In everyday language use, speakers frequently utter and interpret sentences that are semantically underspecified, namely, whose content is insufficient to fully convey their message or interpret them univocally. For example, to interpret the underspecified sentence “Don’t spend too much”, which leaves implicit what (not) to spend, additional linguistic context or outside knowledge is needed. In this work, we propose a novel Dataset of semantically Underspecified Sentences grouped by Type (DUST) and use it to study whether pre-trained language models (LMs) correctly identify and interpret underspecified sentences. We find that newer LMs are reasonably able to identify underspecified sentences when explicitly prompted. However, interpreting them correctly is much harder for any LMs. Our experiments show that when interpreting underspecified sentences, LMs exhibit little uncertainty, contrary to what theoretical accounts of underspecification would predict. Overall, our study reveals limitations in current models’ processing of sentence semantics and highlights the importance of using naturalistic data and communicative scenarios when evaluating LMs’ language capabilities.

Visual storytelling consists in generating a natural language story given a temporally ordered sequence of images. This task is not only challenging for models, but also very difficult to evaluate with automatic metrics since there is no consensus about what makes a story ‘good’. In this paper, we introduce a novel method that measures story quality in terms of human likeness regarding three key aspects highlighted in previous work: visual grounding, coherence, and repetitiveness. We then use this method to evaluate the stories generated by several models, showing that the foundation model LLaVA obtains the best result, but only slightly so compared to TAPM, a 50-times smaller visual storytelling model. Upgrading the visual and language components of TAPM results in a model that yields competitive performance with a relatively low number of parameters. Finally, we carry out a human evaluation study, whose results suggest that a ‘good’ story may require more than a human-like level of visual grounding, coherence, and repetition.

There is an intricate relation between the properties of an image and how humans behave while describing the image. This behavior shows ample variation, as manifested in human signals such as eye movements and when humans start to describe the image. Despite the value of such signals of visuo-linguistic variation, they are virtually disregarded in the training of current pretrained models, which motivates further investigation. Using a corpus of Dutch image descriptions with concurrently collected eye-tracking data, we explore the nature of the variation in visuo-linguistic signals, and find that they correlate with each other. Given this result, we hypothesize that variation stems partly from the properties of the images, and explore whether image representations encoded by pretrained vision encoders can capture such variation. Our results indicate that pretrained models do so to a weak-to-moderate degree, suggesting that the models lack biases about what makes a stimulus complex for humans and what leads to variations in human outputs.

Most languages could be ambiguous, which means the same conveyed text or speech, results in different actions by different readers or listeners. In this project, we propose a method to detect the ambiguity of a sentence using translation by multilingual LLMs. In particular, we hypothesize that a good machine translator should preserve the ambiguity of sentences in all target languages. Therefore, we investigate whether ambiguity is encoded in the hidden representation of a translation model or, instead, if only a single meaning is encoded. In our experiments, we have been able to predict ambiguity of sentences with high accuracy using machine translation without direct use of semantics and only based on the reconstruction error of a function that maps the forward and backward translation hidden representations to each other. The potential applications of the proposed approach span i) detecting ambiguous sentences, ii) fine-tuning existing multilingual LLMs to preserve ambiguous information, and iii) developing AI systems that can generate ambiguity-free languages when needed.

While human speakers use a variety of different expressions when describing the same object in an image, giving rise to a distribution of plausible labels driven by pragmatic constraints, the extent to which current Vision & Language Large Language Models (VLLMs) can mimic this crucial feature of language use is an open question. This applies to common, everyday objects, but it is particularly interesting for uncommon or novel objects for which a category label may be lacking or fuzzy. Furthermore, similar patterns of variation are observed among human speakers for highly context-sensitive expressions, such as the quantifiers ‘few’ or ‘most’. In our work, we evaluate VLLMs (FROMAGe, BLIP-2, LLaVA) on three categories (nouns, attributes, and quantifiers) where humans show great subjective variability concerning the distribution over plausible labels, using datasets and resources mostly under-explored in previous work. Our results reveal mixed evidence on the ability of VLLMs to capture human naming preferences at generation time: while some models are good at mimicking human distributions for nouns and attributes, all of them fail to assign quantifiers, a task that requires more accurate, high-level reasoning.

Intelligent systems that aim at mastering language as humans do must deal with its semantic underspecification, namely, the possibility for a linguistic signal to convey only part of the information needed for communication to succeed. Consider the usages of the pronoun they, which can leave the gender and number of its referent(s) underspecified. Semantic underspecification is not a bug but a crucial language feature that boosts its storage and processing efficiency. Indeed, human speakers can quickly and effortlessly integrate semantically-underspecified linguistic signals with a wide range of non-linguistic information, e.g., the multimodal context, social or cultural conventions, and shared knowledge. Standard NLP models have, in principle, no or limited access to such extra information, while multimodal systems grounding language into other modalities, such as vision, are naturally equipped to account for this phenomenon. However, we show that they struggle with it, which could negatively affect their performance and lead to harmful consequences when used for applications. In this position paper, we argue that our community should be aware of semantic underspecification if it aims to develop language technology that can successfully interact with human users. We discuss some applications where mastering it is crucial and outline a few directions toward achieving this goal.

Dialogue participants may have varying levels of knowledge about the topic under discussion. In such cases, it is essential for speakers to adapt their utterances by taking their audience into account. Yet, it is an open question how such adaptation can be modelled in computational agents. In this paper, we model a visually grounded referential game between a knowledgeable speaker and a listener with more limited visual and linguistic experience. Inspired by psycholinguistic theories, we endow our speaker with the ability to adapt its referring expressions via a simulation module that monitors the effectiveness of planned utterances from the listener’s perspective. We propose an adaptation mechanism building on plug-and-play approaches to controlled language generation, where utterance generation is steered on the fly by the simulator without finetuning the speaker’s underlying language model. Our results and analyses show that our approach is effective: the speaker’s utterances become closer to the listener’s domain of expertise, which leads to higher communicative success.

A proper evaluation of stories generated for a sequence of images—the task commonly referred to as visual storytelling—must consider multiple aspects, such as coherence, grammatical correctness, and visual grounding. In this work, we focus on evaluating the degree of grounding, that is, the extent to which a story is about the entities shown in the images. We analyze current metrics, both designed for this purpose and for general vision-text alignment. Given their observed shortcomings, we propose a novel evaluation tool, GROOViST, that accounts for cross-modal dependencies, temporal misalignments (the fact that the order in which entities appear in the story and the image sequence may not match), and human intuitions on visual grounding. An additional advantage of GROOViST is its modular design, where the contribution of each component can be assessed and interpreted individually.

Despite the impressive performance achieved by pre-trained language-and-vision models in downstream tasks, it remains an open question whether this reflects a proper understanding of image-text interaction. In this work, we explore to what extent they handle basic linguistic constructions—active-passive voice, coordination, and relative clauses—that even preschool children can typically master. We present BLA, a novel, automatically constructed benchmark to evaluate multimodal models on these Basic Language Abilities. We show that different types of Transformer-based systems, such as CLIP, ViLBERT, and BLIP2, generally struggle with BLA in a zero-shot setting, in line with previous findings. Our experiments, in particular, show that most of the tested models only marginally benefit when fine-tuned or prompted with construction-specific samples. Yet, the generative BLIP2 shows promising trends, especially in an in-context learning setting. This opens the door to using BLA not only as an evaluation benchmark but also to improve models’ basic language abilities.

Animacy—whether an entity is alive and sentient—is fundamental to cognitive processing, impacting areas such as memory, vision, and language. However, animacy is not always expressed directly in language: in English it often manifests indirectly, in the form of selectional constraints on verbs and adjectives. This poses a potential issue for transformer language models (LMs): they often train only on text, and thus lack access to extralinguistic information from which humans learn about animacy. We ask: how does this impact LMs’ animacy processing—do they still behave as humans do? We answer this question using open-source LMs. Like previous studies, we find that LMs behave much like humans when presented with entities whose animacy is typical. However, we also show that even when presented with stories about atypically animate entities, such as a peanut in love, LMs adapt: they treat these entities as animate, though they do not adapt as well as humans. Even when the context indicating atypical animacy is very short, LMs pick up on subtle clues and change their behavior. We conclude that despite the limited signal through which LMs can learn about animacy, they are indeed sensitive to the relevant lexical semantic nuances available in English.

This work studies the semantic representations learned by BERT for compounds, that is, expressions such as sunlight or bodyguard. We build on recent studies that explore semantic information in Transformers at the word level and test whether BERT aligns with human semantic intuitions when dealing with expressions (e.g., sunlight) whose overall meaning depends—to a various extent—on the semantics of the constituent words (sun, light). We leverage a dataset that includes human judgments on two psycholinguistic measures of compound semantic analysis: lexeme meaning dominance (LMD; quantifying the weight of each constituent toward the compound meaning) and semantic transparency (ST; evaluating the extent to which the compound meaning is recoverable from the constituents’ semantics). We show that BERT-based measures moderately align with human intuitions, especially when using contextualized representations, and that LMD is overall more predictable than ST. Contrary to the results reported for ‘standard’ words, higher, more contextualized layers are the best at representing compound meaning. These findings shed new light on the abilities of BERT in dealing with fine-grained semantic phenomena. Moreover, they can provide insights into how speakers represent compounds.

In this work, we use a transformer-based pre-trained multimodal model, CLIP, to shed light on the mechanisms employed by human speakers when referring to visual entities. In particular, we use CLIP to quantify the degree of descriptiveness (how well an utterance describes an image in isolation) and discriminativeness (to what extent an utterance is effective in picking out a single image among similar images) of human referring utterances within multimodal dialogues. Overall, our results show that utterances become less descriptive over time while their discriminativeness remains unchanged. Through analysis, we propose that this trend could be due to participants relying on the previous mentions in the dialogue history, as well as being able to distill the most discriminative information from the visual context. In general, our study opens up the possibility of using this and similar models to quantify patterns in human data and shed light on the underlying cognitive mechanisms.

To be trusted and perceived as natural and coherent, conversational systems must adapt to the language of their users. While personalized dialogue is a promising direction, controlling generation for fine-grained language features remains a challenge in this approach. A recent line of research showed the effectiveness of leveraging pre-trained language models toward adapting to a text’s topic or sentiment. In this study, we build on these approaches and focus on a higher-level dimension of language variation: speakers’ age. We frame the task as a dialogue response generation, and test methods based on bag-of-words (BoW) and neural discriminators (Disc) to condition the output of GPT-2 and DialoGPT without altering the parameters of the language models. We show that Disc models achieve a higher degree of detectable control than BoW models based on automatic evaluation. In contrast, humans can partially detect age differences in BoW but not Disc responses. Since BoW responses are deemed better than Disc ones by humans, simple controllable methods thus appear to be a better tradeoff between adaptation and language quality. Our work confirms the challenges of adapting to higher-level dimensions of language variation. Moreover, it highlights the need to evaluate natural language generation thoroughly.

This study carries out a systematic intrinsic evaluation of the semantic representations learned by state-of-the-art pre-trained multimodal Transformers. These representations are claimed to be task-agnostic and shown to help on many downstream language-and-vision tasks. However, the extent to which they align with human semantic intuitions remains unclear. We experiment with various models and obtain static word representations from the contextualized ones they learn. We then evaluate them against the semantic judgments provided by human speakers. In line with previous evidence, we observe a generalized advantage of multimodal representations over language- only ones on concrete word pairs, but not on abstract ones. On the one hand, this confirms the effectiveness of these models to align language and vision, which results in better semantic representations for concepts that are grounded in images. On the other hand, models are shown to follow different representation learning patterns, which sheds some light on how and when they perform multimodal integration.

We propose EASE, a simple diagnostic tool for Visual Question Answering (VQA) which quantifies the difficulty of an image, question sample. EASE is based on the pattern of answers provided by multiple annotators to a given question. In particular, it considers two aspects of the answers: (i) their Entropy; (ii) their Semantic content. First, we prove the validity of our diagnostic to identify samples that are easy/hard for state-of-art VQA models. Second, we show that EASE can be successfully used to select the most-informative samples for training/fine-tuning. Crucially, only information that is readily available in any VQA dataset is used to compute its scores.

We investigate the representations learned by vision and language models in tasks that require relational reasoning. Focusing on the problem of assessing the relative size of objects in abstract visual contexts, we analyse both one-step and two-step reasoning. For the latter, we construct a new dataset of three-image scenes and define a task that requires reasoning at the level of the individual images and across images in a scene. We probe the learned model representations using diagnostic classifiers. Our experiments show that pretrained multimodal transformer-based architectures can perform higher-level relational reasoning, and are able to learn representations for novel tasks and data that are very different from what was seen in pretraining.

This paper introduces BD2BB, a novel language and vision benchmark that requires multimodal models combine complementary information from the two modalities. Recently, impressive progress has been made to develop universal multimodal encoders suitable for virtually any language and vision tasks. However, current approaches often require them to combine redundant information provided by language and vision. Inspired by real-life communicative contexts, we propose a novel task where either modality is necessary but not sufficient to make a correct prediction. To do so, we first build a dataset of images and corresponding sentences provided by human participants. Second, we evaluate state-of-the-art models and compare their performance against human speakers. We show that, while the task is relatively easy for humans, best-performing models struggle to achieve similar results.

Dialogue participants often refer to entities or situations repeatedly within a conversation, which contributes to its cohesiveness. Subsequent references exploit the common ground accumulated by the interlocutors and hence have several interesting properties, namely, they tend to be shorter and reuse expressions that were effective in previous mentions. In this paper, we tackle the generation of first and subsequent references in visually grounded dialogue. We propose a generation model that produces referring utterances grounded in both the visual and the conversational context. To assess the referring effectiveness of its output, we also implement a reference resolution system. Our experiments and analyses show that the model produces better, more effective referring utterances than a model not grounded in the dialogue context, and generates subsequent references that exhibit linguistic patterns akin to humans.

When speakers describe an image, they tend to look at objects before mentioning them. In this paper, we investigate such sequential cross-modal alignment by modelling the image description generation process computationally. We take as our starting point a state-of-the-art image captioning system and develop several model variants that exploit information from human gaze patterns recorded during language production. In particular, we propose the first approach to image description generation where visual processing is modelled sequentially. Our experiments and analyses confirm that better descriptions can be obtained by exploiting gaze-driven attention and shed light on human cognitive processes by comparing different ways of aligning the gaze modality with language production. We find that processing gaze data sequentially leads to descriptions that are better aligned to those produced by speakers, more diverse, and more natural—particularly when gaze is encoded with a dedicated recurrent component.

This work aims at modeling how the meaning of gradable adjectives of size (‘big’, ‘small’) can be learned from visually-grounded contexts. Inspired by cognitive and linguistic evidence showing that the use of these expressions relies on setting a threshold that is dependent on a specific context, we investigate the ability of multi-modal models in assessing whether an object is ‘big’ or ‘small’ in a given visual scene. In contrast with the standard computational approach that simplistically treats gradable adjectives as ‘fixed’ attributes, we pose the problem as relational: to be successful, a model has to consider the full visual context. By means of four main tasks, we show that state-of-the-art models (but not a relatively strong baseline) can learn the function subtending the meaning of size adjectives, though their performance is found to decrease while moving from simple to more complex tasks. Crucially, models fail in developing abstract representations of gradable adjectives that can be used compositionally.

In this paper, we experiment with a recently proposed visual reasoning task dealing with quantities – modeling the multimodal, contextually-dependent meaning of size adjectives (‘big’, ‘small’) – and explore the impact of varying the training data on the learning behavior of a state-of-art system. In previous work, models have been shown to fail in generalizing to unseen adjective-noun combinations. Here, we investigate whether, and to what extent, seeing some of these cases during training helps a model understand the rule subtending the task, i.e., that being big implies being not small, and vice versa. We show that relatively few examples are enough to understand this relationship, and that developing a specific, mutually exclusive representation of size adjectives is beneficial to the task.

Inspired by the literature on multisensory integration, we develop a computational model to ground quantifiers in perception. The model learns to pick, out of nine quantifiers (‘few’, ‘many’, ‘all’, etc.), the one that is more likely to describe the percent of animals in a visual-auditory input containing both animals and artifacts. We show that relying on concurrent sensory inputs increases model performance on the quantification task. Moreover, we evaluate the model in a situation in which only the auditory modality is given, while the visual one is ‘hallucinanted’ either from the auditory input itself or from a linguistic caption describing the quantity of entities in the auditory input. This way, the model exploits prior associations between modalities. We show that the model profits from the prior knowledge and outperforms the auditory-only setting.

The present work investigates whether different quantification mechanisms (set comparison, vague quantification, and proportional estimation) can be jointly learned from visual scenes by a multi-task computational model. The motivation is that, in humans, these processes underlie the same cognitive, non-symbolic ability, which allows an automatic estimation and comparison of set magnitudes. We show that when information about lower-complexity tasks is available, the higher-level proportional task becomes more accurate than when performed in isolation. Moreover, the multi-task model is able to generalize to unseen combinations of target/non-target objects. Consistently with behavioral evidence showing the interference of absolute number in the proportional task, the multi-task model no longer works when asked to provide the number of target objects in the scene.

We study the role of linguistic context in predicting quantifiers (‘few’, ‘all’). We collect crowdsourced data from human participants and test various models in a local (single-sentence) and a global context (multi-sentence) condition. Models significantly out-perform humans in the former setting and are only slightly better in the latter. While human performance improves with more linguistic context (especially on proportional quantifiers), model performance suffers. Models are very effective in exploiting lexical and morpho-syntactic patterns; humans are better at genuinely understanding the meaning of the (global) context.

People can refer to quantities in a visual scene by using either exact cardinals (e.g. one, two, three) or natural language quantifiers (e.g. few, most, all). In humans, these two processes underlie fairly different cognitive and neural mechanisms. Inspired by this evidence, the present study proposes two models for learning the objective meaning of cardinals and quantifiers from visual scenes containing multiple objects. We show that a model capitalizing on a ‘fuzzy’ measure of similarity is effective for learning quantifiers, whereas the learning of exact cardinals is better accomplished when information about number is provided.

In this paper, we aim to understand whether current language and vision (LaVi) models truly grasp the interaction between the two modalities. To this end, we propose an extension of the MS-COCO dataset, FOIL-COCO, which associates images with both correct and ‘foil’ captions, that is, descriptions of the image that are highly similar to the original ones, but contain one single mistake (‘foil word’). We show that current LaVi models fall into the traps of this data and perform badly on three tasks: a) caption classification (correct vs. foil); b) foil word detection; c) foil word correction. Humans, in contrast, have near-perfect performance on those tasks. We demonstrate that merely utilising language cues is not enough to model FOIL-COCO and that it challenges the state-of-the-art by requiring a fine-grained understanding of the relation between text and image.