Humanitarian aid missions, whether emergency famine relief, establishment of medical clinics, or missions in conjunction with peace-keeping operations, require on-demand communication with the indigenous population. If such operations take place in countries with a commonly-spoken major language, such as English or Spanish, it proves relatively easy to find participating personnel with the appropriate linguistic fluency. However, such is not the case when the operations take place in regions where less common languages are spoken, such as Bosnia (language: Serbo-Croatian), Haiti (language: Haitian Creole), Somalia (language: Somali, a.k.a. “Soomaaliga”), or Afghanistan (language: Pashto, with subpopulations of Urdu and Tadjik speakers). Even in Latin America, where Spanish and Portuguese dominate, there are over 100 indigenous languages, including Quechua in Peru, Aymara in Bolivia, Mapudungun in Southern Chile, and the Tucan languages in the Southern Colombian Putumayo region. Many native speakers of these languages are not versant in either Spanish or Portuguese, especially those in remote mountainous or jungle regions, where the need for medical or educational aid, or protection from organized drug gangs, may be paramount.

Abstractive text summarization aims at compressing the information of a long source document into a rephrased, condensed summary. Despite advances in modeling techniques, abstractive summarization models still suffer from several key challenges: (i) layout bias: they overfit to the style of training corpora; (ii) limited abstractiveness: they are optimized to copying n-grams from the source rather than generating novel abstractive summaries; (iii) lack of transparency: they are not interpretable. In this work, we propose a framework based on document-level structure induction for summarization to address these challenges. To this end, we propose incorporating latent and explicit dependencies across sentences in the source document into end-to-end single-document summarization models. Our framework complements standard encoder-decoder summarization models by augmenting them with rich structure-aware document representations based on implicitly learned (latent) structures and externally-derived linguistic (explicit) structures. We show that our summarization framework, trained on the CNN/DM dataset, improves the coverage of content in the source documents, generates more abstractive summaries by generating more novel n-grams, and incorporates interpretable sentence-level structures, while performing on par with standard baselines.

Unsupervised domain adaptation, as a prevalent transfer learning setting, spans many real-world applications. With the increasing representational power and applicability of neural networks, state-of-the-art domain adaptation methods make use of deep architectures to map the input features X to a latent representation Z that has the same marginal distribution across domains. This has been shown to be insufficient for generating optimal representation for classification, and to find conditionally invariant representations, usually strong assumptions are needed. We provide reasoning why when the supports of the source and target data from overlap, any map of X that is fixed across domains may not be suitable for domain adaptation via invariant features. Furthermore, we develop an efficient technique in which the optimal map from X to Z also takes domain-specific information as input, in addition to the features X . By using the property of minimal changes of causal mechanisms across domains, our model also takes into account the domain-specific information to ensure that the latent representation Z does not discard valuable information about Y . We demonstrate the efficacy of our method via synthetic and real-world data experiments. The code is available at: https://github.com/DMIRLAB-Group/DSAN.

Current natural language processing models work well on a single task, yet they often fail to continuously learn new tasks without forgetting previous ones as they are re-trained throughout their lifetime, a challenge known as lifelong learning. State-of-the-art lifelong language learning methods store past examples in episodic memory and replay them at both training and inference time. However, as we show later in our experiments, there are three significant impediments: (1) needing unrealistically large memory module to achieve good performance, (2) suffering from negative transfer, (3) requiring multiple local adaptation steps for each test example that significantly slows down the inference speed. In this paper, we identify three common principles of lifelong learning methods and propose an efficient meta-lifelong framework that combines them in a synergistic fashion. To achieve sample efficiency, our method trains the model in a manner that it learns a better initialization for local adaptation. Extensive experiments on text classification and question answering benchmarks demonstrate the effectiveness of our framework by achieving state-of-the-art performance using merely 1% memory size and narrowing the gap with multi-task learning. We further show that our method alleviates both catastrophic forgetting and negative transfer at the same time.

Cross-lingual entity linking (XEL) is the task of finding referents in a target-language knowledge base (KB) for mentions extracted from source-language texts. The first step of (X)EL is candidate generation, which retrieves a list of plausible candidate entities from the target-language KB for each mention. Approaches based on resources from Wikipedia have proven successful in the realm of relatively high-resource languages, but these do not extend well to low-resource languages with few, if any, Wikipedia pages. Recently, transfer learning methods have been shown to reduce the demand for resources in the low-resource languages by utilizing resources in closely related languages, but the performance still lags far behind their high-resource counterparts. In this paper, we first assess the problems faced by current entity candidate generation methods for low-resource XEL, then propose three improvements that (1) reduce the disconnect between entity mentions and KB entries, and (2) improve the robustness of the model to low-resource scenarios. The methods are simple, but effective: We experiment with our approach on seven XEL datasets and find that they yield an average gain of 16.9% in Top-30 gold candidate recall, compared with state-of-the-art baselines. Our improved model also yields an average gain of 7.9% in in-KB accuracy of end-to-end XEL. 1

How can Artificial Intelligence help a stateless minority from online abuse? Research efforts in hate speech detection thus far have largely focused on identifying and subsequently filtering out negative content that specifically targets them. In this paper, we highlight a recent work [8] which tackles a different aspect of web-vulnerability of marginalized communities: sparsity of prominority voices championing their cause. The highlighted paper advocates that blocking hate alone may not be sufficient in these cases as the internet shapes community perception to a great extent in modern times and supportive comments to a vulnerable community serve a different purpose. Using an Active Sampling approach, the paper constructs a nuanced voice-for-the-voiceless classifier that automatically discovers comments supporting a (allegedly) persecuted minority. In the context of the Rohingya refugee crisis, one of the biggest humanitarian crises of modern times, the paper presents promising results that can substantially aid content moderation efforts in finding positive content supporting the Rohingyas.

Multi-view learning makes use of diverse models arising from multiple sources of input or different feature subsets for the same task. For example, a given natural language processing task can combine evidence from models arising from character, morpheme, lexical, or phrasal views. The most common strategy with multi-view learning, especially popular in the neural network community, is to unify multiple representations into one unified vector through concatenation, averaging, or pooling, and then build a single-view model on top of the unified representation. As an alternative, we examine whether building one model per view and then unifying the different models can lead to improvements, especially in low-resource scenarios. More specifically, taking inspiration from co-training methods, we propose a semi-supervised learning approach based on multi-view models through consensus promotion, and investigate whether this improves overall performance. To test the multi-view hypothesis, we use moderately low-resource scenarios for nine languages and test the performance of the joint model for part-of-speech tagging and dependency parsing. The proposed model shows significant improvements across the test cases, with average gains of -0.9 ∼ +9.3 labeled attachment score (LAS) points. We also investigate the effect of unlabeled data on the proposed model by varying the amount of training data and by using different domains of unlabeled data.

The Rohingya refugee crisis is one of the biggest humanitarian crises of modern times with more than 700,000 Rohingyas rendered homeless according to the United Nations High Commissioner for Refugees. While it has received sustained press attention globally, no comprehensive research has been performed on social media pertaining to this large evolving crisis. In this work, we construct a substantial corpus of YouTube video comments (263,482 comments from 113,250 users in 5,153 relevant videos) with an aim to analyze the possible role of AI in helping a marginalized community. Using a novel combination of multiple Active Learning strategies and a novel active sampling strategy based on nearest-neighbors in the comment-embedding space, we construct a classifier that can detect comments defending the Rohingyas among larger numbers of disparaging and neutral ones. We advocate that beyond the burgeoning field of hate speech detection, automatic detection of help speech can lend voice to the voiceless people and make the internet safer for marginalized communities.

To acquire a new skill, humans learn better and faster if a tutor, based on their current knowledge level, informs them of how much attention they should pay to particular content or practice problems. Similarly, a machine learning model could potentially be trained better with a scorer that "adapts" to its current learning state and estimates the importance of each training data instance. Training such an adaptive scorer efficiently is a challenging problem; in order to precisely quantify the effect of a data instance at a given time during the training, it is typically necessary to first complete the entire training process. To efficiently optimize data usage, we propose a reinforcement learning approach called Differentiable Data Selection (DDS). In DDS, we formulate a scorer network as a learnable function of the training data, which can be efficiently updated along with the main model being trained. Specifically, DDS updates the scorer with an intuitive reward signal: it should up-weigh the data that has a similar gradient with a dev set upon which we would finally like to perform well. Without significant computing overhead, DDS delivers strong and consistent improvements over several strong baselines on two very different tasks of machine translation and image classification.

Mining insights from large volume of social media texts with minimal supervision is a highly challenging Natural Language Processing (NLP) task. While Language Models' (LMs) efficacy in several downstream tasks is well-studied, assessing their applicability in answering relational questions, tracking perception or mining deeper insights is under-explored. Few recent lines of work have scratched the surface by studying pre-trained LMs' (e.g., BERT) capability in answering relational questions through "fill-in-the-blank" cloze statements (e.g., [Dante was born in MASK]). BERT predicts the MASK-ed word with a list of words ranked by probability (in this case, BERT successfully predicts Florence with the highest probability). In this paper, we conduct a feasibility study of fine-tuned LMs with a different focus on tracking polls, tracking community perception and mining deeper insights typically obtained through costly surveys. Our main focus is on a substantial corpus of video comments extracted from YouTube videos (6,182,868 comments on 130,067 videos by 1,518,077 users) posted within 100 days prior to the 2019 Indian General Election. Using fill-in-the-blank cloze statements against a recent high-performance language modeling algorithm, BERT, we present a novel application of this family of tools that is able to (1) aggregate political sentiment (2) reveal community perception and (3) track evolving national priorities and issues of interest.

The recent Pulwama terror attack (February 14, 2019, Pulwama, Kashmir) triggered a chain of escalating events between India and Pakistan adding another episode to their 70-year-old dispute over Kashmir. The present era of ubiquitious social media has never seen nuclear powers closer to war. In this paper, we analyze this evolving international crisis via a substantial corpus constructed using comments on YouTube videos (921,235 English comments posted by 392,460 users out of 2.04 million overall comments by 791,289 users on 2,890 videos). Our main contributions in the paper are three-fold. First, we present an observation that polyglot word-embeddings reveal precise and accurate language clusters, and subsequently construct a document language-identification technique with negligible annotation requirements. We demonstrate the viability and utility across a variety of data sets involving several low-resource languages. Second, we present an analysis on temporal trends of pro-peace and pro-war intent observing that when tensions between the two nations were at their peak, pro-peace intent in the corpus was at its highest point. Finally, in the context of heated discussions in a politically tense situation where two nations are at the brink of a full-fledged war, we argue the importance of automatic identification of user-generated web content that can diffuse hostility and address this prediction task, dubbed \emph{hope-speech detection}.

Learning multilingual representations of text has proven a successful method for many cross-lingual transfer learning tasks. There are two main paradigms for learning such representations: (1) alignment, which maps different independently trained monolingual representations into a shared space, and (2) joint training, which directly learns unified multilingual representations using monolingual and cross-lingual objectives jointly. In this paper, we first conduct direct comparisons of representations learned using both of these methods across diverse cross-lingual tasks. Our empirical results reveal a set of pros and cons for both methods, and show that the relative performance of alignment versus joint training is task-dependent. Stemming from this analysis, we propose a simple and novel framework that combines these two previously mutually-exclusive approaches. Extensive experiments demonstrate that our proposed framework alleviates limitations of both approaches, and outperforms existing methods on the MUSE bilingual lexicon induction (BLI) benchmark. We further show that this framework can generalize to contextualized representations such as Multilingual BERT, and produces state-of-the-art results on the CoNLL cross-lingual NER benchmark.

Traditional named entity recognition models use gazetteers (lists of entities) as features to improve performance. Although modern neural network models do not require such hand-crafted features for strong performance, recent work has demonstrated their utility for named entity recognition on English data. However, designing such features for low-resource languages is challenging, because exhaustive entity gazetteers do not exist in these languages. To address this problem, we propose a method of "soft gazetteers" that incorporates ubiquitously available information from English knowledge bases, such as Wikipedia, into neural named entity recognition models through cross-lingual entity linking. Our experiments on four low-resource languages show an average improvement of 4 points in F1 score. Code and data are available at https://github.com/neulab/soft-gazetteers.

With increasing expectations for flexibility and adaptability of machine learning systems, the importance of automatic model updates and performance stability in the face of various types of concept drift has received significant interest. In this study, we explore how feature selection techniques, mostly neglected in the aforementioned effort, may be used to improve the drift compensation process with no a priori assumptions regarding the type of drift. To this end, we (A) evaluate several feature selection techniques by their potential to minimize the effect of drift while still capturing its essence (predict its near-term course), (B) analyze the factors contributing to the success of our proposed method, and (C) provide empirical drift adaptation results via active learning on an extensive data set of real-life political indicators. The results demonstrate that using L1 regularization in the context of our new sample-driven drift-modeling approach results in improved performance as compared to alternative feature selection techniques. The reduced model also requires fewer additional samples to recover from drift even with existing active-sampling strategies.

Code mixing (or code switching) is a common phenomenon observed in social-media content generated by a linguistically diverse user-base. Studies show that in the Indian sub-continent, a substantial fraction of social media posts exhibit code switching. While the difficulties posed by code mixed documents to further downstream analyses are well-understood, lending visibility to code mixed documents under certain scenarios may have utility that has been previously overlooked. For instance, a document written in a mixture of multiple languages can be partially accessible to a wider audience; this could be particularly useful if a considerable fraction of the audience lacks fluency in one of the component languages. In this paper, we provide a systematic approach to sample code mixed documents leveraging a polyglot embedding based method that requires minimal supervision. In the context of the 2019 India-Pakistan conflict triggered by the Pulwama terror attack, we demonstrate an untapped potential of harnessing code mixing for human well-being: starting from an existing hostility diffusing hope speech classifier solely trained on English documents, code mixed documents are utilized to perform crosslingual sampling and retrieve hope speech content written in a low-resource but widely used language - Romanized Hindi. Our proposed pipeline requires minimal supervision and holds promise in substantially reducing web moderation efforts. A further exploratory study on a new COVID-19 data set introduced in this paper demonstrates the generalizability of our cross-lingual sampling technique.

Practitioners apply neural networks to increasingly complex problems in natural language processing, such as syntactic parsing and semantic role labeling that have rich output structures. Many such structured-prediction problems require deterministic constraints on the output values; for example, in sequence-to-sequence syntactic parsing, we require that the sequential outputs encode valid trees. While hidden units might capture such properties, the network is not always able to learn such constraints from the training data alone, and practitioners must then resort to post-processing. In this paper, we present an inference method for neural networks that enforces deterministic constraints on outputs without performing rule-based post-processing or expensive discrete search. Instead, in the spirit of gradient-based training, we enforce constraints with gradient-based inference (GBI): for each input at test-time, we nudge continuous model weights until the network's unconstrained inference procedure generates an output that satisfies the constraints. We study the efficacy of GBI on three tasks with hard constraints: semantic role labeling, syntactic parsing, and sequence transduction. In each case, the algorithm not only satisfies constraints, but improves accuracy, even when the underlying network is state-of-the-art.

When labeled data is scarce for a specific target task, transfer learning often offers an effective solution by utilizing data from a related source task. However, when transferring knowledge from a less related source, it may inversely hurt the target performance, a phenomenon known as negative transfer. Despite its pervasiveness, negative transfer is usually described in an informal manner, lacking rigorous definition, careful analysis, or systematic treatment. This paper proposes a formal definition of negative transfer and analyzes three important aspects thereof. Stemming from this analysis, a novel technique is proposed to circumvent negative transfer by filtering out unrelated source data. Based on adversarial networks, the technique is highly generic and can be applied to a wide range of transfer learning algorithms. The proposed approach is evaluated on six state-of-the-art deep transfer methods via experiments on four benchmark datasets with varying levels of difficulty. Empirically, the proposed method consistently improves the performance of all baseline methods and largely avoids negative transfer, even when the source data is degenerate.

Existing recurrent neural language models often fail to capture higher-level structure present in text: for example, rhyming patterns present in poetry. Much prior work on poetry generation uses manually defined constraints which are satisfied during decoding using either specialized decoding procedures or rejection sampling. The rhyming constraints themselves are typically not learned by the generator. We propose an alternate approach that uses a structured discriminator to learn a poetry generator that directly captures rhyming constraints in a generative adversarial setup. By causing the discriminator to compare poems based only on a learned similarity matrix of pairs of line ending words, the proposed approach is able to successfully learn rhyming patterns in two different English poetry datasets (Sonnet and Limerick) without explicitly being provided with any phonetic information.

For a long time, designing neural architectures that exhibit high performance was considered a dark art that required expert hand-tuning. One of the few well-known guidelines for architecture design is the avoidance of exploding gradients, though even this guideline has remained relatively vague and circumstantial. We introduce the nonlinearity coefficient (NLC), a measurement of the complexity of the function computed by a neural network that is based on the magnitude of the gradient. Via an extensive empirical study, we show that the NLC is a powerful predictor of test error and that attaining a right-sized NLC is essential for optimal performance. The NLC exhibits a range of intriguing and important properties. It is closely tied to the amount of information gained from computing a single network gradient. It is tied to the error incurred when replacing the nonlinearity operations in the network with linear operations. It is not susceptible to the confounders of multiplicative scaling, additive bias and layer width. It is stable from layer to layer. Hence, we argue that the NLC is the first robust predictor of overfitting in deep networks.