EMNLP 2020 Papers

Graph embedding (GE) methods embed nodes (and/or edges) in graph into a low-dimensional semantic space, and have shown its effectiveness in modeling multi-relational data. However, existing GE models are not practical in real-world applications since it overlooked the streaming nature of incoming data. To address this issue, we study the problem of continual graph representation learning which aims to continually train a GE model on new data to learn incessantly emerging multi-relational data while avoiding catastrophically forgetting old learned knowledge. Moreover, we propose a disentangle-based continual graph representation learning (DiCGRL) framework inspired by the human’s ability to learn procedural knowledge. The experimental results show that DiCGRL could effectively alleviate the catastrophic forgetting problem and outperform state-of-the-art continual learning models. The code and datasets are released on https://github.com/KXY-PUBLIC/DiCGRL.

One key principle for assessing textual similarity is measuring the degree of semantic overlap between texts by considering the word alignment. Such alignment-based approaches are both intuitive and interpretable; however, they are empirically inferior to the simple cosine similarity between general-purpose sentence vectors. We focus on the fact that the norm of word vectors is a good proxy for word importance, and the angle of them is a good proxy for word similarity. However, alignment-based approaches do not distinguish the norm and direction, whereas sentence-vector approaches automatically use the norm as the word importance. Accordingly, we propose decoupling word vectors into their norm and direction then computing the alignment-based similarity with the help of earth mover’s distance (optimal transport), which we refer to as word rotator’s distance. Furthermore, we demonstrate how to grow the norm and direction of word vectors (vector converter); this is a new systematic approach derived from the sentence-vector estimation methods, which can significantly improve the performance of the proposed method. On several STS benchmarks, the proposed methods outperform not only alignment-based approaches but also strong baselines. The source code is avaliable at https://github.com/eumesy/wrd

Few/Zero-shot learning is a big challenge of many classifications tasks, where a classifier is required to recognise instances of classes that have very few or even no training samples. It becomes more difficult in multi-label classification, where each instance is labelled with more than one class. In this paper, we present a simple multi-graph aggregation model that fuses knowledge from multiple label graphs encoding different semantic label relationships in order to study how the aggregated knowledge can benefit multi-label zero/few-shot document classification. The model utilises three kinds of semantic information, i.e., the pre-trained word embeddings, label description, and pre-defined label relations. Experimental results derived on two large clinical datasets (i.e., MIMIC-II and MIMIC-III ) and the EU legislation dataset show that methods equipped with the multi-graph knowledge aggregation achieve significant performance improvement across almost all the measures on few/zero-shot labels.

To scale non-parametric extensions of probabilistic topic models such as Latent Dirichlet allocation to larger data sets, practitioners rely increasingly on parallel and distributed systems. In this work, we study data-parallel training for the hierarchical Dirichlet process (HDP) topic model. Based upon a representation of certain conditional distributions within an HDP, we propose a doubly sparse data-parallel sampler for the HDP topic model. This sampler utilizes all available sources of sparsity found in natural language - an important way to make computation efficient. We benchmark our method on a well-known corpus (PubMed) with 8m documents and 768m tokens, using a single multi-core machine in under four days.

It is challenging to perform lifelong language learning (LLL) on a stream of different tasks without any performance degradation comparing to the multi-task counterparts. To address this issue, we present Lifelong Language Knowledge Distillation (L2KD), a simple but efficient method that can be easily applied to existing LLL architectures in order to mitigate the degradation. Specifically, when the LLL model is trained on a new task, we assign a teacher model to first learn the new task, and pass the knowledge to the LLL model via knowledge distillation. Therefore, the LLL model can better adapt to the new task while keeping the previously learned knowledge. Experiments show that the proposed L2KD consistently improves previous state-of-the-art models, and the degradation comparing to multi-task models in LLL tasks is well mitigated for both sequence generation and text classification tasks.

Bolukbasi et al. (2016) presents one of the first gender bias mitigation techniques for word embeddings. Their method takes pre-trained word embeddings as input and attempts to isolate a linear subspace that captures most of the gender bias in the embeddings. As judged by an analogical evaluation task, their method virtually eliminates gender bias in the embeddings. However, an implicit and untested assumption of their method is that the bias subspace is actually linear. In this work, we generalize their method to a kernelized, non-linear version. We take inspiration from kernel principal component analysis and derive a non-linear bias isolation technique. We discuss and overcome some of the practical drawbacks of our method for non-linear gender bias mitigation in word embeddings and analyze empirically whether the bias subspace is actually linear. Our analysis shows that gender bias is in fact well captured by a linear subspace, justifying the assumption of Bolukbasi et al. (2016).

Generating text from structured data is challenging because it requires bridging the gap between (i) structure and natural language (NL) and (ii) semantically underspecified input and fully specified NL output. Multilingual generation brings in an additional challenge: that of generating into languages with varied word order and morphological properties. In this work, we focus on Abstract Meaning Representations (AMRs) as structured input, where previous research has overwhelmingly focused on generating only into English. We leverage advances in cross-lingual embeddings, pretraining, and multilingual models to create multilingual AMR-to-text models that generate in twenty one different languages. Our multilingual models surpass baselines that generate into one language in eighteen languages, based on automatic metrics. We analyze the ability of our multilingual models to accurately capture morphology and word order using human evaluation, and find that native speakers judge our generations to be fluent.

Neural Natural Language Generation (NLG) systems are well known for their unreliability. To overcome this issue, we propose a data augmentation approach which allows us to restrict the output of a network and guarantee reliability. While this restriction means generation will be less diverse than if randomly sampled, we include experiments that demonstrate the tendency of existing neural generation approaches to produce dull and repetitive text, and we argue that reliability is more important than diversity for this task. The system trained using this approach scored 100% in semantic accuracy on the E2E NLG Challenge dataset, the same as a template system.

Punning is a creative way to make conversation enjoyable and literary writing elegant. In this paper, we focus on the task of generating a pun sentence given a pair of homophones. We first find the constraint words supporting the semantic incongruity for a sentence. Then we rewrite the sentence with explicit positive and negative constraints. Our model achieves the state-of-the-art results in both automatic and human evaluations. We further make an error analysis and discuss the challenges for the computational pun models.

A sequence-to-sequence (seq2seq) learning with neural networks empirically shows to be an effective framework for grammatical error correction (GEC), which takes a sentence with errors as input and outputs the corrected one. However, the performance of GEC models with the seq2seq framework heavily relies on the size and quality of the corpus on hand. We propose a method inspired by adversarial training to generate more meaningful and valuable training examples by continually identifying the weak spots of a model, and to enhance the model by gradually adding the generated adversarial examples to the training set. Extensive experimental results show that such adversarial training can improve both the generalization and robustness of GEC models.

Recent years the task of incomplete utterance rewriting has raised a large attention. Previous works usually shape it as a machine translation task and employ sequence to sequence based architecture with copy mechanism. In this paper, we present a novel and extensive approach, which formulates it as a semantic segmentation task. Instead of generating from scratch, such a formulation introduces edit operations and shapes the problem as prediction of a word-level edit matrix. Benefiting from being able to capture both local and global information, our approach achieves state-of-the-art performance on several public datasets. Furthermore, our approach is four times faster than the standard approach in inference.

Existing pre-trained large language models have shown unparalleled generative capabilities. However, they are not controllable. In this paper, we propose MEGATRON-CNTRL, a novel framework that uses large-scale language models and adds control to text generation by incorporating an external knowledge base. Our framework consists of a keyword predictor, a knowledge retriever, a contextual knowledge ranker, and a conditional text generator. As we do not have access to ground-truth supervision for the knowledge ranker, we make use of weak supervision from sentence embedding. The empirical results show that our model generates more fluent, consistent, and coherent stories with less repetition and higher diversity compared to prior work on the ROC story dataset. We showcase the controllability of our model by replacing the keywords used to generate stories and re-running the generation process. Human evaluation results show that 77.5% of these stories are successfully controlled by the new keywords. Furthermore, by scaling our model from 124 million to 8.3 billion parameters we demonstrate that larger models improve both the quality of generation (from 74.5% to 93.0% for consistency) and controllability (from 77.5% to 91.5%).

This paper proposes to adapt self-attention to discourse level for modeling discourse elements in argumentative student essays. Specifically, we focus on two issues. First, we propose structural sentence positional encodings to explicitly represent sentence positions. Second, we propose to use inter-sentence attentions to capture sentence interactions and enhance sentence representation. We conduct experiments on two datasets: a Chinese dataset and an English dataset. We find that (i) sentence positional encoding can lead to a large improvement for identifying discourse elements; (ii) a structural relative positional encoding of sentences shows to be most effective; (iii) inter-sentence attention vectors are useful as a kind of sentence representations for identifying discourse elements.

Discourse relations describe how two propositions relate to one another, and identifying them automatically is an integral part of natural language understanding. However, annotating discourse relations typically requires expert annotators. Recently, different semantic aspects of a sentence have been represented and crowd-sourced via question-and-answer (QA) pairs. This paper proposes a novel representation of discourse relations as QA pairs, which in turn allows us to crowd-source wide-coverage data annotated with discourse relations, via an intuitively appealing interface for composing such questions and answers. Based on our proposed representation, we collect a novel and wide-coverage QADiscourse dataset, and present baseline algorithms for predicting QADiscourse relations.

As different genres are known to differ in their communicative properties and as previously, for Chinese, discourse relations have only been annotated over news text, we have created the TED-CDB dataset. TED-CDB comprises a large set of TED talks in Chinese that have been manually annotated according to the goals and principles of Penn Discourse Treebank, but adapted to features that are not present in English. It serves as a unique Chinese corpus of spoken discourse. Benchmark experiments show that TED-CDB poses a challenge for state-of-the-art discourse relation classifiers, whose F1 performance on 4-way classification is 60%. This is a dramatic drop of 35% from performance on the news text in the Chinese Discourse Treebank. Transfer learning experiments have been carried out with the TED-CDB for both same-language cross-domain transfer and same-domain cross-language transfer. Both demonstrate that the TED-CDB can improve the performance of systems being developed for languages other than Chinese and would be helpful for insufficient or unbalanced data in other corpora. The dataset and our Chinese annotation guidelines will be made freely available.

This paper focuses on tree-based modeling for the sentence classification task. In existing works, aggregating on a syntax tree usually considers local information of sub-trees. In contrast, in addition to the local information, our proposed Modularized Syntactic Neural Network (MSNN) utilizes the syntax category labels and takes advantage of the global context while modeling sub-trees. In MSNN, each node of a syntax tree is modeled by a label-related syntax module. Each syntax module aggregates the outputs of lower-level modules, and finally, the root module provides the sentence representation. We design a tree-parallel mini-batch strategy for efficient training and predicting. Experimental results on four benchmark datasets show that our MSNN significantly outperforms previous state-of-the-art tree-based methods on the sentence classification task.

This paper reduces discontinuous parsing to sequence labeling. It first shows that existing reductions for constituent parsing as labeling do not support discontinuities. Second, it fills this gap and proposes to encode tree discontinuities as nearly ordered permutations of the input sequence. Third, it studies whether such discontinuous representations are learnable. The experiments show that despite the architectural simplicity, under the right representation, the models are fast and accurate.

Cross-language differences in (universal) dependency parsing performance are mostly attributed to treebank size, average sentence length, average dependency length, morphological complexity, and domain differences. We point at a factor not previously discussed: If we abstract away from words and dependency labels, how many graphs in the test data were seen in the training data? We compute graph isomorphisms, and show that, treebank size aside, overlap between training and test graphs explain more of the observed variation than standard explanations such as the above.

We introduce a novel chart-based algorithm for span-based parsing of discontinuous constituency trees of block degree two, including ill-nested structures. In particular, we show that we can build variants of our parser with smaller search spaces and time complexities ranging from O(nˆ6) down to O(nˆ3). The cubic time variant covers 98% of constituents observed in linguistic treebanks while having the same complexity as continuous constituency parsers. We evaluate our approach on German and English treebanks (Negra, Tiger, and DPTB) and report state-of-the-art results in the fully supervised setting. We also experiment with pre-trained word embeddings and Bert-based neural networks.