One of the biggest challenges that prohibit the use of many current NLP methods in clinical settings is the availability of public datasets. In this work, we present MeDAL, a large medical text dataset curated for abbreviation disambiguation, designed for natural language understanding pre-training in the medical domain. We pre-trained several models of common architectures on this dataset and empirically showed that such pre-training leads to improved performance and convergence speed when fine-tuning on downstream medical tasks.
To quantify how well natural language understanding models can capture consistency in a general conversation, we introduce the DialoguE COntradiction DEtection task (DECODE) and a new conversational dataset containing both human-human and human-bot contradictory dialogues. We then compare a structured utterance-based approach of using pre-trained Transformer models for contradiction detection with the typical unstructured approach. Results reveal that: (i) our newly collected dataset is notably more effective at providing supervision for the dialogue contradiction detection task than existing NLI data including those aimed to cover the dialogue domain; (ii) the structured utterance-based approach is more robust and transferable on both analysis and out-of-distribution dialogues than its unstructured counterpart. We also show that our best contradiction detection model correlates well with human judgments and further provide evidence for its usage in both automatically evaluating and improving the consistency of state-of-the-art generative chatbots.
Off-the-shelf biomedical embeddings obtained from the recently released various pre-trained language models (such as BERT, XLNET) have demonstrated state-of-the-art results (in terms of accuracy) for the various natural language understanding tasks (NLU) in the biomedical domain. Relation Classification (RC) falls into one of the most critical tasks. In this paper, we explore how to incorporate domain knowledge of the biomedical entities (such as drug, disease, genes), obtained from Knowledge Graph (KG) Embeddings, for predicting Drug-Drug Interaction from textual corpus. We propose a new method, BERTKG-DDI, to combine drug embeddings obtained from its interaction with other biomedical entities along with domain-specific BioBERT embedding-based RC architecture. Experiments conducted on the DDIExtraction 2013 corpus clearly indicate that this strategy improves other baselines architectures by 4.1% macro F1-score.
Sources of commonsense knowledge support applications in natural language understanding, computer vision, and knowledge graphs. Given their complementarity, their integration is desired. Yet, their different foci, modeling approaches, and sparse overlap make integration difficult. In this paper, we consolidate commonsense knowledge by following five principles, which we apply to combine seven key sources into a first integrated CommonSense Knowledge Graph (CSKG). We analyze CSKG and its various text and graph embeddings, showing that CSKG is well-connected and that its embeddings provide a useful entry point to the graph. We demonstrate how CSKG can provide evidence for generalizable downstream reasoning and for pre-training of language models. CSKG and all its embeddings are made publicly available to support further research on commonsense knowledge integration and reasoning.
Voice Assistants such as Alexa, Siri, and Google Assistant typically use a two-stage Spoken Language Understanding pipeline; first, an Automatic Speech Recognition (ASR) component to process customer speech and generate text transcriptions, followed by a Natural Language Understanding (NLU) component to map transcriptions to an actionable hypothesis. An end-to-end (E2E) system that goes directly from speech to a hypothesis is a more attractive option. These systems were shown to be smaller, faster, and better optimized. However, they require massive amounts of end-to-end training data and in addition, don't take advantage of the already available ASR and NLU training data. In this work, we propose an E2E system that is designed to jointly train on multiple speech-to-text tasks, such as ASR (speech-transcription) and SLU (speech-hypothesis), and text-to-text tasks, such as NLU (text-hypothesis). We call this the Audio-Text All-Task (AT-AT) Model and we show that it beats the performance of E2E models trained on individual tasks, especially ones trained on limited data. We show this result on an internal music dataset and two public datasets, FluentSpeech and SNIPS Audio, where we achieve state-of-the-art results. Since our model can process both speech and text input sequences and learn to predict a target sequence, it also allows us to do zero-shot E2E SLU by training on only text-hypothesis data (without any speech) from a new domain. We evaluate this ability of our model on the Facebook TOP dataset and set a new benchmark for zeroshot E2E performance. We will soon release the audio data collected for the TOP dataset for future research.
The pre-training models such as BERT have achieved great results in various natural language processing problems. However, a large number of parameters need significant amounts of memory and the consumption of inference time, which makes it difficult to deploy them on edge devices. In this work, we propose a knowledge distillation method LRC-BERT based on contrastive learning to fit the output of the intermediate layer from the angular distance aspect, which is not considered by the existing distillation methods. Furthermore, we introduce a gradient perturbation-based training architecture in the training phase to increase the robustness of LRC-BERT, which is the first attempt in knowledge distillation. Additionally, in order to better capture the distribution characteristics of the intermediate layer, we design a two-stage training method for the total distillation loss. Finally, by verifying 8 datasets on the General Language Understanding Evaluation (GLUE) benchmark, the performance of the proposed LRC-BERT exceeds the existing state-of-the-art methods, which proves the effectiveness of our method.
Despite the progress made in recent years in addressing natural language understanding (NLU) challenges, the majority of this progress remains to be concentrated on resource-rich languages like English. This work focuses on Persian language, one of the widely spoken languages in the world, and yet there are few NLU datasets available for this rich language. The availability of high-quality evaluation datasets is a necessity for reliable assessment of the progress on different NLU tasks and domains. We introduce ParsiNLU, the first benchmark in Persian language that includes a range of high-level tasks -- Reading Comprehension, Textual Entailment, etc. These datasets are collected in a multitude of ways, often involving manual annotations by native speakers. This results in over 14.5$k$ new instances across 6 distinct NLU tasks. Besides, we present the first results on state-of-the-art monolingual and multi-lingual pre-trained language-models on this benchmark and compare them with human performance, which provides valuable insights into our ability to tackle natural language understanding challenges in Persian. We hope ParsiNLU fosters further research and advances in Persian language understanding.
This article contains a proposal to add coinduction to the computational apparatus of natural language understanding. This, we argue, will provide a basis for more realistic, computationally sound, and scalable models of natural language dialogue, syntax and semantics. Given that the bottom up, inductively constructed, semantic and syntactic structures are brittle, and seemingly incapable of adequately representing the meaning of longer sentences or realistic dialogues, natural language understanding is in need of a new foundation. Coinduction, which uses top down constraints, has been successfully used in the design of operating systems and programming languages. Moreover, implicitly it has been present in text mining, machine translation, and in some attempts to model intensionality and modalities, which provides evidence that it works. This article shows high level formalizations of some of such uses. Since coinduction and induction can coexist, they can provide a common language and a conceptual model for research in natural language understanding. In particular, such an opportunity seems to be emerging in research on compositionality. This article shows several examples of the joint appearance of induction and coinduction in natural language processing. We argue that the known individual limitations of induction and coinduction can be overcome in empirical settings by a combination of the the two methods. We see an open problem in providing a theory of their joint use.
For natural language processing systems, two kinds of evidence support the use of text representations from neural language models "pretrained" on large unannotated corpora: performance on application-inspired benchmarks (Peters et al., 2018, inter alia), and the emergence of syntactic abstractions in those representations (Tenney et al., 2019, inter alia). On the other hand, the lack of grounded supervision calls into question how well these representations can ever capture meaning (Bender and Koller, 2020). We apply novel probes to recent language models -- specifically focusing on predicate-argument structure as operationalized by semantic dependencies (Ivanova et al., 2012) -- and find that, unlike syntax, semantics is not brought to the surface by today's pretrained models. We then use convolutional graph encoders to explicitly incorporate semantic parses into task-specific finetuning, yielding benefits to natural language understanding (NLU) tasks in the GLUE benchmark. This approach demonstrates the potential for general-purpose (rather than task-specific) linguistic supervision, above and beyond conventional pretraining and finetuning. Several diagnostics help to localize the benefits of our approach.
Data sparsity is one of the key challenges associated with model development in Natural Language Understanding (NLU) for conversational agents. The challenge is made more complex by the demand for high quality annotated utterances commonly required for supervised learning, usually resulting in weeks of manual labor and high cost. In this paper, we present our results on boosting NLU model performance through training data augmentation using a sequential generative adversarial network (GAN). We explore data generation in the context of two tasks, the bootstrapping of a new language and the handling of low resource features. For both tasks we explore three sequential GAN architectures, one with a token-level reward function, another with our own implementation of a token-level Monte Carlo rollout reward, and a third with sentence-level reward. We evaluate the performance of these feedback models across several sampling methodologies and compare our results to upsampling the original data to the same scale. We further improve the GAN model performance through the transfer learning of the pretrained embeddings. Our experiments reveal synthetic data generated using the sequential generative adversarial network provides significant performance boosts across multiple metrics and can be a major benefit to the NLU tasks.
Metaphors are ubiquitous in human language. The metaphor detection task (MD) aims at detecting and interpreting metaphors from written language, which is crucial in natural language understanding (NLU) research. In this paper, we introduce a pre-trained Transformer-based model into MD. Our model outperforms the previous state-of-the-art models by large margins in our evaluations, with relative improvements on the F-1 score from 5.33% to 28.39%. Second, we extend MD to a classification task about the metaphoricity of an entire piece of text to make MD applicable in more general NLU scenes. Finally, we clean up the improper or outdated annotations in one of the MD benchmark datasets and re-benchmark it with our Transformer-based model. This approach could be applied to other existing MD datasets as well, since the metaphoricity annotations in these benchmark datasets may be outdated. Future research efforts are also necessary to build an up-to-date and well-annotated dataset consisting of longer and more complex texts.
Current voice assistants typically use the best hypothesis yielded by their Automatic Speech Recognition (ASR) module as input to their Natural Language Understanding (NLU) module, thereby losing helpful information that might be stored in lower-ranked ASR hypotheses. We explore the change in performance of NLU associated tasks when utilizing five-best ASR hypotheses when compared to status quo for two language datasets, German and Portuguese. To harvest information from the ASR five-best, we leverage extractive summarization and joint extractive-abstractive summarization models for Domain Classification (DC) experiments while using a sequence-to-sequence model with a pointer generator network for Intent Classification (IC) and Named Entity Recognition (NER) multi-task experiments. For the DC full test set, we observe significant improvements of up to 7.2% and 15.5% in micro-averaged F1 scores, for German and Portuguese, respectively. In cases where the best ASR hypothesis was not an exact match to the transcribed utterance (mismatched test set), we see improvements of up to 6.7% and 8.8% micro-averaged F1 scores, for German and Portuguese, respectively. For IC and NER multi-task experiments, when evaluating on the mismatched test set, we see improvements across all domains in German and in 17 out of 19 domains in Portuguese (improvements based on change in SeMER scores). Our results suggest that the use of multiple ASR hypotheses, as opposed to one, can lead to significant performance improvements in the DC task for these non-English datasets. In addition, it could lead to significant improvement in the performance of IC and NER tasks in cases where the ASR model makes mistakes.
With the recent explosion in popularity of voice assistant devices, there is a growing interest in making them available to user populations in additional countries and languages. However, to provide the highest accuracy and best performance for specific user populations, most existing voice assistant models are developed individually for each region or language, which requires linear investment of effort. In this paper, we propose a general multilingual model framework for Natural Language Understanding (NLU) models, which can help bootstrap new language models faster and reduce the amount of effort required to develop each language separately. We explore how different deep learning architectures affect multilingual NLU model performance. Our experimental results show that these multilingual models can reach same or better performance compared to monolingual models across language-specific test data while require less effort in creating features and model maintenance.
We present a neural model for paraphrasing and train it to generate delexicalized sentences. We achieve this by creating training data in which each input is paired with a number of reference paraphrases. These sets of reference paraphrases represent a weak type of semantic equivalence based on annotated slots and intents. To understand semantics from different types of slots, other than anonymizing slots, we apply convolutional neural networks (CNN) prior to pooling on slot values and use pointers to locate slots in the output. We show empirically that the generated paraphrases are of high quality, leading to an additional 1.29% exact match on live utterances. We also show that natural language understanding (NLU) tasks, such as intent classification and named entity recognition, can benefit from data augmentation using automatically generated paraphrases.
Chatbots are envisioned to dramatically change the future of Software Engineering, allowing practitioners to chat and inquire about their software projects and interact with different services using natural language. At the heart of every chatbot is a Natural Language Understanding (NLU) component that enables the chatbot to understand natural language input. Recently, many NLU platforms were provided to serve as an off-the-shelf NLU component for chatbots, however, selecting the best NLU for Software Engineering chatbots remains an open challenge. Therefore, in this paper, we evaluate four of the most commonly used NLUs, namely IBM Watson, Google Dialogflow, Rasa, and Microsoft LUIS to shed light on which NLU should be used in Software Engineering based chatbots. Specifically, we examine the NLUs' performance in classifying intents, confidence scores stability, and extracting entities. To evaluate the NLUs, we use two datasets that reflect two common tasks performed by Software Engineering practitioners, 1) the task of chatting with the chatbot to ask questions about software repositories 2) the task of asking development questions on Q&A forums (e.g., Stack Overflow). According to our findings, IBM Watson is the best performing NLU when considering the three aspects (intents classification, confidence scores, and entity extraction). However, the results from each individual aspect show that, in intents classification, IBM Watson performs the best with an F1-measure > 84%, but in confidence scores, Rasa comes on top with a median confidence score higher than 0.91. Our results also show that all NLUs, except for Dialogflow, generally provide trustable confidence scores. For entity extraction, Microsoft LUIS and IBM Watson outperform other NLUs in the two SE tasks. Our results provide guidance to software engineering practitioners when deciding which NLU to use in their chatbots.
Recently, leveraging pre-trained Transformer based language models in down stream, task specific models has advanced state of the art results in natural language understanding tasks. However, only a little research has explored the suitability of this approach in low resource settings with less than 1,000 training data points. In this work, we explore fine-tuning methods of BERT -- a pre-trained Transformer based language model -- by utilizing pool-based active learning to speed up training while keeping the cost of labeling new data constant. Our experimental results on the GLUE data set show an advantage in model performance by maximizing the approximate knowledge gain of the model when querying from the pool of unlabeled data. Finally, we demonstrate and analyze the benefits of freezing layers of the language model during fine-tuning to reduce the number of trainable parameters, making it more suitable for low-resource settings.
Universal Turing Machines [29, 10, 18] are well known in computer science but they are about manual programming for general purposes. Although human children perform conscious learning (i.e., learning while being conscious) from infancy [24, 23, 14, 4], it is unknown that Universal Turing Machiness can facilitate not only our understanding of Autonomous Programming For General Purposes (APFGP) by machines, but also enable early-age conscious learning. This work reports a new kind of AI---conscious learning AI from a machine's "baby" time. Instead of arguing what static tasks a conscious machine should be able to do during its "adulthood", this work suggests that APFGP is a computationally clearer and necessary criterion for us to judge whether a machine is capable of conscious learning so that it can autonomously acquire skills along its "career path". The results here report new concepts and experimental studies for early vision, audition, natural language understanding, and emotion, with conscious learning capabilities that are absent from traditional AI systems.
Spoken language understanding (SLU) systems are widely used in handling of customer-care calls.A traditional SLU system consists of an acoustic model (AM) and a language model (LM) that areused to decode the utterance and a natural language understanding (NLU) model that predicts theintent. While AM can be shared across different domains, LM and NLU models need to be trainedspecifically for every new task. However, preparing enough data to train these models is prohibitivelyexpensive. In this paper, we introduce an efficient method to expand the limited in-domain data. Theprocess starts with training a preliminary NLU model based on logistic regression on the in-domaindata. Since the features are based onn= 1,2-grams, we can detect the most informative n-gramsfor each intent class. Using these n-grams, we find the samples in the out-of-domain corpus that1) contain the desired n-gram and/or 2) have similar intent label. The ones which meet the firstconstraint are used to train a new LM model and the ones that meet both constraints are used to train anew NLU model. Our results on two divergent experimental setups show that the proposed approachreduces by 30% the absolute classification error rate (CER) comparing to the preliminary modelsand it significantly outperforms the traditional data expansion algorithms such as the ones based onsemi-supervised learning, TF-IDF and embedding vectors.
In this paper, we propose and experiment with techniques for extreme compression of neural natural language understanding (NLU) models, making them suitable for execution on resource-constrained devices. We propose a task-aware, end-to-end compression approach that performs word-embedding compression jointly with NLU task learning. We show our results on a large-scale, commercial NLU system trained on a varied set of intents with huge vocabulary sizes. Our approach outperforms a range of baselines and achieves a compression rate of 97.4% with less than 3.7% degradation in predictive performance. Our analysis indicates that the signal from the downstream task is important for effective compression with minimal degradation in performance.
We introduce CLIMATE-FEVER, a new publicly available dataset for verification of climate change-related claims. By providing a dataset for the research community, we aim to facilitate and encourage work on improving algorithms for retrieving evidential support for climate-specific claims, addressing the underlying language understanding challenges, and ultimately help alleviate the impact of misinformation on climate change. We adapt the methodology of FEVER [1], the largest dataset of artificially designed claims, to real-life claims collected from the Internet. While during this process, we could rely on the expertise of renowned climate scientists, it turned out to be no easy task. We discuss the surprising, subtle complexity of modeling real-world climate-related claims within the \textsc{fever} framework, which we believe provides a valuable challenge for general natural language understanding. We hope that our work will mark the beginning of a new exciting long-term joint effort by the climate science and AI community.