In dialogue, the addressee may initially misunderstand the speaker and respond erroneously, often prompting the speaker to correct the misunderstanding in the next turn with a Third Position Repair (TPR). The ability to process and respond appropriately to such repair sequences is thus crucial in conversational AI systems. In this paper, we first collect, analyse, and publicly release BlockWorld-Repairs: a dataset of multi-modal TPR sequences in an instruction-following manipulation task that is, by design, rife with referential ambiguity. We employ this dataset to evaluate several state-of-the-art Vision and Language Models (VLM) across multiple settings, focusing on their capability to process and accurately respond to TPRs and thus recover from miscommunication. We find that, compared to humans, all models significantly underperform in this task. We then show that VLMs can benefit from specialised losses targeting relevant tokens during fine-tuning, achieving better performance and generalising better to new scenarios. Our results suggest that these models are not yet ready to be deployed in multi-modal collaborative settings where repairs are common, and highlight the need to design training regimes and objectives that facilitate learning from interaction. Our code and data are available at www.github.com/JChiyah/blockworld-repairs

This study explores replacing Transformers in Visual Language Models (VLMs) with Mamba, a recent structured state space model (SSM) that demonstrates promising performance in sequence modeling. We test models up to 3B parameters under controlled conditions, showing that Mamba-based VLMs outperforms Transformers-based VLMs in captioning, question answering, and reading comprehension. However, we find that Transformers achieve greater performance in visual grounding and the performance gap widens with scale. We explore two hypotheses to explain this phenomenon: 1) the effect of task-agnostic visual encoding on the updates of the hidden states, and 2) the difficulty in performing visual grounding from the perspective of in-context multimodal retrieval. Our results indicate that a task-aware encoding yields minimal performance gains on grounding, however, Transformers significantly outperform Mamba at in-context multimodal retrieval. Overall, Mamba shows promising performance on tasks where the correct output relies on a summary of the image but struggles when retrieval of explicit information from the context is required.

Evaluating the generalisation capabilities of multimodal models based solely on their performance on out-of-distribution data fails to capture their true robustness. This work introduces a comprehensive evaluation framework that systematically examines the role of instructions and inputs in the generalisation abilities of such models, considering architectural design, input perturbations across language and vision modalities, and increased task complexity. The proposed framework uncovers the resilience of multimodal models to extreme instruction perturbations and their vulnerability to observational changes, raising concerns about overfitting to spurious correlations. By employing this evaluation framework on current Transformer-based multimodal models for robotic manipulation tasks, we uncover limitations and suggest future advancements should focus on architectural and training innovations that better integrate multimodal inputs, enhancing a model’s generalisation prowess by prioritising sensitivity to input content over incidental correlations.

Continual learning focuses on incrementally training a model on a sequence of tasks with the aim of learning new tasks while minimizing performance drop on previous tasks. Existing approaches at the intersection of Continual Learning and Visual Question Answering (VQA) do not study how the multimodal nature of the input affects the learning dynamics of a model. In this paper, we demonstrate that each modality evolves at different rates across a continuum of tasks and that this behavior occurs in established encoder-only models as well as modern recipes for developing Vision & Language (VL) models. Motivated by this observation, we propose a modality-aware feature distillation (MAFED) approach which outperforms existing baselines across models of varying scale in three multimodal continual learning settings. Furthermore, we provide ablations showcasing that modality-aware distillation complements experience replay. Overall, our results emphasize the importance of addressing modality-specific dynamics to prevent forgetting in multimodal continual learning.

Recent work showed the possibility of building open-vocabulary large language models (LLMs) that directly operate on pixel representations. These models are implemented as autoencoders that reconstruct masked patches of rendered text.However, these pixel-based LLMs are limited to discriminative tasks (e.g., classification) and, similar to BERT, cannot be used to generate text.Therefore, they cannot be used for generative tasks such as free-form question answering. In this work, we introduce PIXAR, the first pixel-based autoregressive LLM that performs text generation. Consisting of only a decoder, PIXAR can perform free-form generative tasks while keeping the number of parameters on par with previous encoder-decoder models.Furthermore, we highlight the challenges of generating text as non-noisy images and show this is due to using a maximum likelihood objective. To overcome this problem, we propose an adversarial pretraining stage that improves the readability and accuracy of PIXAR by 8.1 on LAMBADA and 8.5 on bAbI— making it comparable to GPT-2 on text generation tasks.This paves the way to build open-vocabulary LLMs that operate on perceptual input only and calls into question the necessity of the usual symbolic input representation, i.e., text as (sub)tokens.

AI personal assistants deployed via robots or wearables require embodied understanding to collaborate with humans effectively. However, current Vision-Language Models (VLMs) primarily focus on third-person view videos, neglecting the richness of egocentric perceptual experience. To address this gap, we propose three key contributions. First, we introduce the Egocentric Video Understanding Dataset (EVUD) for training VLMs on video captioning and question answering tasks specific to egocentric videos. Second, we present , a 7B parameter VLM trained using parameter-efficient methods on EVUD. Finally, we evaluate ‘s capabilities on OpenEQA, a challenging benchmark for embodied video question answering. Our model achieves state-of-the-art performance, outperforming open-source models including strong Socratic models using GPT-4 as a planner by 3.6%.Additionally, we outperform Claude 3 and Gemini Pro Vision 1.0 and showcase competitive results compared to Gemini Pro 1.5 and GPT-4V, even surpassing the latter in spatial reasoning. This research paves the way for building efficient VLMs that can be deployed in robots or wearables, leveraging embodied video understanding to collaborate seamlessly with humans in everyday tasks, contributing to the advancement of next-generation Embodied AI.

An effective method for combining frozen large language models (LLM) and visual encoders involves a resampler module that creates a ‘visual prompt’ which is provided to the LLM, along with the textual prompt. While this approach has enabled impressive performance across many coarse-grained tasks like image captioning and visual question answering, more fine-grained tasks that require spatial understanding have not been thoroughly examined. In this paper, we use diagnostic classifiers to measure the extent to which the visual prompt produced by the resampler encodes spatial information. Our results show that this information is largely absent from the resampler output when kept frozen during training of the classifiers. However, when the resampler and classifier are trained jointly, we observe a significant performance boost. This shows that the compression achieved by the resamplers can in principle encode the requisite spatial information, but that more object-aware objectives are needed at the pretraining stage to facilitate this capability.

Augmenting Large Language Models (LLMs) with image-understanding capabilities has resulted in a boom of high-performing Vision-Language models (VLMs). While studying the alignment of LLMs to human values has received widespread attention, the safety of VLMs has not received the same attention. In this paper, we explore the impact of jailbreaking on three state-of-the-art VLMs, each using a distinct modeling approach. By comparing each VLM to their respective LLM backbone, we find that each VLM is more susceptible to jailbreaking. We consider this as an undesirable outcome from visual instruction-tuning, which imposes a forgetting effect on an LLM’s safety guardrails. Therefore, we provide recommendations for future work based on evaluation strategies that aim to highlight the weaknesses of a VLM, as well as take safety measures into account during visual instruction tuning.

Referential ambiguities arise in dialogue when a referring expression does not uniquely identify the intended referent for the addressee. Addressees usually detect such ambiguities immediately and work with the speaker to repair it using meta-communicative, Clarificational Exchanges (CE): a Clarification Request (CR) and a response. Here, we argue that the ability to generate and respond to CRs imposes specific constraints on the architecture and objective functions of multi-modal, visually grounded dialogue models. We use the SIMMC 2.0 dataset to evaluate the ability of different state-of-the-art model architectures to process CEs, with a metric that probes the contextual updates that arise from them in the model. We find that language-based models are able to encode simple multi-modal semantic information and process some CEs, excelling with those related to the dialogue history, whilst multi-modal models can use additional learning objectives to obtain disentangled object representations, which become crucial to handle complex referential ambiguities across modalities overall.

Interactive and embodied tasks pose at least two fundamental challenges to existing Vision & Language (VL) models, including 1) grounding language in trajectories of actions and observations, and 2) referential disambiguation. To tackle these challenges, we propose an Embodied MultiModal Agent (EMMA): a unified encoder-decoder model that reasons over images and trajectories, and casts action prediction as multimodal text generation. By unifying all tasks as text generation, EMMA learns a language of actions which facilitates transfer across tasks. Different to previous modular approaches with independently trained components, we use a single multitask model where each task contributes to goal completion. EMMA performs on par with similar models on several VL benchmarks and sets a new state-of-the-art performance (36.81% success rate) on the Dialog-guided Task Completion (DTC), a benchmark to evaluate dialog-guided agents in the Alexa Arena.

Compositionality – the ability to combine simpler concepts to understand & generate arbitrarily more complex conceptual structures – has long been thought to be the cornerstone of human language capacity. With the recent, notable success of neural models in various NLP tasks, attention has now naturally turned to the compositional capacity of these models. In this paper, we study the compositional generalization properties of image captioning models. We perform a set experiments under controlled conditions using model and data ablations, each designed to benchmark a particular facet of compositional generalization: systematicity is the ability of a model to create novel combinations of concepts out of those observed during training, productivity is here operationalised as the capacity of a model to extend its predictions beyond the length distribution it has observed during training, and substitutivity is concerned with the robustness of the model against synonym substitutions. While previous work has focused primarily on systematicity, here we provide a more in-depth analysis of the strengths and weaknesses of state of the art captioning models. Our findings demonstrate that the models we study here do not compositionally generalize in terms of systematicity and productivity, however, they are robust to some degree to synonym substitutions

We demonstrate EMMA, an embodied multimodal agent which has been developed for the Alexa Prize SimBot challenge. The agent acts within a 3D simulated environment for household tasks. EMMA is a unified and multimodal generative model aimed at solving embodied tasks. In contrast to previous work, our approach treats multiple multimodal tasks as a single multimodal conditional text generation problem, where a model learns to output text given both language and visual input. Furthermore, we showcase that a single generative agent can solve tasks with visual inputs of varying length, such as answering questions about static images, or executing actions given a sequence of previous frames and dialogue utterances. The demo system will allow users to interact conversationally with EMMA in embodied dialogues in different 3D environments from the TEACh dataset.

Since the advent of Transformer-based, pretrained language models (LM) such as BERT, Natural Language Understanding (NLU) components in the form of Dialogue Act Recognition (DAR) and Slot Recognition (SR) for dialogue systems have become both more accurate and easier to create for specific application domains. Unsurprisingly however, much of this progress has been limited to the English language, due to the existence of very large datasets in both dialogue and written form, while only few corpora are available for lower resourced languages like Italian. In this paper, we present JILDA 2.0, an enhanced version of a Italian task-oriented dialogue dataset, using it to realise a Italian NLU baseline by evaluating three of the most recent pretrained LMs: Italian BERT, Multilingual BERT, and AlBERTo for the DAR and SR tasks. Thus, this paper not only presents an updated version of a dataset characterised by complex dialogues, but it also highlights the challenges that still remain in creating effective NLU components for lower resourced languages, constituting a first step in improving NLU for Italian dialogue.

Artificial agents are nowadays challenged to perform embodied AI tasks. To succeed, agents must understand the meaning of verbs and how their corresponding actions transform the surrounding world. In this work, we propose ACT-Thor, a novel controlled benchmark for embodied action understanding. We use the AI2-THOR simulated environment to produce a controlled setup in which an agent, given a before-image and an associated action command, has to determine what the correct after-image is among a set of possible candidates. First, we assess the feasibility of the task via a human evaluation that resulted in 81.4% accuracy, and very high inter-annotator agreement (84.9%). Second, we design both unimodal and multimodal baselines, using state-of-the-art visual feature extractors. Our evaluation and error analysis suggest that only models that have a very structured representation of the actions together with powerful visual features can perform well on the task. However, they still fall behind human performance in a zero-shot scenario where the model is exposed to unseen (action, object) pairs. This paves the way for a systematic way of evaluating embodied AI agents that understand grounded actions.

Guessing games are a prototypical instance of the “learning by interacting” paradigm. This work investigates how well an artificial agent can benefit from playing guessing games when later asked to perform on novel NLP downstream tasks such as Visual Question Answering (VQA). We propose two ways to exploit playing guessing games: 1) a supervised learning scenario in which the agent learns to mimic successful guessing games and 2) a novel way for an agent to play by itself, called Self-play via Iterated Experience Learning (SPIEL). We evaluate the ability of both procedures to generalise: an in-domain evaluation shows an increased accuracy (+7.79) compared with competitors on the evaluation suite CompGuessWhat?!; a transfer evaluation shows improved performance for VQA on the TDIUC dataset in terms of harmonic average accuracy (+5.31) thanks to more fine-grained object representations learned via SPIEL.

In visual guessing games, a Guesser has to identify a target object in a scene by asking questions to an Oracle. An effective strategy for the players is to learn conceptual representations of objects that are both discriminative and expressive enough to ask questions and guess correctly. However, as shown by Suglia et al. (2020), existing models fail to learn truly multi-modal representations, relying instead on gold category labels for objects in the scene both at training and inference time. This provides an unnatural performance advantage when categories at inference time match those at training time, and it causes models to fail in more realistic “zero-shot” scenarios where out-of-domain object categories are involved. To overcome this issue, we introduce a novel “imagination” module based on Regularized Auto-Encoders, that learns context-aware and category-aware latent embeddings without relying on category labels at inference time. Our imagination module outperforms state-of-the-art competitors by 8.26% gameplay accuracy in the CompGuessWhat?! zero-shot scenario (Suglia et al., 2020), and it improves the Oracle and Guesser accuracy by 2.08% and 12.86% in the GuessWhat?! benchmark, when no gold categories are available at inference time. The imagination module also boosts reasoning about object properties and attributes.

Approaches to Grounded Language Learning are commonly focused on a single task-based final performance measure which may not depend on desirable properties of the learned hidden representations, such as their ability to predict object attributes or generalize to unseen situations. To remedy this, we present GroLLA, an evaluation framework for Grounded Language Learning with Attributes based on three sub-tasks: 1) Goal-oriented evaluation; 2) Object attribute prediction evaluation; and 3) Zero-shot evaluation. We also propose a new dataset CompGuessWhat?! as an instance of this framework for evaluating the quality of learned neural representations, in particular with respect to attribute grounding. To this end, we extend the original GuessWhat?! dataset by including a semantic layer on top of the perceptual one. Specifically, we enrich the VisualGenome scene graphs associated with the GuessWhat?! images with several attributes from resources such as VISA and ImSitu. We then compare several hidden state representations from current state-of-the-art approaches to Grounded Language Learning. By using diagnostic classifiers, we show that current models’ learned representations are not expressive enough to encode object attributes (average F1 of 44.27). In addition, they do not learn strategies nor representations that are robust enough to perform well when novel scenes or objects are involved in gameplay (zero-shot best accuracy 50.06%).