Quantifying word informativeness and its impact on eye-movement reading behavior: Cross-linguistic variability and...

Area of Science:

• Psycholinguistics
• Computational Linguistics
• Cognitive Science

Background:

• Word importance influences reading behavior, but existing quantification methods lack interpretability.
• Subjective ratings and computational solutions have limitations in objectively measuring word centrality.

Purpose of the Study:

• Introduce a novel, objective, and interpretable measure of word significance within a sentence, termed 'informativeness'.
• To assess how this 'informativeness' measure relates to reading behavior and other psycholinguistic variables.

Main Methods:

• Developed a new measure ('informativeness') based on comparing vector representations of full sentences with sentences excluding a target word.
• Utilized large-scale eye-tracking datasets (Multilingual Eye-tracking Corpus, MECO) for first (L1) and second language (L2) readers.

Main Results:

• The 'informativeness' measure shows expected correlations with psycholinguistic variables like frequency and predictability.
• Informativeness uniquely predicts eye-movement reading behavior in both L1 and L2 readers.
• Effects of informativeness generalize across writing systems and are more pronounced in less proficient readers.

Conclusions:

• The 'informativeness' measure provides a valuable tool for studying word significance in reading.
• Findings deepen the understanding of how word importance impacts reading processes across diverse linguistic contexts and reader proficiencies.