PyTorchKR Papers

최신 논문

총 1,841편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2007.05891

HyperGrid: Efficient Multi-Task Transformers with Grid-wise Decomposable Hyper Projections

Achieving state-of-the-art performance on natural language understanding tasks typically relies on fine-tuning a fresh model for every task. Consequently, this approach leads to a higher overall parameter cost, along with higher technical maintenance for serving multiple models. Learning a single multi-task model that is able to do well for all the tasks has been a challenging and yet attractive proposition. In this paper, we propose \textsc{HyperGrid}, a new approach for highly effective multi-task learning. The proposed approach is based on a decomposable hypernetwork that learns grid-wise projections that help to specialize regions in weight matrices for different tasks. In order to construct the proposed hypernetwork, our method learns the interactions and composition between a global (task-agnostic) state and a local task-specific state. We apply our proposed \textsc{HyperGrid} on the current state-of-the-art T5 model, demonstrating strong performance across the GLUE and SuperGLUE benchmarks when using only a single multi-task model. Our method helps bridge the gap between fine-tuning and multi-task learning approaches.

arXiv:2007.05044공식 코드 · Papers with Code

Advances of Transformer-Based Models for News Headline Generation

Pretrained language models based on Transformer architecture are the reason for recent breakthroughs in many areas of NLP, including sentiment analysis, question answering, named entity recognition. Headline generation is a special kind of text summarization task. Models need to have strong natural language understanding that goes beyond the meaning of individual words and sentences and an ability to distinguish essential information to succeed in it. In this paper, we fine-tune two pretrained Transformer-based models (mBART and BertSumAbs) for that task and achieve new state-of-the-art results on the RIA and Lenta datasets of Russian news. BertSumAbs increases ROUGE on average by 2.9 and 2.0 points respectively over previous best score achieved by Phrase-Based Attentional Transformer and CopyNet.

arXiv:2007.04248

Chatbot: A Conversational Agent employed with Named Entity Recognition Model using Artificial Neural Network

Chatbot is a technology that is used to mimic human behavior using natural language. There are different types of Chatbot that can be used as conversational agent in various business domains in order to increase the customer service and satisfaction. For any business domain, it requires a knowledge base to be built for that domain and design an information retrieval based system that can respond the user with a piece of documentation or generated sentences. The core component of a Chatbot is Natural Language Understanding (NLU) which has been impressively improved by deep learning methods. But we often lack such properly built NLU modules and requires more time to build it from scratch for high quality conversations. This may encourage fresh learners to build a Chatbot from scratch with simple architecture and using small dataset, although it may have reduced functionality, rather than building high quality data driven methods. This research focuses on Named Entity Recognition (NER) and Intent Classification models which can be integrated into NLU service of a Chatbot. Named entities will be inserted manually in the knowledge base and automatically detected in a given sentence. The NER model in the proposed architecture is based on artificial neural network which is trained on manually created entities and evaluated using CoNLL-2003 dataset.

arXiv:2007.02484

Logic, Language, and Calculus

The difference between object-language and metalanguage is crucial for logical analysis, but has yet not been examined for the field of computer science. In this paper the difference is examined with regard to inferential relations. It is argued that inferential relations in a metalanguage (like a calculus for propositional logic) cannot represent conceptual relations of natural language. Inferential relations govern our concept use and understanding. Several approaches in the field of Natural Language Understanding (NLU) and Natural Language Inference (NLI) take this insight in account, but do not consider, how an inference can be assessed as a good inference. I present a logical analysis that can assesss the normative dimension of inferences, which is a crucial part of logical understanding and goes beyond formal understanding of metalanguages.

arXiv:2007.01836

Pretrained Semantic Speech Embeddings for End-to-End Spoken Language Understanding via Cross-Modal Teacher-Student Learning

Spoken language understanding is typically based on pipeline architectures including speech recognition and natural language understanding steps. These components are optimized independently to allow usage of available data, but the overall system suffers from error propagation. In this paper, we propose a novel training method that enables pretrained contextual embeddings to process acoustic features. In particular, we extend it with an encoder of pretrained speech recognition systems in order to construct end-to-end spoken language understanding systems. Our proposed method is based on the teacher-student framework across speech and text modalities that aligns the acoustic and the semantic latent spaces. Experimental results in three benchmarks show that our system reaches the performance comparable to the pipeline architecture without using any training data and outperforms it after fine-tuning with ten examples per class on two out of three benchmarks.

arXiv:2007.00747

A Semantic Web Framework for Automated Smart Assistants: COVID-19 Case Study

COVID-19 pandemic elucidated that knowledge systems will be instrumental in cases where accurate information needs to be communicated to a substantial group of people with different backgrounds and technological resources. However, several challenges and obstacles hold back the wide adoption of virtual assistants by public health departments and organizations. This paper presents the Instant Expert, an open-source semantic web framework to build and integrate voice-enabled smart assistants (i.e. chatbots) for any web platform regardless of the underlying domain and technology. The component allows non-technical domain experts to effortlessly incorporate an operational assistant with voice recognition capability into their websites. Instant Expert is capable of automatically parsing, processing, and modeling Frequently Asked Questions pages as an information resource as well as communicating with an external knowledge engine for ontology-powered inference and dynamic data utilization. The presented framework utilizes advanced web technologies to ensure reusability and reliability, and an inference engine for natural language understanding powered by deep learning and heuristic algorithms. A use case for creating an informatory assistant for COVID-19 based on the Centers for Disease Control and Prevention (CDC) data is presented to demonstrate the framework's usage and benefits.

arXiv:2006.15595공식 코드 · Papers with Code

Rethinking Positional Encoding in Language Pre-training

In this work, we investigate the positional encoding methods used in language pre-training (e.g., BERT) and identify several problems in the existing formulations. First, we show that in the absolute positional encoding, the addition operation applied on positional embeddings and word embeddings brings mixed correlations between the two heterogeneous information resources. It may bring unnecessary randomness in the attention and further limit the expressiveness of the model. Second, we question whether treating the position of the symbol \texttt{[CLS]} the same as other words is a reasonable design, considering its special role (the representation of the entire sentence) in the downstream tasks. Motivated from above analysis, we propose a new positional encoding method called \textbf{T}ransformer with \textbf{U}ntied \textbf{P}ositional \textbf{E}ncoding (TUPE). In the self-attention module, TUPE computes the word contextual correlation and positional correlation separately with different parameterizations and then adds them together. This design removes the mixed and noisy correlations over heterogeneous embeddings and offers more expressiveness by using different projection matrices. Furthermore, TUPE unties the \texttt{[CLS]} symbol from other positions, making it easier to capture information from all positions. Extensive experiments and ablation studies on GLUE benchmark demonstrate the effectiveness of the proposed method. Codes and models are released at https://github.com/guolinke/TUPE.

arXiv:2006.14994공식 코드 · Papers with Code

ProVe -- Self-supervised pipeline for automated product replacement and cold-starting based on neural language models

In retail vertical industries, businesses are dealing with human limitation of quickly understanding and adapting to new purchasing behaviors. Moreover, retail businesses need to overcome the human limitation of properly managing a massive selection of products/brands/categories. These limitations lead to deficiencies from both commercial (e.g. loss of sales, decrease in customer satisfaction) and operational perspective (e.g. out-of-stock, over-stock). In this paper, we propose a pipeline approach based on Natural Language Understanding, for recommending the most suitable replacements for products that are out-of-stock. Moreover, we will propose a solution for managing products that were newly introduced in a retailer's portfolio with almost no transactional history. This solution will help businesses: automatically assign the new products to the right category; recommend complementary products for cross-sell from day 1; perform sales predictions even with almost no transactional history. Finally, the vector space model resulted by applying the pipeline presented in this paper is directly used as semantic information in deep learning-based demand forecasting solutions, leading to more accurate predictions. The whole research and experimentation process have been done using real-life private transactional data, however the source code is available on https://github.com/Lummetry/ProVe

arXiv:2006.14223

Neural Machine Translation For Paraphrase Generation

Training a spoken language understanding system, as the one in Alexa, typically requires a large human-annotated corpus of data. Manual annotations are expensive and time consuming. In Alexa Skill Kit (ASK) user experience with the skill greatly depends on the amount of data provided by skill developer. In this work, we present an automatic natural language generation system, capable of generating both human-like interactions and annotations by the means of paraphrasing. Our approach consists of machine translation (MT) inspired encoder-decoder deep recurrent neural network. We evaluate our model on the impact it has on ASK skill, intent, named entity classification accuracy and sentence level coverage, all of which demonstrate significant improvements for unseen skills on natural language understanding (NLU) models, trained on the data augmented with paraphrases.

arXiv:2006.13484

Accelerated Large Batch Optimization of BERT Pretraining in 54 minutes

BERT has recently attracted a lot of attention in natural language understanding (NLU) and achieved state-of-the-art results in various NLU tasks. However, its success requires large deep neural networks and huge amount of data, which result in long training time and impede development progress. Using stochastic gradient methods with large mini-batch has been advocated as an efficient tool to reduce the training time. Along this line of research, LAMB is a prominent example that reduces the training time of BERT from 3 days to 76 minutes on a TPUv3 Pod. In this paper, we propose an accelerated gradient method called LANS to improve the efficiency of using large mini-batches for training. As the learning rate is theoretically upper bounded by the inverse of the Lipschitz constant of the function, one cannot always reduce the number of optimization iterations by selecting a larger learning rate. In order to use larger mini-batch size without accuracy loss, we develop a new learning rate scheduler that overcomes the difficulty of using large learning rate. Using the proposed LANS method and the learning rate scheme, we scaled up the mini-batch sizes to 96K and 33K in phases 1 and 2 of BERT pretraining, respectively. It takes 54 minutes on 192 AWS EC2 P3dn.24xlarge instances to achieve a target F1 score of 90.5 or higher on SQuAD v1.1, achieving the fastest BERT training time in the cloud.

arXiv:2006.11527

Memory Transformer

Transformer-based models have achieved state-of-the-art results in many natural language processing tasks. The self-attention architecture allows transformer to combine information from all elements of a sequence into context-aware representations. However, information about the context is stored mostly in the same element-wise representations. This might limit the processing of properties related to the sequence as a whole more difficult. Adding trainable memory to selectively store local as well as global representations of a sequence is a promising direction to improve the Transformer model. Memory-augmented neural networks (MANNs) extend traditional neural architectures with general-purpose memory for representations. MANNs have demonstrated the capability to learn simple algorithms like Copy or Reverse and can be successfully trained via backpropagation on diverse tasks from question answering to language modeling outperforming RNNs and LSTMs of comparable complexity. In this work, we propose and study few extensions of the Transformer baseline (1) by adding memory tokens to store non-local representations, (2) creating memory bottleneck for the global information, (3) controlling memory update with dedicated layer. We evaluate these memory augmented Transformers and demonstrate that presence of memory positively correlates with the model performance for machine translation and language modelling tasks. Augmentation of pre-trained masked language model with memory tokens shows mixed results for tasks from GLUE benchmark. Visualization of attention patterns over the memory suggest that it improves the model's ability to process a global context.

arXiv:2006.11918공식 코드 · Papers with Code

MaxVA: Fast Adaptation of Step Sizes by Maximizing Observed Variance of Gradients

Adaptive gradient methods such as RMSProp and Adam use exponential moving estimate of the squared gradient to compute adaptive step sizes, achieving better convergence than SGD in face of noisy objectives. However, Adam can have undesirable convergence behaviors due to unstable or extreme adaptive learning rates. Methods such as AMSGrad and AdaBound have been proposed to stabilize the adaptive learning rates of Adam in the later stage of training, but they do not outperform Adam in some practical tasks such as training Transformers \cite{transformer}. In this paper, we propose an adaptive learning rate principle, in which the running mean of squared gradient in Adam is replaced by a weighted mean, with weights chosen to maximize the estimated variance of each coordinate. This results in a faster adaptation to the local gradient variance, which leads to more desirable empirical convergence behaviors than Adam. We prove the proposed algorithm converges under mild assumptions for nonconvex stochastic optimization problems, and demonstrate the improved efficacy of our adaptive averaging approach on machine translation, natural language understanding and large-batch pretraining of BERT. The code is available at https://github.com/zhuchen03/MaxVA.

arXiv:2006.10270

Multi-branch Attentive Transformer

While the multi-branch architecture is one of the key ingredients to the success of computer vision tasks, it has not been well investigated in natural language processing, especially sequence learning tasks. In this work, we propose a simple yet effective variant of Transformer called multi-branch attentive Transformer (briefly, MAT), where the attention layer is the average of multiple branches and each branch is an independent multi-head attention layer. We leverage two training techniques to regularize the training: drop-branch, which randomly drops individual branches during training, and proximal initialization, which uses a pre-trained Transformer model to initialize multiple branches. Experiments on machine translation, code generation and natural language understanding demonstrate that such a simple variant of Transformer brings significant improvements. Our code is available at \url{https://github.com/HA-Transformer}.

arXiv:2006.09479공식 코드 · Papers with Code

EPIE Dataset: A Corpus For Possible Idiomatic Expressions

Idiomatic expressions have always been a bottleneck for language comprehension and natural language understanding, specifically for tasks like Machine Translation(MT). MT systems predominantly produce literal translations of idiomatic expressions as they do not exhibit generic and linguistically deterministic patterns which can be exploited for comprehension of the non-compositional meaning of the expressions. These expressions occur in parallel corpora used for training, but due to the comparatively high occurrences of the constituent words of idiomatic expressions in literal context, the idiomatic meaning gets overpowered by the compositional meaning of the expression. State of the art Metaphor Detection Systems are able to detect non-compositional usage at word level but miss out on idiosyncratic phrasal idiomatic expressions. This creates a dire need for a dataset with a wider coverage and higher occurrence of commonly occurring idiomatic expressions, the spans of which can be used for Metaphor Detection. With this in mind, we present our English Possible Idiomatic Expressions(EPIE) corpus containing 25206 sentences labelled with lexical instances of 717 idiomatic expressions. These spans also cover literal usages for the given set of idiomatic expressions. We also present the utility of our dataset by using it to train a sequence labelling module and testing on three independent datasets with high accuracy, precision and recall scores.

arXiv:2006.08369공식 코드 · Papers with Code

EPIC30M: An Epidemics Corpus Of Over 30 Million Relevant Tweets

Since the start of COVID-19, several relevant corpora from various sources are presented in the literature that contain millions of data points. While these corpora are valuable in supporting many analyses on this specific pandemic, researchers require additional benchmark corpora that contain other epidemics to facilitate cross-epidemic pattern recognition and trend analysis tasks. During our other efforts on COVID-19 related work, we discover very little disease related corpora in the literature that are sizable and rich enough to support such cross-epidemic analysis tasks. In this paper, we present EPIC30M, a large-scale epidemic corpus that contains 30 millions micro-blog posts, i.e., tweets crawled from Twitter, from year 2006 to 2020. EPIC30M contains a subset of 26.2 millions tweets related to three general diseases, namely Ebola, Cholera and Swine Flu, and another subset of 4.7 millions tweets of six global epidemic outbreaks, including 2009 H1N1 Swine Flu, 2010 Haiti Cholera, 2012 Middle-East Respiratory Syndrome (MERS), 2013 West African Ebola, 2016 Yemen Cholera and 2018 Kivu Ebola. Furthermore, we explore and discuss the properties of the corpus with statistics of key terms and hashtags and trends analysis for each subset. Finally, we demonstrate the value and impact that EPIC30M could create through a discussion of multiple use cases of cross-epidemic research topics that attract growing interest in recent years. These use cases span multiple research areas, such as epidemiological modeling, pattern recognition, natural language understanding and economical modeling.

arXiv:2006.07499공식 코드 · Papers with Code

A Generative Model for Joint Natural Language Understanding and Generation

Natural language understanding (NLU) and natural language generation (NLG) are two fundamental and related tasks in building task-oriented dialogue systems with opposite objectives: NLU tackles the transformation from natural language to formal representations, whereas NLG does the reverse. A key to success in either task is parallel training data which is expensive to obtain at a large scale. In this work, we propose a generative model which couples NLU and NLG through a shared latent variable. This approach allows us to explore both spaces of natural language and formal representations, and facilitates information sharing through the latent space to eventually benefit NLU and NLG. Our model achieves state-of-the-art performance on two dialogue datasets with both flat and tree-structured formal representations. We also show that the model can be trained in a semi-supervised fashion by utilising unlabelled data to boost its performance.

arXiv:2006.06114

Consolidating Commonsense Knowledge

Commonsense reasoning is an important aspect of building robust AI systems and is receiving significant attention in the natural language understanding, computer vision, and knowledge graphs communities. At present, a number of valuable commonsense knowledge sources exist, with different foci, strengths, and weaknesses. In this paper, we list representative sources and their properties. Based on this survey, we propose principles and a representation model in order to consolidate them into a Common Sense Knowledge Graph (CSKG). We apply this approach to consolidate seven separate sources into a first integrated CSKG. We present statistics of CSKG, present initial investigations of its utility on four QA datasets, and list learned lessons.

arXiv:2006.05635

Data Augmentation for Training Dialog Models Robust to Speech Recognition Errors

Speech-based virtual assistants, such as Amazon Alexa, Google assistant, and Apple Siri, typically convert users' audio signals to text data through automatic speech recognition (ASR) and feed the text to downstream dialog models for natural language understanding and response generation. The ASR output is error-prone; however, the downstream dialog models are often trained on error-free text data, making them sensitive to ASR errors during inference time. To bridge the gap and make dialog models more robust to ASR errors, we leverage an ASR error simulator to inject noise into the error-free text data, and subsequently train the dialog models with the augmented data. Compared to other approaches for handling ASR errors, such as using ASR lattice or end-to-end methods, our data augmentation approach does not require any modification to the ASR or downstream dialog models; our approach also does not introduce any additional latency during inference time. We perform extensive experiments on benchmark data and show that our approach improves the performance of downstream dialog models in the presence of ASR errors, and it is particularly effective in the low-resource situations where there are constraints on model size or the training data is scarce.

논문 ID: output/1434940

Probabilistic Word Association for Dialogue Act Classification with Recurrent Neural Networks

The identification of Dialogue Act’s (DA) is an important aspect in determining the meaning of an utterance for many applications that require natural language understanding, and recent work using recurrent neural networks (RNN) has shown promising results when applied to the DA classification problem. This work presents a novel probabilistic method of utterance representation and describes a RNN sentence model for out-of-context DA Classification. The utterance representations are generated from keywords selected for their frequency association with certain DA’s. The proposed probabilistic representations are applied to the Switchboard DA corpus and performance is compared with pre-trained word embeddings using the same baseline RNN model. The results indicate that the probabilistic method achieves 75.48% overall accuracy and an improvement over the word embedding representations of 1.8%. This demonstrates the potential utility of using statistical utterance representations, that are able to capture word-DA relationships, for the purpose of DA classification.

arXiv:2006.05744공식 코드 · Papers with Code

MC-BERT: Efficient Language Pre-Training via a Meta Controller

Pre-trained contextual representations (e.g., BERT) have become the foundation to achieve state-of-the-art results on many NLP tasks. However, large-scale pre-training is computationally expensive. ELECTRA, an early attempt to accelerate pre-training, trains a discriminative model that predicts whether each input token was replaced by a generator. Our studies reveal that ELECTRA's success is mainly due to its reduced complexity of the pre-training task: the binary classification (replaced token detection) is more efficient to learn than the generation task (masked language modeling). However, such a simplified task is less semantically informative. To achieve better efficiency and effectiveness, we propose a novel meta-learning framework, MC-BERT. The pre-training task is a multi-choice cloze test with a reject option, where a meta controller network provides training input and candidates. Results over GLUE natural language understanding benchmark demonstrate that our proposed method is both efficient and effective: it outperforms baselines on GLUE semantic tasks given the same computational budget.