The Cognitive Task Ontology (CogiTO)

The aim of COGITO is to relate Tasks from Cognitive Atlas to HED tags. In other words, we define the terms for cognitive tasks by their relation to HED tags, e.g. Reading Task is equivalent to (has some Language-item) AND (has some Read), where Reading Task is a Task from Cognitive Atlas and {Read, Language-item} are HED Tags. The purpose of this ontology is to query partially annotated datasets either by a Cognitive Task or by HED tags. Usage In order to use CogiTO following these steps: Download or Clone this respository Open the cogito.owl file with an OWL-compliant editor, such as Protégé. Check Imports: we reuse existing OWL files from Cognitive Atlas and HED by an OWL import. Sometimes the import fails, in which case it is necessary to reimport the ontologies. Enable ELK Reasoner in Protégé (recommended for reasoning, as other reasoners may time out due to ontology complexity): go to Window → Reasoner → ELK LLM Assistant To generate the mapping between Cognitive Atlas tasks (based on their natural language description) and the corresponding HED tags, we eployed a LLM (Gemini 2.5 Pro). The selection of Gemini 2.5 Pro was based on an evaluation of several LLMs (see llm-assistent/). Specifically, we compared the LLM-generated HED tag sets with expert annotations for 10 representative Cognitive Atlas tasks using the Jaccard similarity coefficient as a performance metric. We chose the following 10 tasks for the evaluation: Stroop task, Semantic decision task, Overt word repetition, Naming covert, Emotion recognition task, Tone counting, Associative memory encoding task, Dual-task weather prediction, Lexical decision task, Oddball task. For the expert annotations see human_annotation.csv. Evaluation In evaluation/ we implemented five different OWL reasoner all supporting the OWL 2 EL profile. To execute each of the reasoners just execute ./evaluate.sh (it might be necessary to execute chmod +x evaluate.sh first). The reasoners output an OWL file including the inferred facts (<reasoner>_inferred_output.owl). In the evaluation script a timeout is set to 1200s, since all reasoners - besides the ELK reasoner - time out. Queries In queries/ we provide some generated test queries in CRPQ syntax for testing. The queries can be divided into five group of queries: G1 - querying for datasets that have a path to a certain taskG2 - querying for datasets that have a path to a certain HED tagG3 - querying for datasets that have paths to two HED tagsG4 - querying for a certain HED tagG5 - querying for a certain task Data The ontology file itself (ontologies/cogito.owl) contains a set of example individuals to demonstrate the use of the Protégé build-in reasoners. However, in theory each BIDS formatted dataset provides potential instances to CogiTO. The RDF data contain the meta-data from OpenNeuro datasets, which we exported by the so-called BIDS Indexer (Research Edition).