Deb Roy

Suyash Fulay, William Brannon, Shrestha Mohanty et al. · mit

Language model alignment research often attempts to ensure that models are not only helpful and harmless, but also truthful and unbiased. However, optimizing these objectives simultaneously can obscure how improving one aspect might impact the others. In this work, we focus on analyzing the relationship between two concepts essential in both language model alignment and political science: truthfulness and political bias. We train reward models on various popular truthfulness datasets and subsequently evaluate their political bias. Our findings reveal that optimizing reward models for truthfulness on these datasets tends to result in a left-leaning political bias. We also find that existing open-source reward models (i.e., those trained on standard human preference datasets) already show a similar bias and that the bias is larger for larger models. These results raise important questions about the datasets used to represent truthfulness, potential limitations of aligning models to be both truthful and politically unbiased, and what language models capture about the relationship between truth and politics.

Eric Chu, Jacob Andreas, Stephen Ansolabehere et al. · microsoft-research, mit

Public opinion reflects and shapes societal behavior, but the traditional survey-based tools to measure it are limited. We introduce a novel approach to probe media diet models -- language models adapted to online news, TV broadcast, or radio show content -- that can emulate the opinions of subpopulations that have consumed a set of media. To validate this method, we use as ground truth the opinions expressed in U.S. nationally representative surveys on COVID-19 and consumer confidence. Our studies indicate that this approach is (1) predictive of human judgements found in survey response distributions and robust to phrasing and channels of media exposure, (2) more accurate at modeling people who follow media more closely, and (3) aligned with literature on which types of opinions are affected by media consumption. Probing language models provides a powerful new method for investigating media effects, has practical applications in supplementing polls and forecasting public opinion, and suggests a need for further study of the surprising fidelity with which neural language models can predict human responses.

Hang Jiang, Doug Beeferman, Brandon Roy et al. · mit

As political attitudes have diverged ideologically in the United States, political speech has diverged lingusitically. The ever-widening polarization between the US political parties is accelerated by an erosion of mutual understanding between them. We aim to make these communities more comprehensible to each other with a framework that probes community-specific responses to the same survey questions using community language models CommunityLM. In our framework we identify committed partisan members for each community on Twitter and fine-tune LMs on the tweets authored by them. We then assess the worldviews of the two groups using prompt-based probing of their corresponding LMs, with prompts that elicit opinions about public figures and groups surveyed by the American National Election Studies (ANES) 2020 Exploratory Testing Survey. We compare the responses generated by the LMs to the ANES survey results, and find a level of alignment that greatly exceeds several baseline methods. Our work aims to show that we can use community LMs to query the worldview of any group of people given a sufficiently large sample of their social media discussions or media diet.

William Brannon, Doug Beeferman, Hang Jiang et al. · mit

Understanding and making use of audience feedback is important but difficult for journalists, who now face an impractically large volume of audience comments online. We introduce AudienceView, an online tool to help journalists categorize and interpret this feedback by leveraging large language models (LLMs). AudienceView identifies themes and topics, connects them back to specific comments, provides ways to visualize the sentiment and distribution of the comments, and helps users develop ideas for subsequent reporting projects. We consider how such tools can be useful in a journalist's workflow, and emphasize the importance of contextual awareness and human judgment.

Hang Jiang, Doug Beeferman, William Brannon et al. · mit

Words often carry different meanings for people from diverse backgrounds. Today's era of social polarization demands that we choose words carefully to prevent miscommunication, especially in political communication and journalism. To address this issue, we introduce the Bridging Dictionary, an interactive tool designed to illuminate how words are perceived by people with different political views. The Bridging Dictionary includes a static, printable document featuring 796 terms with summaries generated by a large language model. These summaries highlight how the terms are used distinctively by Republicans and Democrats. Additionally, the Bridging Dictionary offers an interactive interface that lets users explore selected words, visualizing their frequency, sentiment, summaries, and examples across political divides. We present a use case for journalists and emphasize the importance of human agency and trust in further enhancing this tool. The deployed version of Bridging Dictionary is available at https://dictionary.ccc-mit.org/.

Prashanth Vijayaraghavan, Deb Roy

Narrative is a ubiquitous component of human communication. Understanding its structure plays a critical role in a wide variety of applications, ranging from simple comparative analyses to enhanced narrative retrieval, comprehension, or reasoning capabilities. Prior research in narratology has highlighted the importance of studying the links between cognitive and linguistic aspects of narratives for effective comprehension. This interdependence is related to the textual semantics and mental language in narratives, referring to characters' motivations, feelings or emotions, and beliefs. However, this interdependence is hardly explored for modeling narratives. In this work, we propose the task of automatically detecting prominent elements of the narrative structure by analyzing the role of characters' inferred mental state along with linguistic information at the syntactic and semantic levels. We introduce a STORIES dataset of short personal narratives containing manual annotations of key elements of narrative structure, specifically climax and resolution. To this end, we implement a computational model that leverages the protagonist's mental state information obtained from a pre-trained model trained on social commonsense knowledge and integrates their representations with contextual semantic embed-dings using a multi-feature fusion approach. Evaluating against prior zero-shot and supervised baselines, we find that our model is able to achieve significant improvements in the task of identifying climax and resolution.

Nabeel Gillani, Doug Beeferman, Christine Vega-Pourheydarian et al.

Most US school districts draw "attendance boundaries" to define catchment areas that assign students to schools near their homes, often recapitulating neighborhood demographic segregation in schools. Focusing on elementary schools, we ask: how much might we reduce school segregation by redrawing attendance boundaries? Combining parent preference data with methods from combinatorial optimization, we simulate alternative boundaries for 98 US school districts serving over 3 million elementary-aged students, minimizing White/non-White segregation while mitigating changes to travel times and school sizes. Across districts, we observe a median 14% relative decrease in segregation, which we estimate would require approximately 20\% of students to switch schools and, surprisingly, a slight reduction in travel times. We release a public dashboard depicting these alternative boundaries (https://www.schooldiversity.org/) and invite both school boards and their constituents to evaluate their viability. Our results show the possibility of greater integration without significant disruptions for families.

Suyash Fulay, Prerna Ravi, Emily Kubin et al.

AI is increasingly used to scale collective decision-making, but far less attention has been paid to how such systems can support procedural legitimacy, particularly the conditions shaping losers' consent: whether participants who do not get their preferred outcome still accept it as fair. We ask: (1) how can AI help ground collective decisions in participants' different experiences and beliefs, and (2) whether exposure to these experiences can increase trust, understanding, and social cohesion even when people disagree with the outcome. We built a system that uses a semi-structured AI interviewer to elicit personal experiences on policy topics and an interactive visualization that displays predicted policy support alongside those voiced experiences. In a randomized experiment (n = 181), interacting with the visualization increased perceived legitimacy, trust in outcomes, and understanding of others' perspectives, even though all participants encountered decisions that went against their stated preferences. Our hope is that the design and evaluation of this tool spurs future researchers to focus on how AI can help not only achieve scale and efficiency in democratic processes, but also increase trust and connection between participants.

William Brannon, Wonjune Kang, Suyash Fulay et al.

Learning on text-attributed graphs (TAGs), in which nodes are associated with one or more texts, has been the subject of much recent work. However, most approaches tend to make strong assumptions about the downstream task of interest, are reliant on hand-labeled data, or fail to equally balance the importance of both text and graph representations. In this work, we propose Contrastive Graph-Text pretraining (ConGraT), a general, self-supervised approach for jointly learning separate representations of texts and nodes in a TAG. Our method trains a language model (LM) and a graph neural network (GNN) to align their representations in a common latent space using a batch-wise contrastive learning objective inspired by CLIP. We further propose an extension to the CLIP objective that leverages graph structure to incorporate information about inter-node similarity. Extensive experiments demonstrate that ConGraT outperforms baselines on various downstream tasks, including node and text category classification, link prediction, and language modeling. Finally, we present an application of our method to community detection in social graphs, which enables finding more textually grounded communities, rather than purely graph-based ones. Code and certain datasets are available at https://github.com/wwbrannon/congrat.

Hang Jiang, Yining Hua, Doug Beeferman et al.

Social media data such as Twitter messages ("tweets") pose a particular challenge to NLP systems because of their short, noisy, and colloquial nature. Tasks such as Named Entity Recognition (NER) and syntactic parsing require highly domain-matched training data for good performance. To date, there is no complete training corpus for both NER and syntactic analysis (e.g., part of speech tagging, dependency parsing) of tweets. While there are some publicly available annotated NLP datasets of tweets, they are only designed for individual tasks. In this study, we aim to create Tweebank-NER, an English NER corpus based on Tweebank V2 (TB2), train state-of-the-art (SOTA) Tweet NLP models on TB2, and release an NLP pipeline called Twitter-Stanza. We annotate named entities in TB2 using Amazon Mechanical Turk and measure the quality of our annotations. We train the Stanza pipeline on TB2 and compare with alternative NLP frameworks (e.g., FLAIR, spaCy) and transformer-based models. The Stanza tokenizer and lemmatizer achieve SOTA performance on TB2, while the Stanza NER tagger, part-of-speech (POS) tagger, and dependency parser achieve competitive performance against non-transformer models. The transformer-based models establish a strong baseline in Tweebank-NER and achieve the new SOTA performance in POS tagging and dependency parsing on TB2. We release the dataset and make both the Stanza pipeline and BERTweet-based models available "off-the-shelf" for use in future Tweet NLP research. Our source code, data, and pre-trained models are available at: \url{https://github.com/social-machines/TweebankNLP}.

Hang Jiang, Xiajie Zhang, Robert Mahari et al. · allen-ai

Making legal knowledge accessible to non-experts is crucial for enhancing general legal literacy and encouraging civic participation in democracy. However, legal documents are often challenging to understand for people without legal backgrounds. In this paper, we present a novel application of large language models (LLMs) in legal education to help non-experts learn intricate legal concepts through storytelling, an effective pedagogical tool in conveying complex and abstract concepts. We also introduce a new dataset LegalStories, which consists of 294 complex legal doctrines, each accompanied by a story and a set of multiple-choice questions generated by LLMs. To construct the dataset, we experiment with various LLMs to generate legal stories explaining these concepts. Furthermore, we use an expert-in-the-loop approach to iteratively design multiple-choice questions. Then, we evaluate the effectiveness of storytelling with LLMs through randomized controlled trials (RCTs) with legal novices on 10 samples from the dataset. We find that LLM-generated stories enhance comprehension of legal concepts and interest in law among non-native speakers compared to only definitions. Moreover, stories consistently help participants relate legal concepts to their lives. Finally, we find that learning with stories shows a higher retention rate for non-native speakers in the follow-up assessment. Our work has strong implications for using LLMs in promoting teaching and learning in the legal field and beyond.

Suyash Fulay, Prerna Ravi, Emily Kubin et al.

Deliberative democratic theory suggests that civic competence: the capacity to navigate disagreement, weigh competing values, and arrive at collective decisions is not innate but developed through practice. Yet opportunities to cultivate these skills remain limited, as traditional deliberative processes like citizens' assemblies reach only a small fraction of the population. We present Agora, an early-stage AI-powered platform that uses LLMs to organize authentic human voices on policy issues, helping users build consensus-finding skills by proposing and revising policy recommendations, hearing supporting and opposing perspectives, and receiving feedback on how policy changes affect predicted support. In a preliminary study with 44 university students, participants using the full interface (with access to voice explanations) reported higher levels of problem-solving skills, internal deliberation, and produced higher quality consensus statements compared to a control condition showing only aggregate support distributions. These initial findings point toward a promising direction for scaling civic education.

Suyash Fulay, Dimitra Dimitrakopoulou, Deb Roy

Deliberation is essential to well-functioning democracies, yet physical, economic, and social barriers often exclude certain groups, reducing representativeness and contributing to issues like group polarization. In this work, we explore the use of large language model (LLM) personas to introduce missing perspectives in policy deliberations. We develop and evaluate a tool that transcribes conversations in real-time and simulates input from relevant but absent stakeholders. We deploy this tool in a 19-person student citizens' assembly on campus sustainability. Participants and facilitators found that the tool was useful to spark new discussions and surfaced valuable perspectives they had not previously considered. However, they also raised skepticism about the ability of LLMs to accurately characterize the perspectives of different groups, especially ones that are already underrepresented. Overall, this case study highlights that while AI personas can usefully surface new perspectives and prompt discussion in deliberative settings, their successful deployment depends on clarifying their limitations and emphasizing that they complement rather than replace genuine participation.

Elinor Poole-Dayan, Daniel T Kessler, Hannah Chiou et al. · mit

Public Narratives (PNs) are key tools for leadership development and civic mobilization, yet their systematic analysis remains challenging due to their subjective interpretation and the high cost of expert annotation. In this work, we propose a novel computational framework that leverages large language models (LLMs) to automate the qualitative annotation of public narratives. Using a codebook we co-developed with subject-matter experts, we evaluate LLM performance against that of expert annotators. Our work reveals that LLMs can achieve near-human-expert performance, achieving an average F1 score of 0.80 across 8 narratives and 14 codes. We then extend our analysis to empirically explore how PN framework elements manifest across a larger dataset of 22 stories. Lastly, we extrapolate our analysis to a set of political speeches, establishing a novel lens in which to analyze political rhetoric in civic spaces. This study demonstrates the potential of LLM-assisted annotation for scalable narrative analysis and highlights key limitations and directions for future research in computational civic storytelling.

Margaret Hughes, Brandon Roy, Elinor Poole-Dayan et al. · mit

Growing literature explores toxicity and polarization in discourse, with comparatively less work on characterizing what makes dialogue prosocial and constructive. We explore conversational discourse and investigate a method for characterizing its quality built upon the notion of ``responsivity'' -- whether one person's conversational turn is responding to a preceding turn. We develop and evaluate methods for quantifying responsivity -- first through semantic similarity of speaker turns, and second by leveraging state-of-the-art large language models (LLMs) to identify the relation between two speaker turns. We evaluate both methods against a ground truth set of human-annotated conversations. Furthermore, selecting the better performing LLM-based approach, we characterize the nature of the response -- whether it responded to that preceding turn in a substantive way or not. We view these responsivity links as a fundamental aspect of dialogue but note that conversations can exhibit significantly different responsivity structures. Accordingly, we then develop conversation-level derived metrics to address various aspects of conversational discourse. We use these derived metrics to explore other conversations and show that they support meaningful characterizations and differentiations across a diverse collection of conversations.

Elinor Poole-Dayan, Deb Roy, Jad Kabbara · mit

In an era of increasing societal fragmentation, political polarization, and erosion of public trust in institutions, representative deliberative assemblies are emerging as a promising democratic forum for developing effective policy outcomes on complex global issues. Despite theoretical attention, there remains limited empirical work that systematically traces how specific ideas evolve, are prioritized, or are discarded during deliberation to form policy recommendations. Addressing these gaps, this work poses two central questions: (1) How might we trace the evolution and distillation of ideas into concrete recommendations within deliberative assemblies? (2) How does the deliberative process shape delegate perspectives and influence voting dynamics over the course of the assembly? To address these questions, we develop LLM-based methodologies for empirically analyzing transcripts from a tech-enhanced in-person deliberative assembly. The framework identifies and visualizes the space of expressed suggestions. We also empirically reconstruct each delegate's evolving perspective throughout the assembly. Our methods contribute novel empirical insights into deliberative processes and demonstrate how LLMs can surface high-resolution dynamics otherwise invisible in traditional assembly outputs.

Wonjune Kang, Deb Roy · mit

We introduce the task of expressive speech retrieval, where the goal is to retrieve speech utterances spoken in a given style based on a natural language description of that style. While prior work has primarily focused on performing speech retrieval based on what was said in an utterance, we aim to do so based on how something was said. We train speech and text encoders to embed speech and text descriptions of speaking styles into a joint latent space, which enables using free-form text prompts describing emotions or styles as queries to retrieve matching expressive speech segments. We perform detailed analyses of various aspects of our proposed framework, including encoder architectures, training criteria for effective cross-modal alignment, and prompt augmentation for improved generalization to arbitrary text queries. Experiments on multiple datasets encompassing 22 speaking styles demonstrate that our approach achieves strong retrieval performance as measured by Recall@k.

Shrestha Mohanty, Sarah Xuan, Jacob Jobraeel et al.

We focus on enhancing comprehension in small-group recorded conversations, which serve as a medium to bring people together and provide a space for sharing personal stories and experiences on crucial social matters. One way to parse and convey information from these conversations is by sharing highlighted excerpts in subsequent conversations. This can help promote a collective understanding of relevant issues, by highlighting perspectives and experiences to other groups of people who might otherwise be unfamiliar with and thus unable to relate to these experiences. The primary challenge that arises then is that excerpts taken from one conversation and shared in another setting might be missing crucial context or key elements that were previously introduced in the original conversation. This problem is exacerbated when conversations become lengthier and richer in themes and shared experiences. To address this, we explore how Large Language Models (LLMs) can enrich these excerpts by providing socially relevant context. We present approaches for effective contextualization to improve comprehension, readability, and empathy. We show significant improvements in understanding, as assessed through subjective and objective evaluations. While LLMs can offer valuable context, they struggle with capturing key social aspects. We release the Human-annotated Salient Excerpts (HSE) dataset to support future work. Additionally, we show how context-enriched excerpts can provide more focused and comprehensive conversation summaries.

Elinor Poole-Dayan, Deb Roy, Jad Kabbara

While state-of-the-art large language models (LLMs) have shown impressive performance on many tasks, there has been extensive research on undesirable model behavior such as hallucinations and bias. In this work, we investigate how the quality of LLM responses changes in terms of information accuracy, truthfulness, and refusals depending on three user traits: English proficiency, education level, and country of origin. We present extensive experimentation on three state-of-the-art LLMs and two different datasets targeting truthfulness and factuality. Our findings suggest that undesirable behaviors in state-of-the-art LLMs occur disproportionately more for users with lower English proficiency, of lower education status, and originating from outside the US, rendering these models unreliable sources of information towards their most vulnerable users.

Wonjune Kang, Deb Roy

In this work, we introduce a framework for speech summarization that leverages the processing and reasoning capabilities of large language models (LLMs). We propose an end-to-end system that combines an instruction-tuned LLM with an audio encoder that converts speech into token representations that the LLM can interpret. Using a dataset with paired speech-text data, the overall system is trained to generate consistent responses to prompts with the same semantic information regardless of the input modality. The resulting framework allows the LLM to process speech inputs in the same way as text, enabling speech summarization by simply prompting the LLM. Unlike prior approaches, our method is able to summarize spoken content from any arbitrary domain, and it can produce summaries in different styles by varying the LLM prompting strategy. Experiments demonstrate that our approach outperforms a cascade baseline of speech recognition followed by LLM text processing.

Hang Jiang, Xiajie Zhang, Xubo Cao et al.

Despite the many use cases for large language models (LLMs) in creating personalized chatbots, there has been limited research on evaluating the extent to which the behaviors of personalized LLMs accurately and consistently reflect specific personality traits. We consider studying the behavior of LLM-based agents which we refer to as LLM personas and present a case study with GPT-3.5 and GPT-4 to investigate whether LLMs can generate content that aligns with their assigned personality profiles. To this end, we simulate distinct LLM personas based on the Big Five personality model, have them complete the 44-item Big Five Inventory (BFI) personality test and a story writing task, and then assess their essays with automatic and human evaluations. Results show that LLM personas' self-reported BFI scores are consistent with their designated personality types, with large effect sizes observed across five traits. Additionally, LLM personas' writings have emerging representative linguistic patterns for personality traits when compared with a human writing corpus. Furthermore, human evaluation shows that humans can perceive some personality traits with an accuracy of up to 80%. Interestingly, the accuracy drops significantly when the annotators were informed of AI authorship.

Hang Jiang, Doug Beeferman, Weiquan Mao et al.

The time at which a message is communicated is a vital piece of metadata in many real-world natural language processing tasks such as Topic Detection and Tracking (TDT). TDT systems aim to cluster a corpus of news articles by event, and in that context, stories that describe the same event are likely to have been written at around the same time. Prior work on time modeling for TDT takes this into account, but does not well capture how time interacts with the semantic nature of the event. For example, stories about a tropical storm are likely to be written within a short time interval, while stories about a movie release may appear over weeks or months. In our work, we design a neural method that fuses temporal and textual information into a single representation of news documents for event detection. We fine-tune these time-aware document embeddings with a triplet loss architecture, integrate the model into downstream TDT systems, and evaluate the systems on two benchmark TDT data sets in English. In the retrospective setting, we apply clustering algorithms to the time-aware embeddings and show substantial improvements over baselines on the News2013 data set. In the online streaming setting, we add our document encoder to an existing state-of-the-art TDT pipeline and demonstrate that it can benefit the overall performance. We conduct ablation studies on the time representation and fusion algorithm strategies, showing that our proposed model outperforms alternative strategies. Finally, we probe the model to examine how it handles recurring events more effectively than previous TDT systems.

Ivan Sysoev, James H. Gray, Susan Fine et al.

English has a convoluted relationship between its pronunciation and spelling, which obscures its phonological structure for early literacy learners. This convoluted relationship has implications for early literacy software, particularly for open-ended, child-driven designs. A tempting way to bypass this issue is to use manipulables (blocks) that are directly tied to phonemes. However, creating phoneme-based blocks leads to two design challenges: (a) how to represent phonemes visually in a child-accessible way and (b) how to account for context-dependent spelling. In the present work, we approached these challenges by developing a set of animated, onomatopoeia-based mnemonic characters, one per phoneme, that can take the shape of different graphemes.We applied the characters to a construction-based literacy app to simplify independent word-building for literacy beginners. We tested the app during a 13-week-long period with 4- to 5-year-olds in kindergarten classrooms. Children showed visible interest in the characters and properly grasped the principles of their functioning. However, the blocks were not sufficient to scaffold independent word building, leading children to rely on other scaffolding mechanisms. To test the characters' efficiency as mnemonics, we evaluated their effect on the speed and accuracy of finding phonemes on a keyboard. The results suggest that there were both children who benefitted from the characters in this task and those who performed better without them. The factors that differentiated these two categories are currently unclear. To help further research on phonetic mnemonics in literacy learning software, we are making the characters available to the research community.

Prashanth Vijayaraghavan, Hugo Larochelle, Deb Roy

With growing role of social media in shaping public opinions and beliefs across the world, there has been an increased attention to identify and counter the problem of hate speech on social media. Hate speech on online spaces has serious manifestations, including social polarization and hate crimes. While prior works have proposed automated techniques to detect hate speech online, these techniques primarily fail to look beyond the textual content. Moreover, few attempts have been made to focus on the aspects of interpretability of such models given the social and legal implications of incorrect predictions. In this work, we propose a deep neural multi-modal model that can: (a) detect hate speech by effectively capturing the semantics of the text along with socio-cultural context in which a particular hate expression is made, and (b) provide interpretable insights into decisions of our model. By performing a thorough evaluation of different modeling techniques, we demonstrate that our model is able to outperform the existing state-of-the-art hate speech classification approaches. Finally, we show the importance of social and cultural context features towards unearthing clusters associated with different categories of hate.

Prashanth Vijayaraghavan, Deb Roy

Stories are a very compelling medium to convey ideas, experiences, social and cultural values. Narrative is a specific manifestation of the story that turns it into knowledge for the audience. In this paper, we propose a machine learning approach to capture the narrative elements in movies by bridging the gap between the low-level data representations and semantic aspects of the visual medium. We present a Memory-Augmented Video Semantic Network, called Video SemNet, to encode the semantic descriptors and learn an embedding for the video. The model employs two main components: (i) a neural semantic learner that learns latent embeddings of semantic descriptors and (ii) a memory module that retains and memorizes specific semantic patterns from the video. We evaluate the video representations obtained from variants of our model on two tasks: (a) genre prediction and (b) IMDB Rating prediction. We demonstrate that our model is able to predict genres and IMDB ratings with a weighted F-1 score of 0.72 and 0.63 respectively. The results are indicative of the representational power of our model and the ability of such representations to measure audience engagement.

Eric Chu, Deb Roy, Jacob Andreas

We present a randomized controlled trial for a model-in-the-loop regression task, with the goal of measuring the extent to which (1) good explanations of model predictions increase human accuracy, and (2) faulty explanations decrease human trust in the model. We study explanations based on visual saliency in an image-based age prediction task for which humans and learned models are individually capable but not highly proficient and frequently disagree. Our experimental design separates model quality from explanation quality, and makes it possible to compare treatments involving a variety of explanations of varying levels of quality. We find that presenting model predictions improves human accuracy. However, visual explanations of various kinds fail to significantly alter human accuracy or trust in the model - regardless of whether explanations characterize an accurate model, an inaccurate one, or are generated randomly and independently of the input image. These findings suggest the need for greater evaluation of explanations in downstream decision making tasks, better design-based tools for presenting explanations to users, and better approaches for generating explanations.

Belen Saldias, Deb Roy

Sharing personal narratives is a fundamental aspect of human social behavior as it helps share our life experiences. We can tell stories and rely on our background to understand their context, similarities, and differences. A substantial effort has been made towards developing storytelling machines or inferring characters' features. However, we don't usually find models that compare narratives. This task is remarkably challenging for machines since they, as sometimes we do, lack an understanding of what similarity means. To address this challenge, we first introduce a corpus of real-world spoken personal narratives comprising 10,296 narrative clauses from 594 video transcripts. Second, we ask non-narrative experts to annotate those clauses under Labov's sociolinguistic model of personal narratives (i.e., action, orientation, and evaluation clause types) and train a classifier that reaches 84.7% F-score for the highest-agreed clauses. Finally, we match stories and explore whether people implicitly rely on Labov's framework to compare narratives. We show that actions followed by the narrator's evaluation of these are the aspects non-experts consider the most. Our approach is intended to help inform machine learning methods aimed at studying or representing personal narratives.

Prashanth Vijayaraghavan, Deb Roy

Recently, generating adversarial examples has become an important means of measuring robustness of a deep learning model. Adversarial examples help us identify the susceptibilities of the model and further counter those vulnerabilities by applying adversarial training techniques. In natural language domain, small perturbations in the form of misspellings or paraphrases can drastically change the semantics of the text. We propose a reinforcement learning based approach towards generating adversarial examples in black-box settings. We demonstrate that our method is able to fool well-trained models for (a) IMDB sentiment classification task and (b) AG's news corpus news categorization task with significantly high success rates. We find that the adversarial examples generated are semantics-preserving perturbations to the original text.

Doug Beeferman, William Brannon, Deb Roy

We introduce RadioTalk, a corpus of speech recognition transcripts sampled from talk radio broadcasts in the United States between October of 2018 and March of 2019. The corpus is intended for use by researchers in the fields of natural language processing, conversational analysis, and the social sciences. The corpus encompasses approximately 2.8 billion words of automatically transcribed speech from 284,000 hours of radio, together with metadata about the speech, such as geographical location, speaker turn boundaries, gender, and radio program information. In this paper we summarize why and how we prepared the corpus, give some descriptive statistics on stations, shows and speakers, and carry out a few high-level analyses.

David McClure, Shayne O'Brien, Deb Roy

We formulate coherence modeling as a regression task and propose two novel methods to combine techniques from our setup with pairwise approaches. The first of our methods is a model that we call "first-next," which operates similarly to selection sorting but conditions decision-making on information about already-sorted sentences. The second consists of a technique for adding context to regression-based models by concatenating sentence-level representations with an encoding of its corresponding out-of-order paragraph. This latter model achieves Kendall-tau distance and positional accuracy scores that match or exceed the current state-of-the-art on these metrics. Our results suggest that many of the gains that come from more complex, machine-translation inspired approaches can be achieved with simpler, more efficient models.

Eric Chu, Prashanth Vijayaraghavan, Deb Roy

The ability to infer persona from dialogue can have applications in areas ranging from computational narrative analysis to personalized dialogue generation. We introduce neural models to learn persona embeddings in a supervised character trope classification task. The models encode dialogue snippets from IMDB into representations that can capture the various categories of film characters. The best-performing models use a multi-level attention mechanism over a set of utterances. We also utilize prior knowledge in the form of textual descriptions of the different tropes. We apply the learned embeddings to find similar characters across different movies, and cluster movies according to the distribution of the embeddings. The use of short conversational text as input, and the ability to learn from prior knowledge using memory, suggests these methods could be applied to other domains.

Eric Chu, Deb Roy

Stories can have tremendous power -- not only useful for entertainment, they can activate our interests and mobilize our actions. The degree to which a story resonates with its audience may be in part reflected in the emotional journey it takes the audience upon. In this paper, we use machine learning methods to construct emotional arcs in movies, calculate families of arcs, and demonstrate the ability for certain arcs to predict audience engagement. The system is applied to Hollywood films and high quality shorts found on the web. We begin by using deep convolutional neural networks for audio and visual sentiment analysis. These models are trained on both new and existing large-scale datasets, after which they can be used to compute separate audio and visual emotional arcs. We then crowdsource annotations for 30-second video clips extracted from highs and lows in the arcs in order to assess the micro-level precision of the system, with precision measured in terms of agreement in polarity between the system's predictions and annotators' ratings. These annotations are also used to combine the audio and visual predictions. Next, we look at macro-level characterizations of movies by investigating whether there exist `universal shapes' of emotional arcs. In particular, we develop a clustering approach to discover distinct classes of emotional arcs. Finally, we show on a sample corpus of short web videos that certain emotional arcs are statistically significant predictors of the number of comments a video receives. These results suggest that the emotional arcs learned by our approach successfully represent macroscopic aspects of a video story that drive audience engagement. Such machine understanding could be used to predict audience reactions to video stories, ultimately improving our ability as storytellers to communicate with each other.

Thomas Kollar, Stefanie Tellex, Matthew Walter et al.

Many task domains require robots to interpret and act upon natural language commands which are given by people and which refer to the robot's physical surroundings. Such interpretation is known variously as the symbol grounding problem, grounded semantics and grounded language acquisition. This problem is challenging because people employ diverse vocabulary and grammar, and because robots have substantial uncertainty about the nature and contents of their surroundings, making it difficult to associate the constitutive language elements (principally noun phrases and spatial relations) of the command text to elements of those surroundings. Symbolic models capture linguistic structure but have not scaled successfully to handle the diverse language produced by untrained users. Existing statistical approaches can better handle diversity, but have not to date modeled complex linguistic structure, limiting achievable accuracy. Recent hybrid approaches have addressed limitations in scaling and complexity, but have not effectively associated linguistic and perceptual features. Our framework, called Generalized Grounding Graphs (G^3), addresses these issues by defining a probabilistic graphical model dynamically according to the linguistic parse structure of a natural language command. This approach scales effectively, handles linguistic diversity, and enables the system to associate parts of a command with the specific objects, places, and events in the external world to which they refer. We show that robots can learn word meanings and use those learned meanings to robustly follow natural language commands produced by untrained users. We demonstrate our approach for both mobility commands and mobile manipulation commands involving a variety of semi-autonomous robotic platforms, including a wheelchair, a micro-air vehicle, a forklift, and the Willow Garage PR2.

Soroush Vosoughi, Prashanth Vijayaraghavan, Deb Roy

We present Tweet2Vec, a novel method for generating general-purpose vector representation of tweets. The model learns tweet embeddings using character-level CNN-LSTM encoder-decoder. We trained our model on 3 million, randomly selected English-language tweets. The model was evaluated using two methods: tweet semantic similarity and tweet sentiment categorization, outperforming the previous state-of-the-art in both tasks. The evaluations demonstrate the power of the tweet embeddings generated by our model for various tweet categorization tasks. The vector representations generated by our model are generic, and hence can be applied to a variety of tasks. Though the model presented in this paper is trained on English-language tweets, the method presented can be used to learn tweet embeddings for different languages.

Prashanth Vijayaraghavan, Ivan Sysoev, Soroush Vosoughi et al.

This paper describes our approach for the Detecting Stance in Tweets task (SemEval-2016 Task 6). We utilized recent advances in short text categorization using deep learning to create word-level and character-level models. The choice between word-level and character-level models in each particular case was informed through validation performance. Our final system is a combination of classifiers using word-level or character-level models. We also employed novel data augmentation techniques to expand and diversify our training dataset, thus making our system more robust. Our system achieved a macro-average precision, recall and F1-scores of 0.67, 0.61 and 0.635 respectively.

Soroush Vosoughi, Helen Zhou, Deb Roy

The rise in popularity and ubiquity of Twitter has made sentiment analysis of tweets an important and well-covered area of research. However, the 140 character limit imposed on tweets makes it hard to use standard linguistic methods for sentiment classification. On the other hand, what tweets lack in structure they make up with sheer volume and rich metadata. This metadata includes geolocation, temporal and author information. We hypothesize that sentiment is dependent on all these contextual factors. Different locations, times and authors have different emotional valences. In this paper, we explored this hypothesis by utilizing distant supervision to collect millions of labelled tweets from different locations, times and authors. We used this data to analyse the variation of tweet sentiments across different authors, times and locations. Once we explored and understood the relationship between these variables and sentiment, we used a Bayesian approach to combine these variables with more standard linguistic features such as n-grams to create a Twitter sentiment classifier. This combined classifier outperforms the purely linguistic classifier, showing that integrating the rich contextual information available on Twitter into sentiment classification is a promising direction of research.

Soroush Vosoughi, Helen Zhou, Deb Roy

There is an ever growing number of users with accounts on multiple social media and networking sites. Consequently, there is increasing interest in matching user accounts and profiles across different social networks in order to create aggregate profiles of users. In this paper, we present models for Digital Stylometry, which is a method for matching users through stylometry inspired techniques. We experimented with linguistic, temporal, and combined temporal-linguistic models for matching user accounts, using standard and novel techniques. Using publicly available data, our best model, a combined temporal-linguistic one, was able to correctly match the accounts of 31% of 5,612 distinct users across Twitter and Facebook.

Soroush Vosoughi, Deb Roy

Speech acts are a way to conceptualize speech as action. This holds true for communication on any platform, including social media platforms such as Twitter. In this paper, we explored speech act recognition on Twitter by treating it as a multi-class classification problem. We created a taxonomy of six speech acts for Twitter and proposed a set of semantic and syntactic features. We trained and tested a logistic regression classifier using a data set of manually labelled tweets. Our method achieved a state-of-the-art performance with an average F1 score of more than $0.70$. We also explored classifiers with three different granularities (Twitter-wide, type-specific and topic-specific) in order to find the right balance between generalization and overfitting for our task.

Prashanth Vijayaraghavan, Soroush Vosoughi, Deb Roy

With the rise in popularity of public social media and micro-blogging services, most notably Twitter, the people have found a venue to hear and be heard by their peers without an intermediary. As a consequence, and aided by the public nature of Twitter, political scientists now potentially have the means to analyse and understand the narratives that organically form, spread and decline among the public in a political campaign. However, the volume and diversity of the conversation on Twitter, combined with its noisy and idiosyncratic nature, make this a hard task. Thus, advanced data mining and language processing techniques are required to process and analyse the data. In this paper, we present and evaluate a technical framework, based on recent advances in deep neural networks, for identifying and analysing election-related conversation on Twitter on a continuous, longitudinal basis. Our models can detect election-related tweets with an F-score of 0.92 and can categorize these tweets into 22 topics with an F-score of 0.90.

Soroush Vosoughi, Deb Roy

Twitter has become one of the main sources of news for many people. As real-world events and emergencies unfold, Twitter is abuzz with hundreds of thousands of stories about the events. Some of these stories are harmless, while others could potentially be life-saving or sources of malicious rumors. Thus, it is critically important to be able to efficiently track stories that spread on Twitter during these events. In this paper, we present a novel semi-automatic tool that enables users to efficiently identify and track stories about real-world events on Twitter. We ran a user study with 25 participants, demonstrating that compared to more conventional methods, our tool can increase the speed and the accuracy with which users can track stories about real-world events.