There have been spectacular advances in many tasks of natural language processing (NLP) by making use of artificial intelligence (IA) techniques such as machine/deep learning (M/DL). However, these improvements do not affect all NLP tasks, specifically those that require deep linguistic knowledge, natural language understanding, semantic inference and reasoning. In hybrid architectures, M/DL approaches co-exist with symbol-manipulation as symbolic models seem to play an important role in inference and reasoning about abstract knowledge.
We call hybrid intelligence (HI) those architectures that integrate symbolic information into statistical or neural-based models so as to allow machines to learn new knowledge in a more 'intelligent' way by endowing them with common sense and deep understanding. The main aim of HI in NLP is to inject deep and structured linguistic knowledge (not just annotated text) into M/DL models so as to develop hybrid architectures for NLP tasks. In more general terms, the concept of HI consists of combining machine and human intelligence to overcome the shortcomings of existing AI systems. Abstract and structured knowledge from specialists can be used not just as training data to learn uninterpretable black-box models, but also to design the models themselves by making them more transparent, easy to interpret by humans, and more efficient for specific purposes.
HI4NLP will provide a forum for discussing about exciting research on HI methodology for NLP tasks. It is open to any contribution that requires deep semantic analysis, such as semantic relation extraction, discourse analysis, argument mining, rumour detection, and so on. Strategies can combine statistical or neural-based models with symbolic information based on propositions, regular patterns, rules, or whatever language resource aimed at representing abstract and structured knowledge.
HI4NLP should interest researchers working on the following topics and sub-topics of NLP by making use of HI approaches (the list is not exhaustive):
Submissions should describe original, unpublished work. Authors are invited to submit two kinds of papers:
Full papers – Reporting substantial and completed work, especially those that may contribute in a significant way to the advancement of the area. Wherever appropriate, concrete evaluation results should be included. Full papers can have up to 8 content pages (without references).
Short papers - Reporting small, focused contributions such as ongoing work, position papers, potential ideas to be discussed, negative results, or an interesting application nugget. Short papers can have up to 4 content pages + 1 page for references.
Each submission will be evaluated by at least three reviewers. As usual for the main ECAI conference, submissions are NOT anonymous. According to the template, names of authors should be stated in the manuscript.
Submissions should be written in English. At submission time, only PDF format is accepted. For the final versions, authors of accepted papers will be given 1 extra content page to take the reviews into account. Authors of accepted papers will be requested to send the source files for the production of the proceedings.
All submitted papers must conform to ECAI stylesheet
LyS Group, University of A Corunha (UDC) Associate Profesor
ERC Starting Grant by the European Research Council, project FASTPARSE (Fast Natural Language Processing for Large-Scale NLP).
Incipit- CSIC, Spanish National Research Council Staff Scientist
Leader of a co-research line on Software Engineering and Cultural Heritage, with special interest on Language and Knowledge connections.