Transactions of the Japanese Society for Artificial Intelligence
Online ISSN : 1346-8030
Print ISSN : 1346-0714
ISSN-L : 1346-0714
Original Paper
Pic2PLex: A Test for Assessing Free Productive Vocabulary
Dolc¸a TellolsTakenobu TokunagaHilofumi YamamotoHikaru Yokono
Author information

2023 Volume 38 Issue 1 Pages B-M52_1-11


We present Pic2PLex, a new test for assessing free productive vocabulary, the lexicon people use for writing and speaking. Unlike the existing productive vocabulary tests, Pic2PLex uses images as stimuli to let test-takers say something. A test item comprises a set of six different images with a common theme and two answer sections: a tenword section and a description section. Given the image set, test-takers are instructed to write ten words that come to their minds and a brief description of the images in the corresponding sections. We propose an algorithm to generate image sets for the test items from a large image dataset. The algorithm was designed to simultaneously achieve two conflicting goals: making images of a test item share the same theme and making themes among image sets across test items diverse as much as possible. We conducted experiments with Japanese language learners and native Japanese speakers to collect responses using Pic2PLex and existing productive vocabulary tests for comparison. Quantitative and qualitative investigation revealed that (i) Pic2PLex elicits words based on productive lexical knowledge that the test-takers would have acquired, and (ii) the elicited responses can be used in assessments that measure how much productive vocabulary the test-taker has.

Content from these authors
© The Japanese Society for Artificial Intelligence 2023
Previous article Next article