PyTorchKR Papers

최신 논문

총 5,082편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2102.10246

CDA: a Cost Efficient Content-based Multilingual Web Document Aligner

We introduce a Content-based Document Alignment approach (CDA), an efficient method to align multilingual web documents based on content in creating parallel training data for machine translation (MT) systems operating at the industrial level. CDA works in two steps: (i) projecting documents of a web domain to a shared multilingual space; then (ii) aligning them based on the similarity of their representations in such space. We leverage lexical translation models to build vector representations using TF-IDF. CDA achieves performance comparable with state-of-the-art systems in the WMT-16 Bilingual Document Alignment Shared Task benchmark while operating in multilingual space. Besides, we created two web-scale datasets to examine the robustness of CDA in an industrial setting involving up to 28 languages and millions of documents. The experiments show that CDA is robust, cost-effective, and is significantly superior in (i) processing large and noisy web data and (ii) scaling to new and low-resourced languages.

arXiv:2102.10243공식 코드 · Papers with Code

Machine Translation Customization via Automatic Training Data Selection from the Web

Machine translation (MT) systems, especially when designed for an industrial setting, are trained with general parallel data derived from the Web. Thus, their style is typically driven by word/structure distribution coming from the average of many domains. In contrast, MT customers want translations to be specialized to their domain, for which they are typically able to provide text samples. We describe an approach for customizing MT systems on specific domains by selecting data similar to the target customer data to train neural translation models. We build document classifiers using monolingual target data, e.g., provided by the customers to select parallel training data from Web crawled data. Finally, we train MT models on our automatically selected data, obtaining a system specialized to the target domain. We tested our approach on the benchmark from WMT-18 Translation Task for News domains enabling comparisons with state-of-the-art MT systems. The results show that our models outperform the top systems while using less data and smaller models.

arXiv:2102.10160공식 코드 · Papers with Code

Multi-Domain Adaptation in Neural Machine Translation Through Multidimensional Tagging

While NMT has achieved remarkable results in the last 5 years, production systems come with strict quality requirements in arbitrarily niche domains that are not always adequately covered by readily available parallel corpora. This is typically addressed by training domain specific models, using fine-tuning methods and some variation of back-translation on top of in-domain monolingual corpora. However, industrial practitioners can rarely afford to focus on a single domain. A far more typical scenario includes a set of closely related, yet succinctly different sub-domains. At Booking.com, we need to translate property descriptions, user reviews, as well as messages, (for example those sent between a customer and an agent or property manager). An editor might need to translate articles across a set of different topics. An e-commerce platform would typically need to translate both the description of each item and the user generated content related to them. To this end, we propose MDT: a novel method to simultaneously fine-tune on several sub-domains by passing multidimensional sentence-level information to the model during training and inference. We show that MDT achieves results competitive to N specialist models each fine-tuned on a single constituent domain, while effectively serving all N sub-domains, therefore cutting development and maintenance costs by the same factor. Besides BLEU (industry standard automatic evaluation metric known to only weakly correlate with human judgement) we also report rigorous human evaluation results for all models and sub-domains as well as specific examples that better contextualise the performance of each model in terms of adequacy and fluency. To facilitate further research, we plan to make the code available upon acceptance.

arXiv:2102.10437

Understanding and Enhancing the Use of Context for Machine Translation

To understand and infer meaning in language, neural models have to learn complicated nuances. Discovering distinctive linguistic phenomena from data is not an easy task. For instance, lexical ambiguity is a fundamental feature of language which is challenging to learn. Even more prominently, inferring the meaning of rare and unseen lexical units is difficult with neural networks. Meaning is often determined from context. With context, languages allow meaning to be conveyed even when the specific words used are not known by the reader. To model this learning process, a system has to learn from a few instances in context and be able to generalize well to unseen cases. The learning process is hindered when training data is scarce for a task. Even with sufficient data, learning patterns for the long tail of the lexical distribution is challenging. In this thesis, we focus on understanding certain potentials of contexts in neural models and design augmentation models to benefit from them. We focus on machine translation as an important instance of the more general language understanding problem. To translate from a source language to a target language, a neural model has to understand the meaning of constituents in the provided context and generate constituents with the same meanings in the target language. This task accentuates the value of capturing nuances of language and the necessity of generalization from few observations. The main problem we study in this thesis is what neural machine translation models learn from data and how we can devise more focused contexts to enhance this learning. Looking more in-depth into the role of context and the impact of data on learning models is essential to advance the NLP field. Moreover, it helps highlight the vulnerabilities of current neural networks and provides insights into designing more robust models.

arXiv:2102.11174공식 코드 · Papers with Code

Linear Transformers Are Secretly Fast Weight Programmers

We show the formal equivalence of linearised self-attention mechanisms and fast weight controllers from the early '90s, where a ``slow" neural net learns by gradient descent to program the ``fast weights" of another net through sequences of elementary programming instructions which are additive outer products of self-invented activation patterns (today called keys and values). Such Fast Weight Programmers (FWPs) learn to manipulate the contents of a finite memory and dynamically interact with it. We infer a memory capacity limitation of recent linearised softmax attention variants, and replace the purely additive outer products by a delta rule-like programming instruction, such that the FWP can more easily learn to correct the current mapping from keys to values. The FWP also learns to compute dynamically changing learning rates. We also propose a new kernel function to linearise attention which balances simplicity and effectiveness. We conduct experiments on synthetic retrieval problems as well as standard machine translation and language modelling tasks which demonstrate the benefits of our methods.

arXiv:2102.09377공식 코드 · Papers with Code

Learning Skill Equivalencies Across Platform Taxonomies

Assessment and reporting of skills is a central feature of many digital learning platforms. With students often using multiple platforms, cross-platform assessment has emerged as a new challenge. While technologies such as Learning Tools Interoperability (LTI) have enabled communication between platforms, reconciling the different skill taxonomies they employ has not been solved at scale. In this paper, we introduce and evaluate a methodology for finding and linking equivalent skills between platforms by utilizing problem content as well as the platform's clickstream data. We propose six models to represent skills as continuous real-valued vectors and leverage machine translation to map between skill spaces. The methods are tested on three digital learning platforms: ASSISTments, Khan Academy, and Cognitive Tutor. Our results demonstrate reasonable accuracy in skill equivalency prediction from a fine-grained taxonomy to a coarse-grained one, achieving an average recall@5 of 0.8 between the three platforms. Our skill translation approach has implications for aiding in the tedious, manual process of taxonomy to taxonomy mapping work, also called crosswalks, within the tutoring as well as standardized testing worlds.

arXiv:2102.08934공식 코드 · Papers with Code

Sparsely Factored Neural Machine Translation

The standard approach to incorporate linguistic information to neural machine translation systems consists in maintaining separate vocabularies for each of the annotated features to be incorporated (e.g. POS tags, dependency relation label), embed them, and then aggregate them with each subword in the word they belong to. This approach, however, cannot easily accommodate annotation schemes that are not dense for every word. We propose a method suited for such a case, showing large improvements in out-of-domain data, and comparable quality for the in-domain data. Experiments are performed in morphologically-rich languages like Basque and German, for the case of low-resource scenarios.

arXiv:2102.08597공식 코드 · Papers with Code

Beyond Fully-Connected Layers with Quaternions: Parameterization of Hypercomplex Multiplications with $1/n$ Parameters

Recent works have demonstrated reasonable success of representation learning in hypercomplex space. Specifically, "fully-connected layers with Quaternions" (4D hypercomplex numbers), which replace real-valued matrix multiplications in fully-connected layers with Hamilton products of Quaternions, both enjoy parameter savings with only 1/4 learnable parameters and achieve comparable performance in various applications. However, one key caveat is that hypercomplex space only exists at very few predefined dimensions (4D, 8D, and 16D). This restricts the flexibility of models that leverage hypercomplex multiplications. To this end, we propose parameterizing hypercomplex multiplications, allowing models to learn multiplication rules from data regardless of whether such rules are predefined. As a result, our method not only subsumes the Hamilton product, but also learns to operate on any arbitrary nD hypercomplex space, providing more architectural flexibility using arbitrarily $1/n$ learnable parameters compared with the fully-connected layer counterpart. Experiments of applications to the LSTM and Transformer models on natural language inference, machine translation, text style transfer, and subject verb agreement demonstrate architectural flexibility and effectiveness of the proposed approach.

arXiv:2102.07847공식 코드 · Papers with Code

Meta Back-translation

Back-translation is an effective strategy to improve the performance of Neural Machine Translation~(NMT) by generating pseudo-parallel data. However, several recent works have found that better translation quality of the pseudo-parallel data does not necessarily lead to better final translation models, while lower-quality but more diverse data often yields stronger results. In this paper, we propose a novel method to generate pseudo-parallel data from a pre-trained back-translation model. Our method is a meta-learning algorithm which adapts a pre-trained back-translation model so that the pseudo-parallel data it generates would train a forward-translation model to do well on a validation set. In our evaluations in both the standard datasets WMT En-De'14 and WMT En-Fr'14, as well as a multilingual translation setting, our method leads to significant improvements over strong baselines. Our code will be made available.

arXiv:2102.08098공식 코드 · Papers with Code

GradInit: Learning to Initialize Neural Networks for Stable and Efficient Training

Innovations in neural architectures have fostered significant breakthroughs in language modeling and computer vision. Unfortunately, novel architectures often result in challenging hyper-parameter choices and training instability if the network parameters are not properly initialized. A number of architecture-specific initialization schemes have been proposed, but these schemes are not always portable to new architectures. This paper presents GradInit, an automated and architecture agnostic method for initializing neural networks. GradInit is based on a simple heuristic; the norm of each network layer is adjusted so that a single step of SGD or Adam with prescribed hyperparameters results in the smallest possible loss value. This adjustment is done by introducing a scalar multiplier variable in front of each parameter block, and then optimizing these variables using a simple numerical scheme. GradInit accelerates the convergence and test performance of many convolutional architectures, both with or without skip connections, and even without normalization layers. It also improves the stability of the original Transformer architecture for machine translation, enabling training it without learning rate warmup using either Adam or SGD under a wide range of learning rates and momentum coefficients. Code is available at https://github.com/zhuchen03/gradinit.

arXiv:2102.08220공식 코드 · Papers with Code

Non-Autoregressive Text Generation with Pre-trained Language Models

Non-autoregressive generation (NAG) has recently attracted great attention due to its fast inference speed. However, the generation quality of existing NAG models still lags behind their autoregressive counterparts. In this work, we show that BERT can be employed as the backbone of a NAG model to greatly improve performance. Additionally, we devise mechanisms to alleviate the two common problems of vanilla NAG models: the inflexibility of prefixed output length and the conditional independence of individual token predictions. Lastly, to further increase the speed advantage of the proposed model, we propose a new decoding strategy, ratio-first, for applications where the output lengths can be approximately estimated beforehand. For a comprehensive evaluation, we test the proposed model on three text generation tasks, including text summarization, sentence compression and machine translation. Experimental results show that our model significantly outperforms existing non-autoregressive baselines and achieves competitive performance with many strong autoregressive models. In addition, we also conduct extensive analysis experiments to reveal the effect of each proposed component.

arXiv:2102.07380

MAPGN: MAsked Pointer-Generator Network for sequence-to-sequence pre-training

This paper presents a self-supervised learning method for pointer-generator networks to improve spoken-text normalization. Spoken-text normalization that converts spoken-style text into style normalized text is becoming an important technology for improving subsequent processing such as machine translation and summarization. The most successful spoken-text normalization method to date is sequence-to-sequence (seq2seq) mapping using pointer-generator networks that possess a copy mechanism from an input sequence. However, these models require a large amount of paired data of spoken-style text and style normalized text, and it is difficult to prepare such a volume of data. In order to construct spoken-text normalization model from the limited paired data, we focus on self-supervised learning which can utilize unpaired text data to improve seq2seq models. Unfortunately, conventional self-supervised learning methods do not assume that pointer-generator networks are utilized. Therefore, we propose a novel self-supervised learning method, MAsked Pointer-Generator Network (MAPGN). The proposed method can effectively pre-train the pointer-generator network by learning to fill masked tokens using the copy mechanism. Our experiments demonstrate that MAPGN is more effective for pointer-generator networks than the conventional self-supervised learning methods in two spoken-text normalization tasks.

arXiv:2102.06991공식 코드 · Papers with Code

The first large scale collection of diverse Hausa language datasets

Hausa language belongs to the Afroasiatic phylum, and with more first-language speakers than any other sub-Saharan African language. With a majority of its speakers residing in the Northern and Southern areas of Nigeria and the Republic of Niger, respectively, it is estimated that over 100 million people speak the language. Hence, making it one of the most spoken Chadic language. While Hausa is considered well-studied and documented language among the sub-Saharan African languages, it is viewed as a low resource language from the perspective of natural language processing (NLP) due to limited resources to utilise in NLP-related tasks. This is common to most languages in Africa; thus, it is crucial to enrich such languages with resources that will support and speed the pace of conducting various downstream tasks to meet the demand of the modern society. While there exist useful datasets, notably from news sites and religious texts, more diversity is needed in the corpus. We provide an expansive collection of curated datasets consisting of both formal and informal forms of the language from refutable websites and online social media networks, respectively. The collection is large and more diverse than the existing corpora by providing the first and largest set of Hausa social media data posts to capture the peculiarities in the language. The collection also consists of a parallel dataset, which can be used for tasks such as machine translation with applications in areas such as the detection of spurious or inciteful online content. We describe the curation process -- from the collection, preprocessing and how to obtain the data -- and proffer some research problems that could be addressed using the data.

arXiv:2102.07539

Crowdsourcing Parallel Corpus for English-Oromo Neural Machine Translation using Community Engagement Platform

Even though Afaan Oromo is the most widely spoken language in the Cushitic family by more than fifty million people in the Horn and East Africa, it is surprisingly resource-scarce from a technological point of view. The increasing amount of various useful documents written in English language brings to investigate the machine that can translate those documents and make it easily accessible for local language. The paper deals with implementing a translation of English to Afaan Oromo and vice versa using Neural Machine Translation. But the implementation is not very well explored due to the limited amount and diversity of the corpus. However, using a bilingual corpus of just over 40k sentence pairs we have collected, this study showed a promising result. About a quarter of this corpus is collected via Community Engagement Platform (CEP) that was implemented to enrich the parallel corpus through crowdsourcing translations.

arXiv:2102.06558

Continuous Learning in Neural Machine Translation using Bilingual Dictionaries

While recent advances in deep learning led to significant improvements in machine translation, neural machine translation is often still not able to continuously adapt to the environment. For humans, as well as for machine translation, bilingual dictionaries are a promising knowledge source to continuously integrate new knowledge. However, their exploitation poses several challenges: The system needs to be able to perform one-shot learning as well as model the morphology of source and target language. In this work, we proposed an evaluation framework to assess the ability of neural machine translation to continuously learn new phrases. We integrate one-shot learning methods for neural machine translation with different word representations and show that it is important to address both in order to successfully make use of bilingual dictionaries. By addressing both challenges we are able to improve the ability to translate new, rare words and phrases from 30% to up to 70%. The correct lemma is even generated by more than 90%.

arXiv:2102.06320공식 코드 · Papers with Code

On Automatic Parsing of Log Records

Software log analysis helps to maintain the health of software solutions and ensure compliance and security. Existing software systems consist of heterogeneous components emitting logs in various formats. A typical solution is to unify the logs using manually built parsers, which is laborious. Instead, we explore the possibility of automating the parsing task by employing machine translation (MT). We create a tool that generates synthetic Apache log records which we used to train recurrent-neural-network-based MT models. Models' evaluation on real-world logs shows that the models can learn Apache log format and parse individual log records. The median relative edit distance between an actual real-world log record and the MT prediction is less than or equal to 28%. Thus, we show that log parsing using an MT approach is promising.

arXiv:2102.06578

Improving Zero-shot Neural Machine Translation on Language-specific Encoders-Decoders

Recently, universal neural machine translation (NMT) with shared encoder-decoder gained good performance on zero-shot translation. Unlike universal NMT, jointly trained language-specific encoders-decoders aim to achieve universal representation across non-shared modules, each of which is for a language or language family. The non-shared architecture has the advantage of mitigating internal language competition, especially when the shared vocabulary and model parameters are restricted in their size. However, the performance of using multiple encoders and decoders on zero-shot translation still lags behind universal NMT. In this work, we study zero-shot translation using language-specific encoders-decoders. We propose to generalize the non-shared architecture and universal NMT by differentiating the Transformer layers between language-specific and interlingua. By selectively sharing parameters and applying cross-attentions, we explore maximizing the representation universality and realizing the best alignment of language-agnostic information. We also introduce a denoising auto-encoding (DAE) objective to jointly train the model with the translation task in a multi-task manner. Experiments on two public multilingual parallel datasets show that our proposed model achieves a competitive or better results than universal NMT and strong pivot baseline. Moreover, we experiment incrementally adding new language to the trained model by only updating the new model parameters. With this little effort, the zero-shot translation between this newly added language and existing languages achieves a comparable result with the model trained jointly from scratch on all languages.

arXiv:2102.05766

Fused Acoustic and Text Encoding for Multimodal Bilingual Pretraining and Speech Translation

Recently, representation learning for text and speech has successfully improved many language related tasks. However, all existing methods suffer from two limitations: (a) they only learn from one input modality, while a unified representation for both speech and text is needed by tasks such as end-to-end speech translation, and as a result,(b) they can not exploit various large-scale text and speech data and their performance is limited by the scarcity of parallel speech translation data.To address these problems, we propose a Fused Acoustic and Text Masked Language Model (FAT-MLM) which jointly learns a unified representation for both acoustic and text input from various types of corpora including parallel data for speech recognition and machine translation, and even pure speech and text data. Within this cross-modal representation learning framework, we further present an end-to-end model for Fused Acoustic and Text Speech Translation (FAT-ST). Experiments on three translation directions show that by fine-tuning from FAT-MLM, our proposed speech translation models substantially improve translation quality by up to +5.9 BLEU.

arXiv:2102.04020

Quality Estimation without Human-labeled Data

Quality estimation aims to measure the quality of translated content without access to a reference translation. This is crucial for machine translation systems in real-world scenarios where high-quality translation is needed. While many approaches exist for quality estimation, they are based on supervised machine learning requiring costly human labelled data. As an alternative, we propose a technique that does not rely on examples from human-annotators and instead uses synthetic training data. We train off-the-shelf architectures for supervised quality estimation on our synthetic data and show that the resulting models achieve comparable performance to models trained on human-annotated data, both for sentence and word-level prediction.

arXiv:2102.02955

Understanding Pre-Editing for Black-Box Neural Machine Translation

Pre-editing is the process of modifying the source text (ST) so that it can be translated by machine translation (MT) in a better quality. Despite the unpredictability of black-box neural MT (NMT), pre-editing has been deployed in various practical MT use cases. Although many studies have demonstrated the effectiveness of pre-editing methods for particular settings, thus far, a deep understanding of what pre-editing is and how it works for black-box NMT is lacking. To elicit such understanding, we extensively investigated human pre-editing practices. We first implemented a protocol to incrementally record the minimum edits for each ST and collected 6,652 instances of pre-editing across three translation directions, two MT systems, and four text domains. We then analysed the instances from three perspectives: the characteristics of the pre-edited ST, the diversity of pre-editing operations, and the impact of the pre-editing operations on NMT outputs. Our findings include the following: (1) enhancing the explicitness of the meaning of an ST and its syntactic structure is more important for obtaining better translations than making the ST shorter and simpler, and (2) although the impact of pre-editing on NMT is generally unpredictable, there are some tendencies of changes in the NMT outputs depending on the editing operation types.