which makes domain adaptation a topic of critical importance. However, most domain adaptation methods focus on fine-tuning or training the entire or part of the model on every new domain, which can be costly. On the other hand, semi-parametric models have been shown to successfully perform domain adaptation by retrieving examples from an in-domain datastore (Khandelwal et al., 2021). A drawback of these retrieval-augmented models, however, is that they tend to be substantially slower. In this paper, we explore several approaches to speed up nearest neighbor machine translation. We adapt the methods recently proposed by He et al. (2021) for language modeling, and introduce a simple but effective caching strategy that avoids performing retrieval when similar contexts have been seen before. Translation quality and runtimes for several domains show the effectiveness of the proposed solutions.
Automatic text summarization aims at condensing a document to a shorter version while preserving the key information. Different from extractive summarization which simply selects text fragments from the document, abstractive summarization generates the summary in a word-by-word manner. Most current state-of-the-art (SOTA) abstractive summarization methods are based on the Transformer-based encoder-decoder architecture and focus on novel self-supervised objectives in pre-training. While these models well capture the contextual information among words in documents, little attention has been paid to incorporating global semantics to better fine-tune for the downstream abstractive summarization task. In this study, we propose a topic-aware abstractive summarization (TAAS) framework by leveraging the underlying semantic structure of documents represented by their latent topics. Specifically, TAAS seamlessly incorporates a neural topic modeling into an encoder-decoder based sequence generation procedure via attention for summarization. This design is able to learn and preserve global semantics of documents and thus makes summarization effective, which has been proved by our experiments on real-world datasets. As compared to several cutting-edge baseline methods, we show that TAAS outperforms BART, a well-recognized SOTA model, by 2%, 8%, and 12% regarding the F measure of ROUGE-1, ROUGE-2, and ROUGE-L, respectively. TAAS also achieves comparable performance to PEGASUS and ProphetNet, which is difficult to accomplish given that training PEGASUS and ProphetNet requires enormous computing capacity beyond what we used in this study.
Food is not only a basic human necessity but also a key factor driving a society's health and economic well-being. As a result, the cooking domain is a popular use-case to demonstrate decision-support (AI) capabilities in service of benefits like precision health with tools ranging from information retrieval interfaces to task-oriented chatbots. An AI here should understand concepts in the food domain (e.g., recipes, ingredients), be tolerant to failures encountered while cooking (e.g., browning of butter), handle allergy-based substitutions, and work with multiple data modalities (e.g. text and images). However, the recipes today are handled as textual documents which makes it difficult for machines to read, reason and handle ambiguity. This demands a need for better representation of the recipes, overcoming the ambiguity and sparseness that exists in the current textual documents. In this paper, we discuss the construction of a machine-understandable rich recipe representation (R3), in the form of plans, from the recipes available in natural language. R3 is infused with additional knowledge such as information about allergens and images of ingredients, possible failures and tips for each atomic cooking step. To show the benefits of R3, we also present TREAT, a tool for recipe retrieval which uses R3 to perform multi-modal reasoning on the recipe's content (plan objects - ingredients and cooking tools), food preparation process (plan actions and time), and media type (image, text). R3 leads to improved retrieval efficiency and new capabilities that were hither-to not possible in textual representation.
Emotion Recognition (ER) aims to classify human utterances into different emotion categories. Based on early-fusion and self-attention-based multimodal interaction between text and acoustic modalities, in this paper, we propose a multimodal multitask learning approach for ER from individual utterances in isolation. Experiments on the IEMOCAP benchmark show that our proposed model performs better than our re-implementation of state-of-the-art and achieves better performance than all other unimodal and multimodal approaches in literature. In addition, strong baselines and ablation studies prove the effectiveness of our proposed approach. We make all our codes publicly available on GitHub.
In this work, our objective is to address the problems of generalization and flexibility for text recognition in documents. We introduce a new model that exploits the repetitive nature of characters in languages, and decouples the visual representation learning and linguistic modelling stages. By doing this, we turn text recognition into a shape matching problem, and thereby achieve generalization in appearance and flexibility in classes. We evaluate the new model on both synthetic and real datasets across different alphabets and show that it can handle challenges that traditional architectures are not able to solve without expensive retraining, including: (i) it can generalize to unseen fonts without new exemplars from them; (ii) it can flexibly change the number of classes, simply by changing the exemplars provided; and (iii) it can generalize to new languages and new characters that it has not been trained for by providing a new glyph set. We show significant improvements over state-of-the-art models for all these cases.
Recent trends in natural language processing using pretraining have shifted focus towards pretraining and fine-tuning approaches for text generation. Often the focus has been on task-agnostic approaches that generalize the language modeling objective. We propose question generation as a pretraining method, which better aligns with the text generation objectives. Our text generation models pretrained with this method are better at understanding the essence of the input and are better language models for the target task. When evaluated on two text generation tasks, abstractive summarization and answer-focused question generation, our models result in state-of-the-art performances in terms of automatic metrics. Human evaluators also found our summaries and generated questions to be more natural, concise and informative.
Text style transfer aims to change the style of sentences while preserving the semantic meanings. Due to the lack of parallel data, the Denoising Auto-Encoder (DAE) is widely used in this task to model distributions of different sentence styles. However, because of the conflict between the target of the conventional denoising procedure and the target of style transfer task, the vanilla DAE can not produce satisfying enough results. To improve the transferability of the model, most of the existing works combine DAE with various complicated unsupervised networks, which makes the whole system become over-complex. In this work, we design a novel DAE model named Style-Enhanced DAE (SE-DAE), which is specifically designed for the text style transfer task. Compared with previous complicated style-transfer models, our model do not consist of any complicated unsupervised networks, but only relies on the high-quality pseudo-parallel data generated by a novel data refinement mechanism. Moreover, to alleviate the conflict between the targets of the conventional denoising procedure and the style transfer task, we propose another novel style denoising mechanism, which is more compatible with the target of the style transfer task. We validate the effectiveness of our model on two style benchmark datasets. Both automatic evaluation and human evaluation show that our proposed model is highly competitive compared with previous strong the state of the art (SOTA) approaches and greatly outperforms the vanilla DAE.
We introduce CM3, a family of causally masked generative models trained over a large corpus of structured multi-modal documents that can contain both text and image tokens. Our new causally masked approach generates tokens left to right while also masking out a small number of long token spans that are generated at the end of the string, instead of their original positions. The casual masking object provides a type of hybrid of the more common causal and masked language models, by enabling full generative modeling while also providing bidirectional context when generating the masked spans. We train causally masked language-image models on large-scale web and Wikipedia articles, where each document contains all of the text, hypertext markup, hyperlinks, and image tokens (from a VQVAE-GAN), provided in the order they appear in the original HTML source (before masking). The resulting CM3 models can generate rich structured, multi-modal outputs while conditioning on arbitrary masked document contexts, and thereby implicitly learn a wide range of text, image, and cross modal tasks. They can be prompted to recover, in a zero-shot fashion, the functionality of models such as DALL-E, GENRE, and HTLM. We set the new state-of-the-art in zero-shot summarization, entity linking, and entity disambiguation while maintaining competitive performance in the fine-tuning setting. We can generate images unconditionally, conditioned on text (like DALL-E) and do captioning all in a zero-shot setting with a single model.
We present a framework for learning multimodal representations from unlabeled data using convolution-free Transformer architectures. Specifically, our Video-Audio-Text Transformer (VATT) takes raw signals as inputs and extracts multimodal representations that are rich enough to benefit a variety of downstream tasks. We train VATT end-to-end from scratch using multimodal contrastive losses and evaluate its performance by the downstream tasks of video action recognition, audio event classification, image classification, and text-to-video retrieval. Furthermore, we study a modality-agnostic single-backbone Transformer by sharing weights among the three modalities. We show that the convolution-free VATT outperforms state-of-the-art ConvNet-based architectures in the downstream tasks. Especially, VATT's vision Transformer achieves the top-1 accuracy of 82.1% on Kinetics-400, 83.6% on Kinetics-600,and 41.1% on Moments in Time, new records while avoiding supervised pre-training. Transferring to image classification leads to 78.7% top-1 accuracy on ImageNet compared to 64.7% by training the same Transformer from scratch, showing the generalizability of our model despite the domain gap between videos and images. VATT's audio Transformer also sets a new record on waveform-based audio event recognition by achieving the mAP of 39.4% on AudioSet without any supervised pre-training.
Training an end-to-end (E2E) neural network speech-to-intent (S2I) system that directly extracts intents from speech requires large amounts of intent-labeled speech data, which is time consuming and expensive to collect. Initializing the S2I model with an ASR model trained on copious speech data can alleviate data sparsity. In this paper, we attempt to leverage NLU text resources. We implemented a CTC-based S2I system that matches the performance of a state-of-the-art, traditional cascaded SLU system. We performed controlled experiments with varying amounts of speech and text training data. When only a tenth of the original data is available, intent classification accuracy degrades by 7.6% absolute. Assuming we have additional text-to-intent data (without speech) available, we investigated two techniques to improve the S2I system: (1) transfer learning, in which acoustic embeddings for intent classification are tied to fine-tuned BERT text embeddings; and (2) data augmentation, in which the text-to-intent data is converted into speech-to-intent data using a multi-speaker text-to-speech system. The proposed approaches recover 80% of performance lost due to using limited intent-labeled speech.