Annotating temporal information in texts is a challenging and time-consuming task. It requires an understanding of natural language, as well as knowledge about the various ways in which temporal data can be expressed and structured in a text. However, the ability to access temporal semantics through computer tools is crucial for many applications that involve interpreting and understanding texts.

A corpus available in this field is TimeBank (Pustejovsky et al., 2003), which was annotated using the TimeML annotation scheme (Pustejovsky et al., 2003), a scheme that does not support complex temporal expressions.

We proposed a new annotation scheme for temporal information in scientific texts: TimeInfo (Yahiaoui & Atanassova, 2022) which allows for more precise and directly usable annotations. The corpus presented here, named TimeTank, consists of 1186 sentences containing a total of 1200 temporal expressions annotated according to the TimeInfo annotation scheme. These sentences are drawn from 603 scientific articles from the CORD-19 corpus (Wang et al., 2020). The sentences were identified and annotated automatically, and the quality of the annotations was manually verified. TimeTank can be employed for the evaluation or training of machine learning models focused on the detection, extraction, and annotation of temporal expressions. The corpus offers a reliable dataset labeled to serve as a foundation for supervised learning.

Bibliography

Pustejovsky, James, et al. "The timebank corpus." Corpus linguistics. Vol. 2003. 2003. Pustejovsky, James, et al. "TimeML: Robust specification of event and temporal expressions in text." New directions in question answering 3 (2003): 28-34. Wang, Lucy Lu, et al. "Cord-19: The covid-19 open research dataset." ArXiv (2020). Yahiaoui, Salah, and Iana Atanassova. "TimeInfo: a Semantic Annotation Framework for Temporal Information in Scientific Papers." Terminology & Ontology: Theories and applications (TOTH 2022). 2022.