Jejueo was classified as critically endangered by UNESCO in 2010. Although diverse efforts to revitalize it have been made, there have been few computational approaches. Motivated by this, we construct two new Jejueo datasets: Jejueo Interview Transcripts (JIT) and Jejueo Single Speaker Speech (JSS). The JIT dataset is a parallel corpus containing 170k+ Jejueo-Korean sentences, and the JSS dataset consists of 10k high-quality audio files recorded by a native Jejueo speaker and a transcript file. Subsequently, we build neural systems of machine translation and speech synthesis using them. All resources are publicly available via our GitHub repository. We hope that these datasets will attract interest of both language and machine learning communities.
Pre-trained word embeddings encode general word semantics and lexical regularities of natural language, and have proven useful across many NLP tasks, including word sense disambiguation, machine translation, and sentiment analysis, to name a few. In supervised tasks such as multiclass text classification (the focus of this article) it seems appealing to enhance word representations with ad-hoc embeddings that encode task-specific information. We propose (supervised) word-class embeddings (WCEs), and show that, when concatenated to (unsupervised) pre-trained word embeddings, they substantially facilitate the training of deep-learning models in multiclass classification by topic. We show empirical evidence that WCEs yield a consistent improvement in multiclass classification accuracy, using four popular neural architectures and six widely used and publicly available datasets for multiclass text classification. Our code that implements WCEs is publicly available at https://github.com/AlexMoreo/word-class-embeddings
To acquire a new skill, humans learn better and faster if a tutor, based on their current knowledge level, informs them of how much attention they should pay to particular content or practice problems. Similarly, a machine learning model could potentially be trained better with a scorer that "adapts" to its current learning state and estimates the importance of each training data instance. Training such an adaptive scorer efficiently is a challenging problem; in order to precisely quantify the effect of a data instance at a given time during the training, it is typically necessary to first complete the entire training process. To efficiently optimize data usage, we propose a reinforcement learning approach called Differentiable Data Selection (DDS). In DDS, we formulate a scorer network as a learnable function of the training data, which can be efficiently updated along with the main model being trained. Specifically, DDS updates the scorer with an intuitive reward signal: it should up-weigh the data that has a similar gradient with a dev set upon which we would finally like to perform well. Without significant computing overhead, DDS delivers strong and consistent improvements over several strong baselines on two very different tasks of machine translation and image classification.
The key challenge of multi-domain translation lies in simultaneously encoding both the general knowledge shared across domains and the particular knowledge distinctive to each domain in a unified model. Previous work shows that the standard neural machine translation (NMT) model, trained on mixed-domain data, generally captures the general knowledge, but misses the domain-specific knowledge. In response to this problem, we augment NMT model with additional domain transformation networks to transform the general representations to domain-specific representations, which are subsequently fed to the NMT decoder. To guarantee the knowledge transformation, we also propose two complementary supervision signals by leveraging the power of knowledge distillation and adversarial learning. Experimental results on several language pairs, covering both balanced and unbalanced multi-domain translation, demonstrate the effectiveness and universality of the proposed approach. Encouragingly, the proposed unified model achieves comparable results with the fine-tuning approach that requires multiple models to preserve the particular knowledge. Further analyses reveal that the domain transformation networks successfully capture the domain-specific knowledge as expected.
A relation tuple consists of two entities and the relation between them, and often such tuples are found in unstructured text. There may be multiple relation tuples present in a text and they may share one or both entities among them. Extracting such relation tuples from a sentence is a difficult task and sharing of entities or overlapping entities among the tuples makes it more challenging. Most prior work adopted a pipeline approach where entities were identified first followed by finding the relations among them, thus missing the interaction among the relation tuples in a sentence. In this paper, we propose two approaches to use encoder-decoder architecture for jointly extracting entities and relations. In the first approach, we propose a representation scheme for relation tuples which enables the decoder to generate one word at a time like machine translation models and still finds all the tuples present in a sentence with full entity names of different length and with overlapping entities. Next, we propose a pointer network-based decoding approach where an entire tuple is generated at every time step. Experiments on the publicly available New York Times corpus show that our proposed approaches outperform previous work and achieve significantly higher F1 scores.
In sequence to sequence learning, the self-attention mechanism proves to be highly effective, and achieves significant improvements in many tasks. However, the self-attention mechanism is not without its own flaws. Although self-attention can model extremely long dependencies, the attention in deep layers tends to overconcentrate on a single token, leading to insufficient use of local information and difficultly in representing long sequences. In this work, we explore parallel multi-scale representation learning on sequence data, striving to capture both long-range and short-range language structures. To this end, we propose the Parallel MUlti-Scale attEntion (MUSE) and MUSE-simple. MUSE-simple contains the basic idea of parallel multi-scale sequence representation learning, and it encodes the sequence in parallel, in terms of different scales with the help from self-attention, and pointwise transformation. MUSE builds on MUSE-simple and explores combining convolution and self-attention for learning sequence representations from more different scales. We focus on machine translation and the proposed approach achieves substantial performance improvements over Transformer, especially on long sequences. More importantly, we find that although conceptually simple, its success in practice requires intricate considerations, and the multi-scale attention must build on unified semantic space. Under common setting, the proposed model achieves substantial performance and outperforms all previous models on three main machine translation tasks. In addition, MUSE has potential for accelerating inference due to its parallelism. Code will be available at https://github.com/lancopku/MUSE
Non-Autoregressive Neural Machine Translation (NAT) achieves significant decoding speedup through generating target words independently and simultaneously. However, in the context of non-autoregressive translation, the word-level cross-entropy loss cannot model the target-side sequential dependency properly, leading to its weak correlation with the translation quality. As a result, NAT tends to generate influent translations with over-translation and under-translation errors. In this paper, we propose to train NAT to minimize the Bag-of-Ngrams (BoN) difference between the model output and the reference sentence. The bag-of-ngrams training objective is differentiable and can be efficiently calculated, which encourages NAT to capture the target-side sequential dependency and correlates well with the translation quality. We validate our approach on three translation tasks and show that our approach largely outperforms the NAT baseline by about 5.0 BLEU scores on WMT14 En$\leftrightarrow$De and about 2.5 BLEU scores on WMT16 En$\leftrightarrow$Ro.
This work aims to produce translations that convey source language content at a formality level that is appropriate for a particular audience. Framing this problem as a neural sequence-to-sequence task ideally requires training triplets consisting of a bilingual sentence pair labeled with target language formality. However, in practice, available training examples are limited to English sentence pairs of different styles, and bilingual parallel sentences of unknown formality. We introduce a novel training scheme for multi-task models that automatically generates synthetic training triplets by inferring the missing element on the fly, thus enabling end-to-end training. Comprehensive automatic and human assessments show that our best model outperforms existing models by producing translations that better match desired formality levels while preserving the source meaning.
Non-autoregressive translation (NAT) models remove the dependence on previous target tokens and generate all target tokens in parallel, resulting in significant inference speedup but at the cost of inferior translation accuracy compared to autoregressive translation (AT) models. Considering that AT models have higher accuracy and are easier to train than NAT models, and both of them share the same model configurations, a natural idea to improve the accuracy of NAT models is to transfer a well-trained AT model to an NAT model through fine-tuning. However, since AT and NAT models differ greatly in training strategy, straightforward fine-tuning does not work well. In this work, we introduce curriculum learning into fine-tuning for NAT. Specifically, we design a curriculum in the fine-tuning process to progressively switch the training from autoregressive generation to non-autoregressive generation. Experiments on four benchmark translation datasets show that the proposed method achieves good improvement (more than $1$ BLEU score) over previous NAT baselines in terms of translation accuracy, and greatly speed up (more than $10$ times) the inference process over AT baselines.
Layer normalization (LayerNorm) is a technique to normalize the distributions of intermediate layers. It enables smoother gradients, faster training, and better generalization accuracy. However, it is still unclear where the effectiveness stems from. In this paper, our main contribution is to take a step further in understanding LayerNorm. Many of previous studies believe that the success of LayerNorm comes from forward normalization. Unlike them, we find that the derivatives of the mean and variance are more important than forward normalization by re-centering and re-scaling backward gradients. Furthermore, we find that the parameters of LayerNorm, including the bias and gain, increase the risk of over-fitting and do not work in most cases. Experiments show that a simple version of LayerNorm (LayerNorm-simple) without the bias and gain outperforms LayerNorm on four datasets. It obtains the state-of-the-art performance on En-Vi machine translation. To address the over-fitting problem, we propose a new normalization method, Adaptive Normalization (AdaNorm), by replacing the bias and gain with a new transformation function. Experiments show that AdaNorm demonstrates better results than LayerNorm on seven out of eight datasets.
The dominant graph-to-sequence transduction models employ graph neural networks for graph representation learning, where the structural information is reflected by the receptive field of neurons. Unlike graph neural networks that restrict the information exchange between immediate neighborhood, we propose a new model, known as Graph Transformer, that uses explicit relation encoding and allows direct communication between two distant nodes. It provides a more efficient way for global graph structure modeling. Experiments on the applications of text generation from Abstract Meaning Representation (AMR) and syntax-based neural machine translation show the superiority of our proposed model. Specifically, our model achieves 27.4 BLEU on LDC2015E86 and 29.7 BLEU on LDC2017T10 for AMR-to-text generation, outperforming the state-of-the-art results by up to 2.2 points. On the syntax-based translation tasks, our model establishes new single-model state-of-the-art BLEU scores, 21.3 for English-to-German and 14.1 for English-to-Czech, improving over the existing best results, including ensembles, by over 1 BLEU.
Large-scale pre-trained language model such as BERT has achieved great success in language understanding tasks. However, it remains an open question how to utilize BERT for language generation. In this paper, we present a novel approach, Conditional Masked Language Modeling (C-MLM), to enable the finetuning of BERT on target generation tasks. The finetuned BERT (teacher) is exploited as extra supervision to improve conventional Seq2Seq models (student) for better text generation performance. By leveraging BERT's idiosyncratic bidirectional nature, distilling knowledge learned in BERT can encourage auto-regressive Seq2Seq models to plan ahead, imposing global sequence-level supervision for coherent text generation. Experiments show that the proposed approach significantly outperforms strong Transformer baselines on multiple language generation tasks such as machine translation and text summarization. Our proposed model also achieves new state of the art on IWSLT German-English and English-Vietnamese MT datasets. Code is available at https://github.com/ChenRocks/Distill-BERT-Textgen.
Automatic post-editing (APE), which aims to correct errors in the output of machine translation systems in a post-processing step, is an important task in natural language processing. While recent work has achieved considerable performance gains by using neural networks, how to model the copying mechanism for APE remains a challenge. In this work, we propose a new method for modeling copying for APE. To better identify translation errors, our method learns the representations of source sentences and system outputs in an interactive way. These representations are used to explicitly indicate which words in the system outputs should be copied, which is useful to help CopyNet (Gu et al., 2016) better generate post-edited translations. Experiments on the datasets of the WMT 2016-2017 APE shared tasks show that our approach outperforms all best published results.
In this work, we present several deep learning models for the automatic diacritization of Arabic text. Our models are built using two main approaches, viz. Feed-Forward Neural Network (FFNN) and Recurrent Neural Network (RNN), with several enhancements such as 100-hot encoding, embeddings, Conditional Random Field (CRF) and Block-Normalized Gradient (BNG). The models are tested on the only freely available benchmark dataset and the results show that our models are either better or on par with other models, which require language-dependent post-processing steps, unlike ours. Moreover, we show that diacritics in Arabic can be used to enhance the models of NLP tasks such as Machine Translation (MT) by proposing the Translation over Diacritization (ToD) approach.
The Transformer model is widely successful on many natural language processing tasks. However, the quadratic complexity of self-attention limit its application on long text. In this paper, adopting a fine-to-coarse attention mechanism on multi-scale spans via binary partitioning (BP), we propose BP-Transformer (BPT for short). BPT yields $O(k\cdot n\log (n/k))$ connections where $k$ is a hyperparameter to control the density of attention. BPT has a good balance between computation complexity and model capacity. A series of experiments on text classification, machine translation and language modeling shows BPT has a superior performance for long text than previous self-attention models. Our code, hyperparameters and CUDA kernels for sparse attention are available in PyTorch.
Traditional methods for deep NLG adopt pipeline approaches comprising stages such as constructing syntactic input, predicting function words, linearizing the syntactic input and generating the surface forms. Though easier to visualize, pipeline approaches suffer from error propagation. In addition, information available across modules cannot be leveraged by all modules. We construct a transition-based model to jointly perform linearization, function word prediction and morphological generation, which considerably improves upon the accuracy compared to a pipelined baseline system. On a standard deep input linearization shared task, our system achieves the best results reported so far.
Translating text that diverges from the training domain is a key challenge for machine translation. Domain robustness---the generalization of models to unseen test domains---is low for both statistical (SMT) and neural machine translation (NMT). In this paper, we study the performance of SMT and NMT models on out-of-domain test sets. We find that in unknown domains, SMT and NMT suffer from very different problems: SMT systems are mostly adequate but not fluent, while NMT systems are mostly fluent, but not adequate. For NMT, we identify such hallucinations (translations that are fluent but unrelated to the source) as a key reason for low domain robustness. To mitigate this problem, we empirically compare methods that are reported to improve adequacy or in-domain robustness in terms of their effectiveness at improving domain robustness. In experiments on German to English OPUS data, and German to Romansh (a low-resource setting) we find that several methods improve domain robustness. While those methods do lead to higher BLEU scores overall, they only slightly increase the adequacy of translations compared to SMT.
We introduce Data Diversification: a simple but effective strategy to boost neural machine translation (NMT) performance. It diversifies the training data by using the predictions of multiple forward and backward models and then merging them with the original dataset on which the final NMT model is trained. Our method is applicable to all NMT models. It does not require extra monolingual data like back-translation, nor does it add more computations and parameters like ensembles of models. Our method achieves state-of-the-art BLEU scores of 30.7 and 43.7 in the WMT'14 English-German and English-French translation tasks, respectively. It also substantially improves on 8 other translation tasks: 4 IWSLT tasks (English-German and English-French) and 4 low-resource translation tasks (English-Nepali and English-Sinhala). We demonstrate that our method is more effective than knowledge distillation and dual learning, it exhibits strong correlation with ensembles of models, and it trades perplexity off for better BLEU score. We have released our source code at https://github.com/nxphi47/data_diversification
This work introduces a machine translation task where the output is aimed at audiences of different levels of target language proficiency. We collect a high quality dataset of news articles available in English and Spanish, written for diverse grade levels and propose a method to align segments across comparable bilingual articles. The resulting dataset makes it possible to train multi-task sequence-to-sequence models that translate Spanish into English targeted at an easier reading grade level than the original Spanish. We show that these multi-task models outperform pipeline approaches that translate and simplify text independently.
The dominant approach to sequence generation is to produce a sequence in some predefined order, e.g. left to right. In contrast, we propose a more general model that can generate the output sequence by inserting tokens in any arbitrary order. Our model learns decoding order as a result of its training procedure. Our experiments show that this model is superior to fixed order models on a number of sequence generation tasks, such as Machine Translation, Image-to-LaTeX and Image Captioning.