About this Research Topic
An increasing number of works in the NLP community is focused on the automatic analysis of highly subjective phenomena, where the perception and socio-cultural background of the recipient of the messages play a critical role. In order to achieve this, the analysis of subtle features of communication is needed, including sentiment, emotion, identity, stance, irony, persuasive and deceptive language, toxic language, among others. The interpretation of natural language beyond its standing meaning becomes key to understanding the speaker’s stance or the veracity of a news item.
Modeling different perspectives is also the next step for Natural Language Understanding (NLU). On the one hand, next generation NLU systems need to be able to discern objective statements of factual events from opinions and loaded questions and statements. In other words, the meaning of a natural language expression needs to be modeled as a function of the personal point of view of its author. On the other hand, modern NLU models are largely based on Language Resources (LRs) and Machine Learning (ML). LRs are usually annotated by humans, who inevitably project the perspective of their personal background into the data. Moreover, the raw data, often collected automatically from Web sources, may reflect different perspectives due to the source itself, or the collection strategies adopted to create the corpora. Such factors are subsequently picked up by supervised ML models. Therefore, there is a growing need to capture and understand possible bias in the data and eliminate or keep under control its effect on the prediction capabilities of the models.
All of the above sources of subjectivity in textual data force us to re-think the very basics of the current practices employed in the design and evaluation of NLP systems, as well as their possible impact upon deployment in the real world. Furthermore, this line of research highlights the need to develop new technologies to assist people in conducting healthier online social interactions and obtaining reliable information.
The field is in dire need of a venue for research in the above directions, which would welcome a balanced combination of NLP technical rigor with fine-grained analysis of the social aspects of meaning and interpretation. At present, this is not contemplated by mainstream NLP conferences. This Research Topic aims at painting a wide picture of the state of studies on the expression and modeling of subjectivity in natural language from a computational standpoint, and fostering an interdisciplinary community of scholars actively working on this topic from different angles.
We welcome contributions on topics such as:
• Bias in language data
• Bias in NLP models
• Ethics and fairness in AI models
• Real-world effects of biased models
• Perspective extraction
• Opinion Mining & Aspect-based Sentiment Analysis
• Bias through affect expression in text
• Fact-checking and features of disinformation
• Stylometry (Author attribution and author profiling)
• Disagreement-aware Crowdsourcing
• Narrative Framing
• Narrative Understanding
• Computational Discourse Analysis
• Irony and Sarcasm
• Abusive Language Processing
• Contextualized meaning representations
• Language Resources for modeling social phenomena and perspectives
The Research Topic welcomes contributions in the form of theoretical or experimental papers, system demonstrations, as well as position papers as long as they fit within the topics of the call.
Keywords: NLP, Sentiment Analysis, Bias, Subjectivity, Natural Language Understanding, Computational Linguistics, Language Resources, Machine Learning, Opinion Mining, Discourse Analysis
Important Note: All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements. Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review.