NAACL 2019 Papers

Most modern Information Extraction (IE) systems are implemented as sequential taggers and only model local dependencies. Non-local and non-sequential context is, however, a valuable source of information to improve predictions. In this paper, we introduce GraphIE, a framework that operates over a graph representing a broad set of dependencies between textual units (i.e. words or sentences). The algorithm propagates information between connected nodes through graph convolutions, generating a richer representation that can be exploited to improve word-level predictions. Evaluation on three different tasks — namely textual, social media and visual information extraction — shows that GraphIE consistently outperforms the state-of-the-art sequence tagging model by a significant margin.

This paper introduces improved methods for sub-event detection in social media streams, by applying neural sequence models not only on the level of individual posts, but also directly on the stream level. Current approaches to identify sub-events within a given event, such as a goal during a soccer match, essentially do not exploit the sequential nature of social media streams. We address this shortcoming by framing the sub-event detection problem in social media streams as a sequence labeling task and adopt a neural sequence architecture that explicitly accounts for the chronological order of posts. Specifically, we (i) establish a neural baseline that outperforms a graph-based state-of-the-art method for binary sub-event detection (2.7% micro-F1 improvement), as well as (ii) demonstrate superiority of a recurrent neural network model on the posts sequence level for labeled sub-events (2.4% bin-level F1 improvement over non-sequential models).

The goal of event detection (ED) is to detect the occurrences of events and categorize them. Previous work solved this task by recognizing and classifying event triggers, which is defined as the word or phrase that most clearly expresses an event occurrence. As a consequence, existing approaches required both annotated triggers and event types in training data. However, triggers are nonessential to event detection, and it is time-consuming for annotators to pick out the “most clearly” word from a given sentence, especially from a long sentence. The expensive annotation of training corpus limits the application of existing approaches. To reduce manual effort, we explore detecting events without triggers. In this work, we propose a novel framework dubbed as Type-aware Bias Neural Network with Attention Mechanisms (TBNNAM), which encodes the representation of a sentence based on target event types. Experimental results demonstrate the effectiveness. Remarkably, the proposed approach even achieves competitive performances compared with state-of-the-arts that used annotated triggers.

Supervised approaches to named entity recognition (NER) are largely developed based on the assumption that the training data is fully annotated with named entity information. However, in practice, annotated data can often be imperfect with one typical issue being the training data may contain incomplete annotations. We highlight several pitfalls associated with learning under such a setup in the context of NER and identify limitations associated with existing approaches, proposing a novel yet easy-to-implement approach for recognizing named entities with incomplete data annotations. We demonstrate the effectiveness of our approach through extensive experiments.

Contextual string embeddings are a recent type of contextualized word embedding that were shown to yield state-of-the-art results when utilized in a range of sequence labeling tasks. They are based on character-level language models which treat text as distributions over characters and are capable of generating embeddings for any string of characters within any textual context. However, such purely character-based approaches struggle to produce meaningful embeddings if a rare string is used in a underspecified context. To address this drawback, we propose a method in which we dynamically aggregate contextualized embeddings of each unique string that we encounter. We then use a pooling operation to distill a ”global” word representation from all contextualized instances. We evaluate these ”pooled contextualized embeddings” on common named entity recognition (NER) tasks such as CoNLL-03 and WNUT and show that our approach significantly improves the state-of-the-art for NER. We make all code and pre-trained models available to the research community for use and reproduction.

We recast dependency parsing as a sequence labeling problem, exploring several encodings of dependency trees as labels. While dependency parsing by means of sequence labeling had been attempted in existing work, results suggested that the technique was impractical. We show instead that with a conventional BILSTM-based model it is possible to obtain fast and accurate parsers. These parsers are conceptually simple, not needing traditional parsing algorithms or auxiliary structures. However, experiments on the PTB and a sample of UD treebanks show that they provide a good speed-accuracy tradeoff, with results competitive with more complex approaches.

We propose a novel transition-based algorithm that straightforwardly parses sentences from left to right by building n attachments, with n being the length of the input sentence. Similarly to the recent stack-pointer parser by Ma et al. (2018), we use the pointer network framework that, given a word, can directly point to a position from the sentence. However, our left-to-right approach is simpler than the original top-down stack-pointer parser (not requiring a stack) and reduces transition sequence length in half, from 2n-1 actions to n. This results in a quadratic non-projective parser that runs twice as fast as the original while achieving the best accuracy to date on the English PTB dataset (96.04% UAS, 94.43% LAS) among fully-supervised single-model dependency parsers, and improves over the former top-down transition system in the majority of languages tested.

We introduce a new syntax-aware model for dependency-based semantic role labeling that outperforms syntax-agnostic models for English and Spanish. We use a BiLSTM to tag the text with supertags extracted from dependency parses, and we feed these supertags, along with words and parts of speech, into a deep highway BiLSTM for semantic role labeling. Our model combines the strengths of earlier models that performed SRL on the basis of a full dependency parse with more recent models that use no syntactic information at all. Our local and non-ensemble model achieves state-of-the-art performance on the CoNLL 09 English and Spanish datasets. SRL models benefit from syntactic information, and we show that supertagging is a simple, powerful, and robust way to incorporate syntax into a neural SRL system.

Concept map-based multi-document summarization has recently been proposed as a variant of the traditional summarization task with graph-structured summaries. As shown by previous work, the grouping of coreferent concept mentions across documents is a crucial subtask of it. However, while the current state-of-the-art method suggested a new grouping method that was shown to improve the summary quality, its use of pairwise comparisons leads to polynomial runtime complexity that prohibits the application to large document collections. In this paper, we propose two alternative grouping techniques based on locality sensitive hashing, approximate nearest neighbor search and a fast clustering algorithm. They exhibit linear and log-linear runtime complexity, making them much more scalable. We report experimental results that confirm the improved runtime behavior while also showing that the quality of the summary concept maps remains comparable.

Serial recall experiments study the ability of humans to recall words in the order in which they occurred. The following serial recall effects are generally investigated in studies with humans: word length and frequency, primacy and recency, semantic confusion, repetition, and transposition effects. In this research, we investigate neural language models in the context of these serial recall effects. Our work provides a framework to better understand and analyze neural language models and opens a new window to develop accurate language models.

Conducting a manual evaluation is considered an essential part of summary evaluation methodology. Traditionally, the Pyramid protocol, which exhaustively compares system summaries to references, has been perceived as very reliable, providing objective scores. Yet, due to the high cost of the Pyramid method and the required expertise, researchers resorted to cheaper and less thorough manual evaluation methods, such as Responsiveness and pairwise comparison, attainable via crowdsourcing. We revisit the Pyramid approach, proposing a lightweight sampling-based version that is crowdsourcable. We analyze the performance of our method in comparison to original expert-based Pyramid evaluations, showing higher correlation relative to the common Responsiveness method. We release our crowdsourced Summary-Content-Units, along with all crowdsourcing scripts, for future evaluations.

Neural sequence-to-sequence models are currently the dominant approach in several natural language processing tasks, but require large parallel corpora. We present a sequence-to-sequence-to-sequence autoencoder (SEQˆ3), consisting of two chained encoder-decoder pairs, with words used as a sequence of discrete latent variables. We apply the proposed model to unsupervised abstractive sentence compression, where the first and last sequences are the input and reconstructed sentences, respectively, while the middle sequence is the compressed sentence. Constraining the length of the latent word sequences forces the model to distill important information from the input. A pretrained language model, acting as a prior over the latent sequences, encourages the compressed sentences to be human-readable. Continuous relaxations enable us to sample from categorical distributions, allowing gradient-based optimization, unlike alternatives that rely on reinforcement learning. The proposed model does not require parallel text-summary pairs, achieving promising results in unsupervised sentence compression on benchmark datasets.

Authors’ keyphrases assigned to scientific articles are essential for recognizing content and topic aspects. Most of the proposed supervised and unsupervised methods for keyphrase generation are unable to produce terms that are valuable but do not appear in the text. In this paper, we explore the possibility of considering the keyphrase string as an abstractive summary of the title and the abstract. First, we collect, process and release a large dataset of scientific paper metadata that contains 2.2 million records. Then we experiment with popular text summarization neural architectures. Despite using advanced deep learning models, large quantities of data and many days of computation, our systematic evaluation on four test datasets reveals that the explored text summarization methods could not produce better keyphrases than the simpler unsupervised methods, or the existing supervised ones.

The automatic detection of satire vs. regular news is relevant for downstream applications (for instance, knowledge base population) and to improve the understanding of linguistic characteristics of satire. Recent approaches build upon corpora which have been labeled automatically based on article sources. We hypothesize that this encourages the models to learn characteristics for different publication sources (e.g., “The Onion” vs. “The Guardian”) rather than characteristics of satire, leading to poor generalization performance to unseen publication sources. We therefore propose a novel model for satire detection with an adversarial component to control for the confounding variable of publication source. On a large novel data set collected from German news (which we make available to the research community), we observe comparable satire classification performance and, as desired, a considerable drop in publication classification performance with adversarial training. Our analysis shows that the adversarial component is crucial for the model to learn to pay attention to linguistic properties of satire.

Authorship verification is the problem of inferring whether two texts were written by the same author. For this task, unmasking is one of the most robust approaches as of today with the major shortcoming of only being applicable to book-length texts. In this paper, we present a generalized unmasking approach which allows for authorship verification of texts as short as four printed pages with very high precision at an adjustable recall tradeoff. Our generalized approach therefore reduces the required material by orders of magnitude, making unmasking applicable to authorship cases of more practical proportions. The new approach is on par with other state-of-the-art techniques that are optimized for texts of this length: it achieves accuracies of 75–80%, while also allowing for easy adjustment to forensic scenarios that require higher levels of confidence in the classification.

The development of a fictional plot is centered around characters who closely interact with each other forming dynamic social networks. In literature analysis, such networks have mostly been analyzed without particular relation types or focusing on roles which the characters take with respect to each other. We argue that an important aspect for the analysis of stories and their development is the emotion between characters. In this paper, we combine these aspects into a unified framework to classify emotional relationships of fictional characters. We formalize it as a new task and describe the annotation of a corpus, based on fan-fiction short stories. The extraction pipeline which we propose consists of character identification (which we treat as given by an oracle here) and the relation classification. For the latter, we provide results using several approaches previously proposed for relation identification with neural methods. The best result of 0.45 F1 is achieved with a GRU with character position indicators on the task of predicting undirected emotion relations in the associated social network graph.

Opinion role labeling (ORL) is an important task for fine-grained opinion mining, which identifies important opinion arguments such as holder and target for a given opinion trigger. The task is highly correlative with semantic role labeling (SRL), which identifies important semantic arguments such as agent and patient for a given predicate. As predicate agents and patients usually correspond to opinion holders and targets respectively, SRL could be valuable for ORL. In this work, we propose a simple and novel method to enhance ORL by utilizing SRL, presenting semantic-aware word representations which are learned from SRL. The representations are then fed into a baseline neural ORL model as basic inputs. We verify the proposed method on a benchmark MPQA corpus. Experimental results show that the proposed method is highly effective. In addition, we compare the method with two representative methods of SRL integration as well, finding that our method can outperform the two methods significantly, achieving 1.47% higher F-scores than the better one.

When assigning quantitative labels to a dataset, different methodologies may rely on different scales. In particular, when assigning polarities to words in a sentiment lexicon, annotators may use binary, categorical, or continuous labels. Naturally, it is of interest to unify these labels from disparate scales to both achieve maximal coverage over words and to create a single, more robust sentiment lexicon while retaining scale coherence. We introduce a generative model of sentiment lexica to combine disparate scales into a common latent representation. We realize this model with a novel multi-view variational autoencoder (VAE), called SentiVAE. We evaluate our approach via a downstream text classification task involving nine English-Language sentiment analysis datasets; our representation outperforms six individual sentiment lexica, as well as a straightforward combination thereof.

In this paper, we quantify, analyze and mitigate gender bias exhibited in ELMo’s contextualized word vectors. First, we conduct several intrinsic analyses and find that (1) training data for ELMo contains significantly more male than female entities, (2) the trained ELMo embeddings systematically encode gender information and (3) ELMo unequally encodes gender information about male and female entities. Then, we show that a state-of-the-art coreference system that depends on ELMo inherits its bias and demonstrates significant bias on the WinoBias probing corpus. Finally, we explore two methods to mitigate such gender bias and show that the bias demonstrated on WinoBias can be eliminated.