Recent studies suggest that similarity in emotional features and concreteness are critical cues underlying word association in native speakers. However, the lexical organization of a foreign language is less understood. This study aims to examine the structure of word associations within the mental lexicon of a foreign (English) and a native language. To this end, 145 native Spanish-speakers produced three lexical associates to cue words in both the foreign and native language. We observed that the associates were more neutrally valenced in the foreign language. Moreover, as cue words increased in their arousal, the produced associates were less arousing in the foreign language. Thus, the structure of these lexical associations could account for prior evidence of emotional detachment in foreign languages. Finally, as cues were more abstract, the foreign language associates were more abstract. Our findings revealed that the linguistic context modulated the lexical associations.

Current approaches to learning semantic representations of sentences often use prior word-level knowledge. The current study aims to leverage visual information in order to capture sentence level semantics without the need for word embeddings. We use a multimodal sentence encoder trained on a corpus of images with matching text captions to produce visually grounded sentence embeddings. Deep Neural Networks are trained to map the two modalities to a common embedding space such that for an image the corresponding caption can be retrieved and vice versa. We show that our model achieves results comparable to the current state of the art on two popular image-caption retrieval benchmark datasets: Microsoft Common Objects in Context (MSCOCO) and Flickr8k. We evaluate the semantic content of the resulting sentence embeddings using the data from the Semantic Textual Similarity (STS) benchmark task and show that the multimodal embeddings correlate well with human semantic similarity judgements. The system achieves state-of-the-art results on several of these benchmarks, which shows that a system trained solely on multimodal data, without assuming any word representations, is able to capture sentence level semantics. Importantly, this result shows that we do not need prior knowledge of lexical level semantics in order to model sentence level semantics. These findings demonstrate the importance of visual information in semantics.

Oral language in individuals with nonverbal learning disabilities (NLD) has been described as empty of meaning, despite apparently average word knowledge. The present study explored the hypothesis that depth but not breadth of semantic representations would be reduced in NLD, and that depth but not breadth would be related to nonverbal gestalt perception. A cross-sectional design compared breadth and depth of vocabulary in 50 adults with or without a diagnosis of NLD. Vocabulary results were also compared with a visual closure test. Participants with NLD had reduced vocabulary depth in comparison with controls. The NLD group also had lower scores for gestalt perception, the ability to perceive a meaningful whole from unrelated parts. Across the sample, this measure predicted scores for vocabulary depth, but not breadth. The NLD group was also less able than the Control group to estimate the size of unknown, physical features of everyday objects. Results supported clinical observations that semantic representations are unconventional and imprecise in individuals with NLD, and suggested specific cognitive underpinnings for such difficulties. Results were also compatible with separate theories of embodied and lateralized semantics. A proposal uniting these theories in a designation over elaboration model is presented.