JWSAN: Japanese word similarity and association norm

We present a new Japanese dataset, Japanese Word Similarity and Association Norm (JWSAN), comprising human rating scores of similarity and association for 2145 word pairs, with a clear distinction between word similarity and word association. Computational models of human semantic memory or mental lexicon, such as distributed semantic models, must predict not only association but also similarity. People can distinguish between word similarity and association. However, although the SimLex-999 dataset is publicly available for English, there is no Japanese similarity dataset with a clear distinction between the two types of word relatedness. JWSAN is the first large Japanese dataset with similarity and association ratings, containing noun, verb, and adjective word pairs. It is also characterized by data collection from a sufficient number of age- and-gender-controlled assessors, with similarity and association ratings obtained via a web-based survey conducted of 6450 native speakers of Japanese. In addition, the effects of the gender and age of the raters were also examined; these factors were only given scant consideration in the past. This dataset can act as a benchmark for improving distributed semantic models in Japanese.

The Nativelikeness Problem in L2 Word-association Tasks: Examining Word Class and Trials

Significant variation exists in how native speakers respond to word association tasks and challenges the usage of nativelikeness as a benchmark to gauge second language (L2) performance. However, the influence of word class and trials of elicitation is not sufficiently addressed in previous work. With controlled stimuli from multiple word classes, repeated elicitations, and analytic approaches aiming to tease apart their interactions, this study compared the extent to which native speaker controls and late L2 learners generated associates that converged to a large-scale association norm, and examined the influence of word class and trial on the likelihood to elicit idiosyncratic responses within the two language groups. During initial elicitation, only adjectives elicited greater convergence to the norm among native speakers than L2 learners. Furthermore, native speakers were more likely to generate synonyms whereas L2 learners were more likely to generate antonyms to adjectives in the initial elicitation. For nouns and verbs, 30% of associates produced by the native speaker controls failed to converge to the norm. In fact, the native speaker controls were not more “nativelike” than L2 learners for nouns and verbs until later elicitations. Finally, despite reports of significant variation among native speakers in previous work, the amount of response idiosyncrasy was consistently lower in native speakers than in L2 learners, regardless of word class or elicitation trial. By revealing the effects of word class and trials on association performance, findings from this study suggest potential means to ameliorate the issue with nativelikeness in L2 word association studies.