Pragmatic inferencing influences the referential status of all potential referents in word learning

After observing a word uttered in the presence of multiple objects, how do learners represent these potential referents? Given a select object from a learning scenario for a novel word, all theories predict that this referent is to be remembered as the referent, but diverge on how the co-present objects are represented. Theory-building is also hampered by the fact that different studies used different stimuli, in addition to whether the learning instances contained referential information on which the referent is intended. Here, we report four experiments with approximately 100 adult participants in each, manipulating both factors. We find that, when the potential referents are known conceptually, the referential status of these objects is influenced by whether the learning instances contain referential information: With information guiding referential selection, learners perform mutual exclusivity and reject the co-occurring objects as potential referents; without this information, learners represent all co-occurring objects as a potential referent. However, when the potential referents are unfamiliar objects, even with the explicit instruction on which object is the intended referent, learners represent both objects as potential referents, as both objects can plausibly belong to the same category that the word refers to. This set of results can be explained by a pragmatic inference process where learners either believe all potential referents belong to different categories and are thus mutually exclusive, or that all potential referents plausibly belong to the same category referred to by the same word. We discuss the implication of these results, which move away from a debate about the sheer number of referents and onto the more mechanistic question of how learners represent referents in word learning.