Measuring the importance of context when modeling language comprehension.

It is widely accepted that language requires context in order to function as communication between speakers and listeners. As listeners, we make use of background knowledge - about the speaker, about entities and concepts, about previous utterances - in order to infer the speaker's intended meaning. But even if there is consensus that these sources of information are a necessary component of linguistic communication, it is another matter entirely to provide a thorough, quantitative accounting for context's interaction with language. When does context matter? What kinds of context matter in which kinds of domains? The empirical investigation of these questions is inhibited by a number of factors: the challenge of quantifying language, the boundless combinations of domains and types of context to be measured, and the challenge of selecting and applying a given construct to natural language data. In response to these factors, we introduce and demonstrate a methodological framework for testing the importance of contextual information in inferring speaker intentions from text. We apply Long Short-term Memory (LSTM) networks, a standard for representing language in its natural, sequential state, and conduct a set of experiments for predicting the persuasive intentions of speakers in political debates using different combinations of text and background information about the speaker. We show, in our modeling and discussion, that the proposed framework is suitable for empirically evaluating the manner and magnitude of context's relevance for any number of domains and constructs.

Processing Narratives Concerning Protected Values: A Cross-Cultural Investigation of Neural Correlates.

Narratives are an important component of culture and play a central role in transmitting social values. Little is known, however, about how the brain of a listener/reader processes narratives. A receiver's response to narration is influenced by the narrator's framing and appeal to values. Narratives that appeal to "protected values," including core personal, national, or religious values, may be particularly effective at influencing receivers. Protected values resist compromise and are tied with identity, affective value, moral decision-making, and other aspects of social cognition. Here, we investigated the neural mechanisms underlying reactions to protected values in narratives. During fMRI scanning, we presented 78 American, Chinese, and Iranian participants with real-life stories distilled from a corpus of over 20 million weblogs. Reading these stories engaged the posterior medial, medial prefrontal, and temporo-parietal cortices. When participants believed that the protagonist was appealing to a protected value, signal in these regions was increased compared with when no protected value was perceived, possibly reflecting the intensive and iterative search required to process this material. The effect strength also varied across groups, potentially reflecting cultural differences in the degree of concern for protected values.

Analyzing Public Conversations About Heart Disease and Heart Health on Facebook From 2016 to 2021: Retrospective Observational Study Applying Latent Dirichlet Allocation Topic Modeling.

BACKGROUND: Heart disease continues to be the leading cause of death in men and women in the United States. The COVID-19 pandemic has further led to increases in various long-term cardiovascular complications. OBJECTIVE: This study analyzed public conversations related to heart disease and heart health on Facebook in terms of their thematic topics and sentiments. In addition, it provided in-depth analyses of 2 subtopics with important practical implications: heart health for women and heart health during the COVID-19 pandemic. METHODS: We collected 34,885 posts and 51,835 comments spanning from June 2016 to June 2021 that were related to heart disease and health from public Facebook pages and groups. We used latent Dirichlet allocation topic modeling to extract discussion topics illuminating the public's interests and concerns regarding heart disease and heart health. We also used Linguistic Inquiry and Word Count (Pennebaker Conglomerates, Inc) to identify public sentiments regarding heart health. RESULTS: We observed an increase in discussions related to heart health on Facebook. Posts and comments increased from 3102 and 3632 in 2016 to 8550 (176% increase) and 14,617 (302% increase) in 2021, respectively. Overall, 35.37% (12,340/34,885) of the posts were created after January 2020, the start of the COVID-19 pandemic. In total, 39.21% (13,677/34,885) of the posts were by nonprofit health organizations. We identified 6 topics in the posts (heart health promotion, personal experiences, risk-reduction education, heart health promotion for women, educational information, and physicians' live discussion sessions). We identified 6 topics in the comments (personal experiences, survivor stories, risk reduction, religion, medical questions, and appreciation of physicians and information on heart health). During the pandemic (from January 2020 to June 2021), risk reduction was a major topic in both posts and comments. Unverified information on alternative treatments and promotional content was also prevalent. Among all posts, 14.91% (5200/34,885) were specifically about heart health for women centering on local event promotion and distinctive symptoms of heart diseases for women. CONCLUSIONS: Our results tracked the public's ongoing discussions on heart disease and heart health on one prominent social media platform, Facebook. The public's discussions and information sharing on heart health increased over time, especially since the start of the COVID-19 pandemic. Various levels of health organizations on Facebook actively promoted heart health information and engaged a large number of users. Facebook presents opportunities for more targeted heart health interventions that can reach and engage diverse populations.

Neural correlates of word representation vectors in natural language processing models: Evidence from representational similarity analysis of event-related brain potentials.

Natural language processing models based on machine learning (ML-NLP models) have been developed to solve practical problems, such as interpreting an Internet search query. These models are not intended to reflect human language comprehension mechanisms, and the word representations used by ML-NLP models and human brains might therefore be quite different. However, because ML-NLP models are trained with the same kinds of inputs that humans must process, and they must solve many of the same computational problems as the human brain, ML-NLP models and human brains may end up with similar word representations. To distinguish between these hypotheses, we used representational similarity analysis to compare the representational geometry of word representations in two ML-NLP models with the representational geometry of the human brain, as indexed with event-related potentials (ERPs). Participants listened to stories while the electroencephalogram was recorded. We extracted averaged ERPs for each of the 100 words that occurred most frequently in the stories, and we calculated the similarity of the neural response for each pair of words. We compared this 100 × 100 similarity matrix to the 100 × 100 similarity matrix for the word pairs according to two ML-NLP models. We found significant representational similarity between the neural data and each ML-NLP model, beginning within 250 ms of word onset. These results indicate that ML-NLP systems that are designed to solve practical technology problems have a representational geometry that is correlated with that of the human brain, presumably because both are influenced by the structural properties and statistics of language.

An intelligent virtual human system for providing healthcare information and support.

Over the last 15 years, a virtual revolution has taken place in the use of Virtual Reality simulation technology for clinical purposes. Shifts in the social and scientific landscape have now set the stage for the next major movement in Clinical Virtual Reality with the "birth" of intelligent virtual humans. Seminal research and development has appeared in the creation of highly interactive, artificially intelligent and natural language capable virtual human agents that can engage real human users in a credible fashion. No longer at the level of a prop to add context or minimal faux interaction in a virtual world, virtual humans can be designed to perceive and act in a 3D virtual world, engage in spoken dialogues with real users and can be capable of exhibiting human-like emotional reactions. This paper will present an overview of the SimCoach project that aims to develop virtual human support agents to serve as online guides for promoting access to psychological healthcare information and for assisting military personnel and family members in breaking down barriers to initiating care. The SimCoach experience is being designed to attract and engage military Service Members, Veterans and their significant others who might not otherwise seek help with a live healthcare provider. It is expected that this experience will motivate users to take the first step--to empower themselves to seek advice and information regarding their healthcare and general personal welfare and encourage them to take the next step towards seeking more formal resources if needed.

Tracking Child Language Development With Neural Network Language Models.

Recent work on the application of neural networks to language modeling has shown that models based on certain neural architectures can capture syntactic information from utterances and sentences even when not given an explicitly syntactic objective. We examine whether a fully data-driven model of language development that uses a recurrent neural network encoder for utterances can track how child language utterances change over the course of language development in a way that is comparable to what is achieved using established language assessment metrics that use language-specific information carefully designed by experts. Given only transcripts of child language utterances from the CHILDES Database and no pre-specified information about language, our model captures not just the structural characteristics of child language utterances, but how these structures reflect language development over time. We establish an evaluation methodology with which we can examine how well our model tracks language development compared to three known approaches: Mean Length of Utterance, the Developmental Sentence Score, and the Index of Productive Syntax. We discuss the applicability of our model to data-driven assessment of child language development, including how a fully data-driven approach supports the possibility of increased research in multilingual and cross-lingual issues.

Evaluating contributions of natural language parsers to protein-protein interaction extraction.

MOTIVATION: While text mining technologies for biomedical research have gained popularity as a way to take advantage of the explosive growth of information in text form in biomedical papers, selecting appropriate natural language processing (NLP) tools is still difficult for researchers who are not familiar with recent advances in NLP. This article provides a comparative evaluation of several state-of-the-art natural language parsers, focusing on the task of extracting protein-protein interaction (PPI) from biomedical papers. We measure how each parser, and its output representation, contributes to accuracy improvement when the parser is used as a component in a PPI system. RESULTS: All the parsers attained improvements in accuracy of PPI extraction. The levels of accuracy obtained with these different parsers vary slightly, while differences in parsing speed are larger. The best accuracy in this work was obtained when we combined Miyao and Tsujii's Enju parser and Charniak and Johnson's reranking parser, and the accuracy is better than the state-of-the-art results on the same data. AVAILABILITY: The PPI extraction system used in this work (AkanePPI) is available online at http://www-tsujii.is.s.u-tokyo.ac.jp/downloads/downloads.cgi. The evaluated parsers are also available online from each developer's site.

Morphosyntactic annotation of CHILDES transcripts.

Corpora of child language are essential for research in child language acquisition and psycholinguistics. Linguistic annotation of the corpora provides researchers with better means for exploring the development of grammatical constructions and their usage. We describe a project whose goal is to annotate the English section of the CHILDES database with grammatical relations in the form of labeled dependency structures. We have produced a corpus of over 18,800 utterances (approximately 65,000 words) with manually curated gold-standard grammatical relation annotations. Using this corpus, we have developed a highly accurate data-driven parser for the English CHILDES data, which we used to automatically annotate the remainder of the English section of CHILDES. We have also extended the parser to Spanish, and are currently working on supporting more languages. The parser and the manually and automatically annotated data are freely available for research purposes.

Incorporating Demographic Embeddings Into Language Understanding.

Meaning depends on context. This applies in obvious cases like deictics or sarcasm as well as more subtle situations like framing or persuasion. One key aspect of this is the identity of the participants in an interaction. Our interpretation of an utterance shifts based on a variety of factors, including personal history, background knowledge, and our relationship to the source. While obviously an incomplete model of individual differences, demographic factors provide a useful starting point and allow us to capture some of this variance. However, the relevance of specific demographic factors varies between situations-where age might be the key factor in one context, ideology might dominate in another. To address this challenge, we introduce a method for combining demographics and context into situated demographic embeddings-mapping representations into a continuous geometric space appropriate for the given domain, showing the resulting representations to be functional and interpretable. We further demonstrate how to make use of related external data so as to apply this approach in low-resource situations. Finally, we show how these representations can be incorporated into improve modeling of real-world natural language understanding tasks, improving model performance and helping with issues of data sparsity.

Automatic parsing of parental verbal input.

To evaluate theoretical proposals regarding the course of child language acquisition, researchers often need to rely on the processing of large numbers of syntactically parsed utterances, both from children and from their parents. Because it is so difficult to do this by hand, there are currently no parsed corpora of child language input data. To automate this process, we developed a system that combined the MOR tagger, a rule-based parser, and statistical disambiguation techniques. The resultant system obtained nearly 80% correct parses for the sentences spoken to children. To achieve this level, we had to construct a particular processing sequence that minimizes problems caused by the coverage/ambiguity tradeoff in parser design. These procedures are particularly appropriate for use with the CHILDES database, an international corpus of transcripts. The data and programs are now freely available over the Internet.