Live presentation

Asynchronous discussion


Online assessments of larger contexts modulation in word Predictability in Short Stories

Lena Valeska Peroni

  • Buenos Aires,
  • Argentina
  • Lena V Peroni ¹
  • , Bruno Bianchi ¹
  • , Diego E Shalom ²
  • , Juan E Kamienkowski ¹
  • , Juan E Kamienkowski ²
  • 1 Laboratorio de Inteligencia Artificial Aplicada, Instituto de Ciencias de la Computación, FCEyN, UBA - CONICET
  • 2 Departamento de Física, FCEyN, UBA

When reading, we are continuously predicting upcoming words, allowing us to integrate information and to guide eye movements. These predictions are operationalized as the Predictability and measured as the probability of guessing the next word from the previous context. Predictability reflects high-level processes taking place during reading and is a strong predictor of eye movements and brain activity. It is estimated by asking many participants to complete the most probable word that follows a given context. This task is called cloze-task and is usually collected in-lab, resulting in a time-consuming and expensive experiment.Here, we present an analysis of three corpora of online cloze-task experiments: (C1) a corpus of Common and Memory-Encoded Sentences collected both online and in-lab, (C2) a corpus with similar material, collected in two independent online experiments, and (C3) a corpus of sentences drawn from short stories collected online, both isolated or contextualized.We observed that the online cloze-task replicates the results from the in-lab one (C1) and is consistent between independent experiments (C2) Interestingly, these results clearly show that Predictability is highly dependent on large contexts (C3) Thus, online cloze-task makes the collection for larger samples easy and generates robust and more precise measures. Moreover, it allows us to explore the effects of larger contexts (up to 3000 words) in the Predictability, that could be impossible otherwise.