Towards the task of factuality assessment

Factuality concerns the extent to which the propositional content of a sentence conforms with the real world, according to the speaker.Factuality assessment is of practical interest for those NLP applications that have to do with textual information analysis: information extraction, information retrieval, text summarization, question-answering systems.Existing practical approaches address the problem of factuality assessment using the "quantitative" measures of certainty and probability and thus have limited possibilities for annotating real texts.There is a need for a more appropriate model of factuality that could serve as the basis of the annotation scheme.We suggest a model of factuality that makes use of speaker's cognitive attitudes as one of the parameters for discriminating between degrees of factuality.We also present a dataset that was manually annotated with factuality values in terms of the model.