MegaIntensionality Validation Study

Authors: Benjamin Kane, William Gantt, and Aaron Steven White


Version: 1.0

Release date: 8 May 2021


This dataset contains the raw annotator judgments from the validation study described in the following paper:

Kane, Benjamin & W. Gantt & A. S. White. 2021. Intensional gaps: relating veridicality, factivity, doxasticity, bouleticity, and neg-raising. Accepted to Semantics and Linguistic Theory 31.

This study aimed to validate a “bleaching” method for collecting judgments about lexically triggered belief and desire inferences. We presented participants on Amazon Mechanical Turk with both traditional contentful sentences (e.g. the soccer player loved that her team had scored the winning goal) and with low-content, templatic ones (e.g. A loved that C happened) and asked about the likelihood that subject or object believes or desires the content of the embedded clause. Details of the item construction and the experimental setup can be found in that paper, and we kindly ask that you cite it if you make use of this data in a presentation or publication.

Version history:

1.0: first public release, 8 May 2021



Column Description Values
participant anonymous integer identifier for participant that provided the response 0…319
listid integer identifier for list participant was responding to 0…31
question the text of the question (antecedent and consequent) posed to annotators see paper
root lemma of clause-embedding verb found in the antecedent sentence see paper
subject the subject of the consequent sentence see paper
target the target of the inference judgment subject, object
consequent lemma of the verb in the consequent sentence (denoting the inference type) believe, want
polarity the matrix polarity of the antecedent sentence positive, negative
valence the valence of the content of the antecedent sentence positive, negative, neutral
tense the tense of the consequent verb future, past
transitivity transitivity of the frame of the antecedent transitive, intransitive
response the participant’s raw response giving the likelihood of the inference 0.0…1.0

Note: A value of neutral for the valence property indicates that the item for that row featured a templatic antecedent (e.g. A loved that C happened) and consequent (e.g. A wanted C to have happened). By contrast, a value of positive or negative indicates that the item for that row featured a contentful antecedent (e.g. the soccer player loved that her team had scored the winning goal) and consequent (e.g. the soccer player wanted her team to have scored the winning goal).