EMNLP 2019 Papers

Vector averaging remains one of the most popular sentence embedding methods in spite of its obvious disregard for syntactic structure. While more complex sequential or convolutional networks potentially yield superior classification performance, the improvements in classification accuracy are typically mediocre compared to the simple vector averaging. As an efficient alternative, we propose the use of discrete cosine transform (DCT) to compress word sequences in an order-preserving manner. The lower order DCT coefficients represent the overall feature patterns in sentences, which results in suitable embeddings for tasks that could benefit from syntactic features. Our results in semantic probing tasks demonstrate that DCT embeddings indeed preserve more syntactic information compared with vector averaging. With practically equivalent complexity, the model yields better overall performance in downstream classification tasks that correlate with syntactic features, which illustrates the capacity of DCT to preserve word order information.

In transductive learning, an unlabeled test set is used for model training. Although this setting deviates from the common assumption of a completely unseen test set, it is applicable in many real-world scenarios, wherein the texts to be processed are known in advance. However, despite its practical advantages, transductive learning is underexplored in natural language processing. Here we conduct an empirical study of transductive learning for neural models and demonstrate its utility in syntactic and semantic tasks. Specifically, we fine-tune language models (LMs) on an unlabeled test set to obtain test-set-specific word representations. Through extensive experiments, we demonstrate that despite its simplicity, transductive LM fine-tuning consistently improves state-of-the-art neural models in in-domain and out-of-domain settings.

State-of-the-art semantic parsers rely on auto-regressive decoding, emitting one symbol at a time. When tested against complex databases that are unobserved at training time (zero-shot), the parser often struggles to select the correct set of database constants in the new database, due to the local nature of decoding. %since their decisions are based on weak, local information only. In this work, we propose a semantic parser that globally reasons about the structure of the output query to make a more contextually-informed selection of database constants. We use message-passing through a graph neural network to softly select a subset of database constants for the output query, conditioned on the question. Moreover, we train a model to rank queries based on the global alignment of database constants to question words. We apply our techniques to the current state-of-the-art model for Spider, a zero-shot semantic parsing dataset with complex databases, increasing accuracy from 39.4% to 47.4%.

The task of semantic parsing is highly useful for dialogue and question answering systems. Many datasets have been proposed to map natural language text into SQL, among which the recent Spider dataset provides cross-domain samples with multiple tables and complex queries. We build a Spider dataset for Chinese, which is currently a low-resource language in this task area. Interesting research questions arise from the uniqueness of the language, which requires word segmentation, and also from the fact that SQL keywords and columns of DB tables are typically written in English. We compare character- and word-based encoders for a semantic parser, and different embedding schemes. Results show that word-based semantic parser is subject to segmentation errors and cross-lingual word embeddings are useful for text-to-SQL.

We present PaLM, a hybrid parser and neural language model. Building on an RNN language model, PaLM adds an attention layer over text spans in the left context. An unsupervised constituency parser can be derived from its attention weights, using a greedy decoding algorithm. We evaluate PaLM on language modeling, and empirically show that it outperforms strong baselines. If syntactic annotations are available, the attention component can be trained in a supervised manner, providing syntactically-informed representations of the context, and further improving language modeling performance.

Spoken Language Understanding (SLU) converts user utterances into structured semantic representations. Data sparsity is one of the main obstacles of SLU due to the high cost of human annotation, especially when domain changes or a new domain comes. In this work, we propose a data augmentation method with atomic templates for SLU, which involves minimum human efforts. The atomic templates produce exemplars for fine-grained constituents of semantic representations. We propose an encoder-decoder model to generate the whole utterance from atomic exemplars. Moreover, the generator could be transferred from source domains to help a new domain which has little data. Experimental results show that our method achieves significant improvements on DSTC 2&3 dataset which is a domain adaptation setting of SLU.

We propose a practical scheme to train a single multilingual sequence labeling model that yields state of the art results and is small and fast enough to run on a single CPU. Starting from a public multilingual BERT checkpoint, our final model is 6x smaller and 27x faster, and has higher accuracy than a state-of-the-art multilingual baseline. We show that our model especially outperforms on low-resource languages, and works on codemixed input text without being explicitly trained on codemixed examples. We showcase the effectiveness of our method by reporting on part-of-speech tagging and morphological prediction on 70 treebanks and 48 languages.

One way to reduce network traffic in multi-node data-parallel stochastic gradient descent is to only exchange the largest gradients. However, doing so damages the gradient and degrades the model’s performance. Transformer models degrade dramatically while the impact on RNNs is smaller. We restore gradient quality by combining the compressed global gradient with the node’s locally computed uncompressed gradient. Neural machine translation experiments show that Transformer convergence is restored while RNNs converge faster. With our method, training on 4 nodes converges up to 1.5x as fast as with uncompressed gradients and scales 3.5x relative to single-node training.

Much previous work has been done in attempting to identify humor in text. In this paper we extend that capability by proposing a new task: assessing whether or not a joke is humorous. We present a novel way of approaching this problem by building a model that learns to identify humorous jokes based on ratings gleaned from Reddit pages, consisting of almost 16,000 labeled instances. Using these ratings to determine the level of humor, we then employ a Transformer architecture for its advantages in learning from sentence context. We demonstrate the effectiveness of this approach and show results that are comparable to human performance. We further demonstrate our model’s increased capabilities on humor identification problems, such as the previously created datasets for short jokes and puns. These experiments show that this method outperforms all previous work done on these tasks, with an F-measure of 93.1% for the Puns dataset and 98.6% on the Short Jokes dataset.

Obtaining large-scale annotated data for NLP tasks in the scientific domain is challenging and expensive. We release SciBERT, a pretrained language model based on BERT (Devlin et. al., 2018) to address the lack of high-quality, large-scale labeled scientific data. SciBERT leverages unsupervised pretraining on a large multi-domain corpus of scientific publications to improve performance on downstream scientific NLP tasks. We evaluate on a suite of tasks including sequence tagging, sentence classification and dependency parsing, with datasets from a variety of scientific domains. We demonstrate statistically significant improvements over BERT and achieve new state-of-the-art results on several of these tasks. The code and pretrained models are available at https://github.com/allenai/scibert/.

When trained effectively, the Variational Autoencoder (VAE) is both a powerful language model and an effective representation learning framework. In practice, however, VAEs are trained with the evidence lower bound (ELBO) as a surrogate objective to the intractable marginal data likelihood. This approach to training yields unstable results, frequently leading to a disastrous local optimum known as posterior collapse. In this paper, we investigate a simple fix for posterior collapse which yields surprisingly effective results. The combination of two known heuristics, previously considered only in isolation, substantially improves held-out likelihood, reconstruction, and latent representation learning when compared with previous state-of-the-art methods. More interestingly, while our experiments demonstrate superiority on these principle evaluations, our method obtains a worse ELBO. We use these results to argue that the typical surrogate objective for VAEs may not be sufficient or necessarily appropriate for balancing the goals of representation learning and data distribution modeling.

In this paper, we present a fast and reliable method based on PCA to select the number of dimensions for word embeddings. First, we train one embedding with a generous upper bound (e.g. 1,000) of dimensions. Then we transform the embeddings using PCA and incrementally remove the lesser dimensions one at a time while recording the embeddings’ performance on language tasks. Lastly, we select the number of dimensions, balancing model size and accuracy. Experiments using various datasets and language tasks demonstrate that we are able to train about 10 times fewer sets of embeddings while retaining optimal performance. Researchers interested in training the best-performing embeddings for downstream tasks, such as sentiment analysis, question answering and hypernym extraction, as well as those interested in embedding compression should find the method helpful.

Bilinear models such as DistMult and ComplEx are effective methods for knowledge graph (KG) completion. However, they require large batch sizes, which becomes a performance bottleneck when training on large scale datasets due to memory constraints. In this paper we use occurrences of entity-relation pairs in the dataset to construct a joint learning model and to increase the quality of sampled negatives during training. We show on three standard datasets that when these two techniques are combined, they give a significant improvement in performance, especially when the batch size and the number of generated negative examples are low relative to the size of the dataset. We then apply our techniques to a dataset containing 2 million entities and demonstrate that our model outperforms the baseline by 2.8% absolute on hits@1.

In this paper, we study differentiable neural architecture search (NAS) methods for natural language processing. In particular, we improve differentiable architecture search by removing the softmax-local constraint. Also, we apply differentiable NAS to named entity recognition (NER). It is the first time that differentiable NAS methods are adopted in NLP tasks other than language modeling. On both the PTB language modeling and CoNLL-2003 English NER data, our method outperforms strong baselines. It achieves a new state-of-the-art on the NER task.

The objective of non-parallel text style transfer, or controllable text generation, is to alter specific attributes (e.g. sentiment, mood, tense, politeness, etc) of a given text while preserving its remaining attributes and content. Generative adversarial network (GAN) is a popular model to ensure the transferred sentences are realistic and have the desired target styles. However, training GAN often suffers from mode collapse problem, which causes that the transferred text is little related to the original text. In this paper, we propose a new GAN model with a word-level conditional architecture and a two-phase training procedure. By using a style-related condition architecture before generating a word, our model is able to maintain style-unrelated words while changing the others. By separating the training procedure into reconstruction and transfer phases, our model is able to learn a proper text generation process, which further improves the content preservation. We test our model on polarity sentiment transfer and multiple-attribute transfer tasks. The empirical results show that our model achieves comparable evaluation scores in both transfer accuracy and fluency but significantly outperforms other state-of-the-art models in content compatibility on three real-world datasets.

Formality text style transfer plays an important role in various NLP applications, such as non-native speaker assistants and child education. Early studies normalize informal sentences with rules, before statistical and neural models become a prevailing method in the field. While a rule-based system is still a common preprocessing step for formality style transfer in the neural era, it could introduce noise if we use the rules in a naive way such as data preprocessing. To mitigate this problem, we study how to harness rules into a state-of-the-art neural network that is typically pretrained on massive corpora. We propose three fine-tuning methods in this paper and achieve a new state-of-the-art on benchmark datasets

Out-of-domain (OOD) detection for low-resource text classification is a realistic but understudied task. The goal is to detect the OOD cases with limited in-domain (ID) training data, since in machine learning applications we observe that training data is often insufficient. In this work, we propose an OOD-resistant Prototypical Network to tackle this zero-shot OOD detection and few-shot ID classification task. Evaluations on real-world datasets show that the proposed solution outperforms state-of-the-art methods in zero-shot OOD detection task, while maintaining a competitive performance on ID classification task.

In this paper we present a novel approach to simultaneously representing multiple languages in a common space. Procrustes Analysis (PA) is commonly used to find the optimal orthogonal word mapping in the bilingual case. The proposed Multi Pairwise Procrustes Analysis (MPPA) is a natural extension of the PA algorithm to multilingual word mapping. Unlike previous PA extensions that require a k-way dictionary, this approach requires only pairwise bilingual dictionaries that are much easier to construct.

In low-resource settings, the performance of supervised labeling models can be improved with automatically annotated or distantly supervised data, which is cheap to create but often noisy. Previous works have shown that significant improvements can be reached by injecting information about the confusion between clean and noisy labels in this additional training data into the classifier training. However, for noise estimation, these approaches either do not take the input features (in our case word embeddings) into account, or they need to learn the noise modeling from scratch which can be difficult in a low-resource setting. We propose to cluster the training data using the input features and then compute different confusion matrices for each cluster. To the best of our knowledge, our approach is the first to leverage feature-dependent noise modeling with pre-initialized confusion matrices. We evaluate on low-resource named entity recognition settings in several languages, showing that our methods improve upon other confusion-matrix based methods by up to 9%.