Gendered Mental Health Stigma in Masked Language Models

Mental health stigma prevents many individuals from receiving the appropriate care, and social psychology studies have shown that mental health tends to be overlooked in men. In this work, we investigate gendered mental health stigma in masked language models. In doing so, we operationalize mental health stigma by developing a framework grounded in psychology research: we use clinical psychology literature to curate prompts, then evaluate the models' propensity to generate gendered words. We find that masked language models capture societal stigma about gender in mental health: models are consistently more likely to predict female subjects than male in sentences about having a mental health condition (32% vs. 19%), and this disparity is exacerbated for sentences that indicate treatment-seeking behavior. Furthermore, we find that different models capture dimensions of stigma differently for men and women, associating stereotypes like anger, blame, and pity more with women with mental health conditions than with men. In showing the complex nuances of models' gendered mental health stigma, we demonstrate that context and overlapping dimensions of identity are important considerations when assessing computational models' social biases.

Referee: Reference-Free Sentence Summarization with Sharper Controllability through Symbolic Knowledge Distillation

We present Referee, a novel framework for sentence summarization that can be trained reference-free (i.e., requiring no gold summaries for supervision), while allowing direct control for compression ratio. Our work is the first to demonstrate that reference-free, controlled sentence summarization is feasible via the conceptual framework of Symbolic Knowledge Distillation (West et al., 2022), where latent knowledge in pre-trained language models is distilled via explicit examples sampled from the teacher models, further purified with three types of filters: length, fidelity, and Information Bottleneck. Moreover, we uniquely propose iterative distillation of knowledge, where student models from the previous iteration of distillation serve as teacher models in the next iteration. Starting off from a relatively modest set of GPT3-generated summaries, we demonstrate how iterative knowledge distillation can lead to considerably smaller, but better summarizers with sharper controllability. A useful by-product of this iterative distillation process is a high-quality dataset of sentence-summary pairs with varying degrees of compression ratios. Empirical results demonstrate that the final student models vastly outperform the much larger GPT3-Instruct model in terms of the controllability of compression ratios, without compromising the quality of resulting summarization.

Language Generation Models Can Cause Harm: So What Can We Do About It? An Actionable Survey

Recent advances in the capacity of large language models to generate human-like text have resulted in their increased adoption in user-facing settings. In parallel, these improvements have prompted a heated discourse around the risks of societal harms they introduce, whether inadvertent or malicious. Several studies have identified potential causes of these harms and called for their mitigation via development of safer and fairer models. Going beyond enumerating the risks of harms, this work provides a survey of practical methods for addressing potential threats and societal harms from language generation models. We draw on several prior works' taxonomies of language model risks to present a structured overview of strategies for detecting and ameliorating different kinds of risks/harms of language generators. Bridging diverse strands of research, this survey aims to serve as a practical guide for both LM researchers and practitioners with explanations of motivations behind different mitigation strategies, their limitations, and open problems for future research.

KALM: Knowledge-Aware Integration of Local, Document, and Global Contexts for Long Document Understanding

With the advent of pre-trained language models (LMs), increasing research efforts have been focusing on infusing commonsense and domain-specific knowledge to prepare LMs for downstream tasks. These works attempt to leverage knowledge graphs, the de facto standard of symbolic knowledge representation, along with pre-trained LMs. While existing approaches leverage external knowledge, it remains an open question how to jointly incorporate knowledge graphs representing varying contexts, from local (e.g., sentence), to document-level, to global knowledge, to enable knowledge-rich and interpretable exchange across these contexts. Such rich contextualization can be especially beneficial for long document understanding tasks since standard pre-trained LMs are typically bounded by the input sequence length. In light of these challenges, we propose KALM, a Knowledge-Aware Language Model that jointly leverages knowledge in local, document-level, and global contexts for long document understanding. KALM first encodes long documents and knowledge graphs into the three knowledge-aware context representations. It then processes each context with context-specific layers, followed by a context fusion layer that facilitates interpretable knowledge exchange to derive an overarching document representation. Extensive experiments demonstrate that KALM achieves state-of-the-art performance on three long document understanding tasks across 6 datasets/settings. Further analyses reveal that the three knowledge-aware contexts are complementary and they all contribute to model performance, while the importance and information exchange patterns of different contexts vary with respect to different tasks and datasets.

ORCA: Interpreting Prompted Language Models via Locating Supporting Data Evidence in the Ocean of Pretraining Data

Large pretrained language models have been performing increasingly well in a variety of downstream tasks via prompting. However, it remains unclear from where the model learns the task-specific knowledge, especially in a zero-shot setup. In this work, we want to find evidence of the model's task-specific competence from pretraining and are specifically interested in locating a very small subset of pretraining data that directly supports the model in the task. We call such a subset supporting data evidence and propose a novel method ORCA to effectively identify it, by iteratively using gradient information related to the downstream task. This supporting data evidence offers interesting insights about the prompted language models: in the tasks of sentiment analysis and textual entailment, BERT shows a substantial reliance on BookCorpus, the smaller corpus of BERT's two pretraining corpora, as well as on pretraining examples that mask out synonyms to the task verbalizers.

Constrained Sampling from Language Models via Langevin Dynamics in Embedding Spaces

Large pre-trained language models are well-established for their ability to generate text seemingly indistinguishable from humans. In this work, we study the problem of constrained sampling from such language models. That is, generating text that satisfies user-defined constraints. Typical decoding strategies which generate samples left-to-right are not always conducive to imposing such constraints globally. Instead, we propose MuCoLa -- a sampling procedure that combines the log-likelihood of the language model with arbitrary differentiable constraints into a single energy function; and generates samples by initializing the entire output sequence with noise and following a Markov chain defined by Langevin Dynamics using the gradients of this energy. We evaluate our approach on different text generation tasks with soft and hard constraints as well as their combinations with competitive results for toxicity avoidance, sentiment control, and keyword-guided generation.

VoynaSlov: A Data Set of Russian Social Media Activity during the 2022 Ukraine-Russia War

In this report, we describe a new data set called VoynaSlov which contains 21M+ Russian-language social media activities (i.e. tweets, posts, comments) made by Russian media outlets and by the general public during the time of war between Ukraine and Russia. We scraped the data from two major platforms that are widely used in Russia: Twitter and VKontakte (VK), a Russian social media platform based in Saint Petersburg commonly referred to as "Russian Facebook". We provide descriptions of our data collection process and data statistics that compare state-affiliated and independent Russian media, and also the two platforms, VK and Twitter. The main differences that distinguish our data from previously released data related to the ongoing war are its focus on Russian media and consideration of state-affiliation as well as the inclusion of data from VK, which is more suitable than Twitter for understanding Russian public sentiment considering its wide use within Russia. We hope our data set can facilitate future research on information warfare and ultimately enable the reduction and prevention of disinformation and opinion manipulation campaigns. The data set is available at https://github.com/chan0park/VoynaSlov and will be regularly updated as we continuously collect more data.

Speaker Information Can Guide Models to Better Inductive Biases: A Case Study On Predicting Code-Switching

Natural language processing (NLP) models trained on people-generated data can be unreliable because, without any constraints, they can learn from spurious correlations that are not relevant to the task. We hypothesize that enriching models with speaker information in a controlled, educated way can guide them to pick up on relevant inductive biases. For the speaker-driven task of predicting code-switching points in English--Spanish bilingual dialogues, we show that adding sociolinguistically-grounded speaker features as prepended prompts significantly improves accuracy. We find that by adding influential phrases to the input, speaker-informed models learn useful and explainable linguistic information. To our knowledge, we are the first to incorporate speaker characteristics in a neural model for code-switching, and more generally, take a step towards developing transparent, personalized models that use speaker information in a controlled way.

Unsupervised Keyphrase Extraction via Interpretable Neural Networks

Keyphrase extraction aims at automatically extracting a list of "important" phrases which represent the key concepts in a document. Prior approaches for unsupervised keyphrase extraction resort to heuristic notions of phrase importance via embedding similarities or graph centrality, requiring extensive domain expertise to develop them. Our work proposes an alternative operational definition: phrases that are most useful for predicting the topic of a text are important keyphrases. To this end, we propose INSPECT -- a self-explaining neural framework for identifying influential keyphrases by measuring the predictive impact of input phrases on the downstream task of topic classification. We show that this novel approach not only alleviates the need for ad-hoc heuristics but also achieves state-of-the-art results in unsupervised keyphrase extraction across four diverse datasets in two domains: scientific publications and news articles. Ultimately, our study suggests a new usage of interpretable neural networks as an intrinsic component in NLP systems, and not only as a tool for explaining model predictions to humans.

Improving the Diversity of Unsupervised Paraphrasing with Embedding Outputs

We present a novel technique for zero-shot paraphrase generation. The key contribution is an end-to-end multilingual paraphrasing model that is trained using translated parallel corpora to generate paraphrases into "meaning spaces" -- replacing the final softmax layer with word embeddings. This architectural modification, plus a training procedure that incorporates an autoencoding objective, enables effective parameter sharing across languages for more fluent monolingual rewriting, and facilitates fluency and diversity in generation. Our continuous-output paraphrase generation models outperform zero-shot paraphrasing baselines when evaluated on two languages using a battery of computational metrics as well as in human assessment.