Recent studies have shown that deep vision-only and language-only models—trained on disjoint modalities—nonetheless project their inputs into a partially aligned representational space. Yet we still lack a clear picture of where in each network this convergence emerges, what visual or linguistic cues support it, and whether it endures the many-to-many nature of real image–text relationships. In this work, we systematically investigate these questions. We show that representational alignment emerges most strongly in mid-to-late layers of both vision and language models, suggesting a hierarchical progression from modality-specific to conceptually shared representations. Second, this alignment is robust to appearance-only changes but collapses when semantic content is altered—e.g., object removal in images or word order shuffling that disrupts thematic roles in sentences—highlighting that the shared code is truly semantic rather than form-based. Critically, we move beyond the conventional one-to-one image-caption paradigm to investigate alignment in many-to-many contexts, acknowledging that neither modality uniquely determines the other. Using a forced-choice "Pick-a-Pic" task, we find that human preferences for image-caption matches are mirrored in the learned embedding spaces across all vision-language model pairs. This pattern holds bidirectionally when multiple captions correspond to a single image, demonstrating that models capture fine-grained semantic distinctions similar to human judgments. Surprisingly, aggregating embeddings across multiple images or phrases referring to the same concept amplifies alignment. Rather than "blurring" representational detail, aggregation appears to distill a more universal semantic core. Together, these results demonstrate that vision and language networks converge on a shared semantic code, where the alignment mirrors human judgements, and becomes more pronounced when multiple exemplars of the same concept within a single modality are averaged in representational space. Our work provides compelling evidence for a universal code of meaning that transcends modality, offering critical insights into how neural networks represent and align semantic information across the vision-language divide.

Seeing Through Words, Speaking Through Pixels: Deep Representational Alignment Between Vision and Language Models

Lexical ambiguity---where a single wordform takes on distinct, context-dependent meanings--serves as a useful tool to compare across different language models' (LMs') ability to form distinct, contextualized representations of the same stimulus. Few studies have systematically compared LMs' contextualized word embeddings for languages beyond English. Here, we evaluate semantic representations of Spanish ambiguous nouns in context in a suite of Spanish-language monolingual and multilingual BERT-based models. We develop a novel dataset of minimal-pair sentences evoking the same or different sense for a target ambiguous noun. In a pre-registered study, we collect contextualized human relatedness judgments for each sentence pair. We find that various BERT-based LMs' contextualized semantic representations capture some variance in human judgments but fall short of the human benchmark. In exploratory work, we find that performance scales with model size. We also identify stereotyped trajectories of target noun disambiguation as a proportion of traversal through a given LM family's architecture, which we partially replicate in English. We contribute (1) a dataset of controlled, Spanish sentence stimuli with human relatedness norms, and (2) to our evolving understanding of the impact that LM specification (architectures, training protocols) exerts on contextualized embeddings.

Evaluating Contextualized Representations of (Spanish) Ambiguous Words: A New Lexical Resource and Empirical Analysis

The success of Large Language Models (LLMs) in other domains has raised the question of whether LLMs can reliably assess and manipulate the readability of text. We approach this question empirically. First, using a published corpus of 4,724 English text excerpts, we find that readability estimates produced “zero- shot” from GPT-4 Turbo and GPT-4o mini exhibit relatively high correlation with human judgments (r = 0.76 and r = 0.74, respec- tively), outperforming estimates derived from traditional readability formulas and various psycholinguistic indices. Then, in a pre-registered human experiment (N = 59), we ask whether Turbo can reliably make text easier or harder to read. We find evidence to support this hypothesis, though considerable variance in human judgments remains unexplained. We conclude by discussing the limitations of this approach, including limited scope, as well as the validity of the “readability” construct and its depen- dence on context, audience, and goal.

Measuring and Modifying the Readability of English Texts with GPT-4

Large Language Models (LLMs) have been criticized for failing to connect linguistic meaning to the worldâfor failing to solve the "symbol grounding problem." Multimodal Large Language Models (MLLMs) offer a potential solution to this challenge by combining linguistic representations and processing with other modalities. However, much is still unknown about exactly how and to what degree MLLMs integrate their distinct modalitiesâand whether the way they do so mirrors the mechanisms believed to underpin grounding in humans. In humans, it has been hypothesized that linguistic meaning is grounded through "embodied simulation," the activation of sensorimotor and affective representations reflecting described experiences. Across four pre-registered studies, we adapt experimental techniques originally developed to investigate embodied simulation in human comprehenders to ask whether MLLMs are sensitive to sensorimotor features that are implied but not explicit in descriptions of an event. In Experiment 1, we find sensitivity to some features (color and shape) and not others (size, orientation, and volume). In Experiment 2, we identify likely bottlenecks to explain an MLLM's lack of sensitivity. In Experiment 3, we find that despite sensitivity to implicit sensorimotor features, MLLMs cannot fully account for human behavior on the same task. Finally, in Experiment 4, we compare the psychometric predictive power of different MLLM architectures and find that ViLT, a single-stream architecture, is more predictive of human performance than CLIP, a dual-encoder architectureâdespite being trained on orders of magnitude less data. These results reveal strengths and limitations in the ability of current MLLMs to integrate language with other modalities, and also shed light on the likely mechanisms underlying human language comprehension.

Do Multimodal Large Language Models and Humans Ground Language Similarly?

Words in human languages cluster together in phonological neighborhoods more closely than would be expected by chance. But why? One explanation is that large neighborhoods are directly selected for, possibly because they scaffold word learning and production. But it's also possible that they emerge as a byproduct of other constraints or selection pressures operating over real lexica. We advance one such selection pressure as a candidate explanation. A pressure to avoid overloading unique wordforms with homophones may lead to clusters of words that are not identical but similar. Using simulated baselines, we test the viability of this alternative account. We find that a pressure against loading too many meanings on unique wordforms––paired with the phonotactics of a target language––produces lexica with neighborhoods that are at least as large on average as those in real lexica. This does not rule out the possibility of a pro-neighborhood pressure, but it does demonstrate the viability of a parsimonious alternative account based on a pressure against homonymy for which there is independent evidence.

Can a pressure against homophones explain phonological neighborhoods?

Most words are ambiguous––i.e., they convey distinct meanings in different contexts––and even the meanings of unambiguous words are context-dependent. Both phenomena present a challenge for NLP. Recently, the advent of contextualized word embeddings has led to success on tasks involving lexical ambiguity, such as Word Sense Disambiguation. However, there are few tasks that directly evaluate how well these contextualized embeddings accommodate the more continuous, dynamic nature of word meaning––particularly in a way that matches human intuitions. We introduce RAW-C, a dataset of graded, human relatedness judgments for 112 ambiguous words in context (with 672 sentence pairs total), as well as human estimates of sense dominance. The average inter-annotator agreement (assessed using a leave-one-annotator-out method) was 0.79. We then show that a measure of cosine distance, computed using contextualized embeddings from BERT and ELMo, correlates with human judgments, but that cosine distance also systematically underestimates how similar humans find uses of the same sense of a word to be, and systematically overestimates how similar humans find uses of different-sense homonyms. Finally, we propose a synthesis between psycholinguistic theories of the mental lexicon and computational models of lexical semantics.

Sean Trott

6

SHORT BIO

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES