Contextualized word embeddings such as ELMo and BERT provide a foundation for strong performance across a range of natural language processing tasks, in part by pretraining on a large and topically-diverse corpus. However, the applicability of this approach is unknown when the target domain varies substantially from the text used during pretraining. Specifically, we are interested the scenario in which labeled data is available in only a canonical source domain such as newstext, and the target domain is distinct from both the labeled corpus and the pretraining data. To address this scenario, we propose domain-adaptive fine-tuning, in which the contextualized embeddings are adapted by masked language modeling on the target domain. We test this approach on the challenging domain of Early Modern English, which differs substantially from existing pretraining corpora. Domain-adaptive fine-tuning yields an improvement of 4\% in part-of-speech tagging accuracy over a BERT baseline, substantially improving on prior work on this task.
Character-level models have been used extensively in recent years in NLP tasks as both supplements and replacements for closed-vocabulary token-level word representations. In one popular architecture, character-level LSTMs are used to feed token representations into a sequence tagger predicting token-level annotations such as part-of-speech (POS) tags. In this work, we examine the behavior of POS taggers across languages from the perspective of individual hidden units within the character LSTM. We aggregate the behavior of these units into language-level metrics which quantify the challenges that taggers face on languages with different morphological properties, and identify links between synthesis and affixation preference and emergent behavior of the hidden tagger layer. In a comparative experiment, we show how modifying the balance between forward and backward hidden units affects model arrangement and performance in these types of languages.
We present a new architecture for storing and accessing entity mentions during online text processing. While reading the text, entity references are identified, and may be stored by either updating or overwriting a cell in a fixed-length memory. The update operation implies coreference with the other mentions that are stored in the same cell; the overwrite operations causes these mentions to be forgotten. By encoding the memory operations as differentiable gates, it is possible to train the model end-to-end, using both a supervised anaphora resolution objective as well as a supplementary language modeling objective. Evaluation on a dataset of pronoun-name anaphora demonstrates that the model achieves state-of-the-art performance with purely left-to-right processing of the text.
We consider the problem of making machine translation more robust to character-level variation at the source side, such as typos. Existing methods achieve greater coverage by applying subword models such as byte-pair encoding (BPE) and character-level encoders, but these methods are highly sensitive to spelling mistakes. We show how training on a mild amount of random synthetic noise can dramatically improve robustness to these variations, without diminishing performance on clean text. We focus on translation performance on natural noise, as captured by frequent corrections in Wikipedia edit logs, and show that robustness to such noise can be achieved using a balanced diet of simple synthetic noises at training time, without access to the natural noise data or distribution.
Wikipedia has a strong norm of writing in a 'neutral point of view' (NPOV). Articles that violate this norm are tagged, and editors are encouraged to make corrections. But the impact of this tagging system has not been quantitatively measured. Does NPOV tagging help articles to converge to the desired style? Do NPOV corrections encourage editors to adopt this style? We study these questions using a corpus of NPOV-tagged articles and a set of lexicons associated with biased language. An interrupted time series analysis shows that after an article is tagged for NPOV, there is a significant decrease in biased language in the article, as measured by several lexicons. However, for individual editors, NPOV corrections and talk page discussions yield no significant change in the usage of words in most of these lexicons, including Wikipedia's own list of 'words to watch.' This suggests that NPOV tagging and discussion does improve content, but has less success enculturating editors to the site's linguistic norms.
Semantic graphs, such as WordNet, are resources which curate natural language on two distinguishable layers. On the local level, individual relations between synsets (semantic building blocks) such as hypernymy and meronymy enhance our understanding of the words used to express their meanings. Globally, analysis of graph-theoretic properties of the entire net sheds light on the structure of human language as a whole. In this paper, we combine global and local properties of semantic graphs through the framework of Max-Margin Markov Graph Models (M3GM), a novel extension of Exponential Random Graph Model (ERGM) that scales to large multi-relational graphs. We demonstrate how such global modeling improves performance on the local task of predicting semantic relations between synsets, yielding new state-of-the-art results on the WN18RR dataset, a challenging version of WordNet link prediction in which "easy" reciprocal cases are removed. In addition, the M3GM model identifies multirelational motifs that are characteristic of well-formed lexical semantic ontologies.
Social media features substantial stylistic variation, raising new challenges for syntactic analysis of online writing. However, this variation is often aligned with author attributes such as age, gender, and geography, as well as more readily-available social network metadata. In this paper, we report new evidence on the link between language and social networks in the task of part-of-speech tagging. We find that tagger error rates are correlated with network structure, with high accuracy in some parts of the network, and lower accuracy elsewhere. As a result, tagger accuracy depends on training from a balanced sample of the network, rather than training on texts from a narrow subcommunity. We also describe our attempts to add robustness to stylistic variation, by building a mixture-of-experts model in which each expert is associated with a region of the social network. While prior work found that similar approaches yield performance improvements in sentiment analysis and entity linking, we were unable to obtain performance improvements in part-of-speech tagging, despite strong evidence for the link between part-of-speech error rates and social network structure.
Clinical notes are text documents that are created by clinicians for each patient encounter. They are typically accompanied by medical codes, which describe the diagnosis and treatment. Annotating these codes is labor intensive and error prone; furthermore, the connection between the codes and the text is not annotated, obscuring the reasons and details behind specific diagnoses and treatments. We present an attentional convolutional network that predicts medical codes from clinical text. Our method aggregates information across the document using a convolutional neural network, and uses an attention mechanism to select the most relevant segments for each of the thousands of possible codes. The method is accurate, achieving precision@8 of 0.71 and a Micro-F1 of 0.54, which are both better than the prior state of the art. Furthermore, through an interpretability evaluation by a physician, we show that the attention mechanism identifies meaningful explanations for each code assignment
Political identity is often manifested in language variation, but the relationship between the two is still relatively unexplored from a quantitative perspective. This study examines the use of Catalan, a language local to the semi-autonomous region of Catalonia in Spain, on Twitter in discourse related to the 2017 independence referendum. We corroborate prior findings that pro-independence tweets are more likely to include the local language than anti-independence tweets. We also find that Catalan is used more often in referendum-related discourse than in other contexts, contrary to prior findings on language variation. This suggests a strong role for the Catalan language in the expression of Catalonian political identity.
The global dynamics of event cascades are often governed by the local dynamics of peer influence. However, detecting social influence from observational data is challenging, due to confounds like homophily and practical issues like missing data. In this work, we propose a novel discriminative method to detect influence from observational data. The core of the approach is to train a ranking algorithm to predict the source of the next event in a cascade, and compare its out-of-sample accuracy against a competitive baseline which lacks access to features corresponding to social influence. Using synthetically generated data, we provide empirical evidence that this method correctly identifies influence in the presence of confounds, and is robust to both missing data and misspecification --- unlike popular alternatives. We also apply the method to two real-world datasets: (1) cascades of co-sponsorship of legislation in the U.S. House of Representatives, on a social network of shared campaign donors; (2) rumors about the Higgs boson discovery, on a follower network of $10^5$ Twitter accounts. Our model identifies the role of peer influence in these scenarios, and uses it to make more accurate predictions about the future trajectory of cascades.