Despite the excellent performance of black box approaches to modeling sentiment and emotion, lexica (sets of informative words and associated weights) that characterize different emotions are indispensable to the NLP community because they allow for interpretable and robust predictions. Emotion analysis of text is increasing in popularity in NLP; however, manually creating lexica for psychological constructs such as empathy has proven difficult. This paper automatically creates empathy word ratings from document-level ratings. The underlying problem of learning word ratings from higher-level supervision has to date only been addressed in an ad hoc fashion and is missing deep learning methods. We systematically compare a number of approaches to learning word ratings from higher-level supervision against a Mixed-Level Feed Forward Network (MLFFN), which we find performs best, and use the MLFFN to create the first-ever empathy lexicon. We then use Signed Spectral Clustering to gain insights into the resulting words.
Social media is increasingly used for large-scale population predictions, such as estimating community health statistics. However, social media users are not typically a representative sample of the intended population --- a "selection bias". Across five tasks for predicting US county population health statistics from Twitter, we explore standard restratification techniques --- bias mitigation approaches that reweight people-specific variables according to how under-sampled their socio-demographic groups are. We found standard restratification provided no improvement and often degraded population prediction accuracy. The core reason for this seemed to be both shrunken and sparse estimates of each population's socio-demographics for which we thus develop and evaluate three methods to address: predictive redistribution to account for shrinking, as well as adaptive binning and informed smoothing to handle sparse socio-demographic estimates. We show each of our methods can significantly improve over the standard restratification approaches. Combining approaches, we find substantial improvements over non-restratified models as well, yielding a 35.4% increase in variance explained for predicting surveyed life satisfaction, and an 10.0% average increase across all tasks.
Social media currently provide a window on our lives, making it possible to learn how people from different places, with different backgrounds, ages, and genders use language. In this work we exploit a newly-created Arabic dataset with ground truth age and gender labels to learn these attributes both individually and in a multi-task setting at the sentence level. Our models are based on variations of deep bidirectional neural networks. More specifically, we build models with gated recurrent units and bidirectional encoder representations from transformers (BERT). We show the utility of multi-task learning (MTL) on the two tasks and identify task-specific attention as a superior choice in this context. We also find that a single-task BERT model outperform our best MTL models on the two tasks. We report tweet-level accuracy of 51.43% for the age task (three-way) and 65.30% on the gender task (binary), both of which outperforms our baselines with a large margin. Our models are language-agnostic, and so can be applied to other languages.
Prediction of language varieties and dialects is an important language processing task, with a wide range of applications. For Arabic, the native tongue of ~ 300 million people, most varieties remain unsupported. To ease this bottleneck, we present a very large scale dataset covering 319 cities from all 21 Arab countries. We introduce a hierarchical attention multi-task learning (HA-MTL) approach for dialect identification exploiting our data at the city, state, and country levels. We also evaluate use of BERT on the three tasks, comparing it to the MTL approach. We benchmark and release our data and models.
Bias in word embeddings such as Word2Vec has been widely investigated, and many efforts made to remove such bias. We show how to use conceptors debiasing to post-process both traditional and contextualized word embeddings. Our conceptor debiasing can simultaneously remove racial and gender biases and, unlike standard debiasing methods, can make effect use of heterogeneous lists of biased words. We show that conceptor debiasing diminishes racial and gender bias of word representations as measured using the Word Embedding Association Test (WEAT) of Caliskan et al. (2017).
Distributed representations of sentences have become ubiquitous in natural language processing tasks. In this paper, we consider a continual learning scenario for sentence representations: Given a sequence of corpora, we aim to optimize the sentence encoder with respect to the new corpus while maintaining its accuracy on the old corpora. To address this problem, we propose to initialize sentence encoders with the help of corpus-independent features, and then sequentially update sentence encoders using Boolean operations of conceptor matrices to learn corpus-dependent features. We evaluate our approach on semantic textual similarity tasks and show that our proposed sentence encoder can continually learn features from new corpora while retaining its competence on previously encountered corpora.
Word vectors are at the core of many natural language processing tasks. Recently, there has been interest in post-processing word vectors to enrich their semantic information. In this paper, we introduce a novel word vector post-processing technique based on matrix conceptors (Jaeger2014), a family of regularized identity maps. More concretely, we propose to use conceptors to suppress those latent features of word vectors having high variances. The proposed method is purely unsupervised: it does not rely on any corpus or external linguistic database. We evaluate the post-processed word vectors on a battery of intrinsic lexical evaluation tasks, showing that the proposed method consistently outperforms existing state-of-the-art alternatives. We also show that post-processed word vectors can be used for the downstream natural language processing task of dialogue state tracking, yielding improved results in different dialogue domains.
A body of literature has demonstrated that users' mental health conditions, such as depression and anxiety, can be predicted from their social media language. There is still a gap in the scientific understanding of how psychological stress is expressed on social media. Stress is one of the primary underlying causes and correlates of chronic physical illnesses and mental health conditions. In this paper, we explore the language of psychological stress with a dataset of 601 social media users, who answered the Perceived Stress Scale questionnaire and also consented to share their Facebook and Twitter data. Firstly, we find that stressed users post about exhaustion, losing control, increased self-focus and physical pain as compared to posts about breakfast, family-time, and travel by users who are not stressed. Secondly, we find that Facebook language is more predictive of stress than Twitter language. Thirdly, we demonstrate how the language based models thus developed can be adapted and be scaled to measure county-level trends. Since county-level language is easily available on Twitter using the Streaming API, we explore multiple domain adaptation algorithms to adapt user-level Facebook models to Twitter language. We find that domain-adapted and scaled social media-based measurements of stress outperform sociodemographic variables (age, gender, race, education, and income), against ground-truth survey-based stress measurements, both at the user- and the county-level in the U.S. Twitter language that scores higher in stress is also predictive of poorer health, less access to facilities and lower socioeconomic status in counties. We conclude with a discussion of the implications of using social media as a new tool for monitoring stress levels of both individuals and counties.
Distributed representations of words, better known as word embeddings, have become important building blocks for natural language processing tasks. Numerous studies are devoted to transferring the success of unsupervised word embeddings to sentence embeddings. In this paper, we introduce a simple representation of sentences in which a sentence embedding is represented as a weighted average of word vectors followed by a soft projection. We demonstrate the effectiveness of this proposed method on the clinical semantic textual similarity task of the BioCreative/OHNLP Challenge 2018.
Deep Learning has drastically reshaped virtually all areas of NLP. Yet on the downside, it is commonly thought to be dependent on vast amounts of training data. As such, these techniques appear ill-suited for areas where annotated data is limited, like emotion analysis, with its many nuanced and hard-to-acquire annotation formats, or other low-data scenarios encountered in under-resourced languages. In contrast to this popular notion, we provide empirical evidence from three typologically diverse languages that today's favorite neural architectures can be trained on a few hundred observations only. Our results suggest that high-quality, pre-trained word embeddings are crucial for achieving high performance despite such strong data limitations.