Multilevel Annotation of Informativeness in Linguistic Corpora: INFOLEXIS

Multilevel Annotation of Informativeness in Linguistic Corpora: INFOLEXIS Olga Batiukova, Keigh Rim & James Pustejovsky Autonomous University of Madrid, Brandeis University [email protected], [email protected], [email protected] Acknowledgements Overview •Semantic annotation of informativeness-based constraints on predicate construction. •Acceptable phrases must be consistent and informative. •Annotated corpora: COCA and Corpus del Español Web/Dialects Annotation scheme Annotated constructions (1) a. Eat ravioli – eat ??(organic) food b. {Rewritten/ published} novel - ??(recently) written novel c. This novel was {rewritten / published} – This novel was written ??(recently) d. This book is yellowing - This book reads ??(fast). Annotation examples STEP 1. All the examples in (1) are consistent. Inconsistent example: The novel was devoured. STEP 2. First member of each pair in (1): informative. Second member: uninformative. STEP 3. Additional modifiers in uninformative expressions: ‘recently read novel’, ‘this book reads fast’. Contrastive interpretation of uninformative expressions: ‘a WRITTEN novel’ is opposed to the novels that never came to exist. Verb classes: acquisition, caused motion, communication, consumption, creation, physical change of state, perception, psychological state, putting, transfer, light. INFOLEXIS Corpus Searcher (Step 1) Main contributions of INFOLEXIS Related to the CLARIN objectives: 1. Novel multilevel annotation scheme applicable to all natural languages. 2. 3 manually annotated datasets: •Syntactic constructions classified as (in)consistent and linked to their original syntactic context (STEP 1). •Consistent constructions classified as (un)informative (STEP 2). •Uninformative constructions with syntactic components that make them acceptable (STEP 3). 3. Annotation tools that can be reused in other manual annotation projects. Theoretical contributon: •A valuable empirical tool for the study of informativeness as an interface phenomenon, involving lexical-semantic, syntactic, and pragmatic factors. •INFOLEXIS contributes to the understanding of the mechanisms underlying selectional constraints at the phrasal and sentential level. Selected references •Batiukova, O. (2024). “Restricciones léxicas sobre la informatividad de la construcción <nombre + participio>”, Verba, 51. •Batiukova, O., & Pustejovsky, J. (2013). “Informativeness constraints and compositionality”, Proceedings of GL2013, 92–100. •Lahiri, S. (2015). “Squinky! a corpus of sentence-level formality, informativeness, and implicature”, arXiv preprint arXiv:1506.02306 •Molina, A., et al. (2013). “Discursive sentence compression”, International Conference on Intelligent Text Processing and Computational Linguistics. •Nishikawa, H., et al. (2010). “Optimizing informativeness and readability for sentiment summarization”, Proceedings of the ACL 2010 Conference. •Pustejovsky, J., & Stubbs, A. (2012). Natural language annotation for machine learning, O’Reilly Media. •Rennie, J. D., & Jaakkola, T. (2005). “Using term informativeness for named entity detection”. Proceedings of the 28th ACM SIGIR conference on Research and development in information retrieval. This presentation has been funded by: • MCIN/AEI/10.13039/501100011033 and FSE+ under the project PID2022-138135NB-I00. •Comunidad de Madrid under the agreement “The node CLARIAHCM: Digital Humanities and Language Techonologies” (4180134).