Distributional semantic models (DSMs) are a primary method for distilling semantic information from corpora. However, a key question remains: What types of semantic relations among words do DSMs detect? Prior work typically has addressed this question using limited human data that are restricted to semantic similarity and/or general semantic relatedness. We tested eight DSMs that are popular in current cognitive and psycholinguistic research (positive pointwise mutual information; global vectors; and three variations each of Skip-gram and continuous bag of words (CBOW) using word, context, and mean embeddings) on a theoretically motivated, rich set of semantic relations involving words from multiple syntactic classes and spanning the abstract-concrete continuum (19 sets of ratings). We found that, overall, the DSMs are best at capturing overall semantic similarity and also can capture verb-noun thematic role relations and noun-noun event-based relations that play important roles in sentence comprehension. Interestingly, Skip-gram and CBOW performed the best in terms of capturing similarity, whereas GloVe dominated the thematic role and event-based relations. We discuss the theoretical and practical implications of our results, make recommendations for users of these models, and demonstrate significant differences in model performance on event-based relations.
In addition to the many easily observable differences between people, there are also differences in people’s subjective experiences that are harder to observe, and which, as a consequence, remain hidden. For example, people vary widely in how much visual imagery they experience. But those who cannot see in their mind’s eye, tend to assume everyone is like them. Those who can, assume everyone else can as well. We argue that a study of such hidden phenomenal differences has much to teach cognitive science. Uncovering and describing this variation (a search for unknown unknowns) may help predict otherwise puzzling differences in human behavior. The very existence of certain differences can also act as a stress test for some cognitive theories. Finally, studying hidden phenomenal differences is the first step toward understanding what kinds of environments may mask or unmask links between phenomenal experience and observable behavior.
Theoretical accounts of negative expressives such as damn have ascribed two main properties to this type of adjective, namely that they are typically speaker-oriented, and that they can be flexible with regard to their syntactic attachment. However, it is not clear what this means during online sentence processing. For example, is it effortful for comprehenders to derive the speaker's negative attitude conveyed by an expressive adjective, or is it a rapid, automatic process? And do comprehenders understand the speaker's attitude regardless of the expressive's syntactic position? The current work provides the first evidence supporting theoretical claims by investigating the incremental processing of Italian negative expressive adjectives. In an eye-tracking study, we show that expressive content is rapidly integrated with information about the speaker's attitude, resulting in the anticipation of an upcoming referent, regardless of the expressive's syntactic realization. We argue that comprehenders use expressives as an ostensive cue that allows for automatic retrieval of the speaker's negative attitude.
Recent work in cognitive psychology and experimental semantics indicates that people do not categorize natural kinds solely by virtue of their purported scientific essence. Two attempts have been made to explain away the data by appealing to the idea that participants in these studies are reasoning diagnostically. I will argue that an appeal to diagnostic reasoning will likely not help to explain away the data.
Interpreting a seemingly simple function word like "or," "behind," or "more" can require logical, numerical, and relational reasoning. How are such words learned by children? Prior acquisition theories have often relied on positing a foundation of innate knowledge. Yet recent neural-network-based visual question answering models apparently can learn to use function words as part of answering questions about complex visual scenes. In this paper, we study what these models learn about function words, in the hope of better understanding how the meanings of these words can be learned by both models and children. We show that recurrent models trained on visually grounded language learn gradient semantics for function words requiring spatial and numerical reasoning. Furthermore, we find that these models can learn the meanings of logical connectives and and or without any prior knowledge of logical reasoning as well as early evidence that they are sensitive to alternative expressions when interpreting language. Finally, we show that word learning difficulty is dependent on the frequency of models' input. Our findings offer proof-of-concept evidence that it is possible to learn the nuanced interpretations of function words in a visually grounded context by using non-symbolic general statistical learning algorithms, without any prior knowledge of linguistic meaning.
