Shruti Palaskar

Shruti Palaskar, Akshita Bhagia, Yonatan Bisk et al. · allen-ai, cmu

Combining the visual modality with pretrained language models has been surprisingly effective for simple descriptive tasks such as image captioning. More general text generation however remains elusive. We take a step back and ask: How do these models work for more complex generative tasks, i.e. conditioning on both text and images? Are multimodal models simply visually adapted language models, or do they combine they reason jointly over modalities? We investigate these questions in the context of self-rationalization (jointly generating task labels/answers and free-text explanations) of three tasks: (i) visual question answering in VQA-X, (ii) visual commonsense reasoning in VCR, and (iii) visual-textual entailment in e-SNLI-VE. We show that recent unimodal advances, CLIP image representations and scaling of language models, do not consistently improve self-rationalization in multimodal tasks. We find that no single model type works universally best across tasks, datasets, and finetuning data sizes. Our findings motivate the need for novel general backbones approach that move text generation from images and text beyond image captioning.

Anirudh Mani, Shruti Palaskar, Nimshi Venkat Meripo et al.

Off-the-shelf pre-trained Automatic Speech Recognition (ASR) systems are an increasingly viable service for companies of any size building speech-based products. While these ASR systems are trained on large amounts of data, domain mismatch is still an issue for many such parties that want to use this service as-is leading to not so optimal results for their task. We propose a simple technique to perform domain adaptation for ASR error correction via machine translation. The machine translation model is a strong candidate to learn a mapping from out-of-domain ASR errors to in-domain terms in the corresponding reference files. We use two off-the-shelf ASR systems in this work: Google ASR (commercial) and the ASPIRE model (open-source). We observe 7% absolute improvement in word error rate and 4 point absolute improvement in BLEU score in Google ASR output via our proposed method. We also evaluate ASR error correction via a downstream task of Speaker Diarization that captures speaker style, syntax, structure and semantic improvements we obtain via ASR correction.

Alec Helbling, Shruti Palaskar, Kundan Krishna et al.

What exactly makes a particular image unsafe? Systematically differentiating between benign and problematic images is a challenging problem, as subtle changes to an image, such as an insulting gesture or symbol, can drastically alter its safety implications. However, existing image safety datasets are coarse and ambiguous, offering only broad safety labels without isolating the specific features that drive these differences. We introduce SafetyPairs, a scalable framework for generating counterfactual pairs of images, that differ only in the features relevant to the given safety policy, thus flipping their safety label. By leveraging image editing models, we make targeted changes to images that alter their safety labels while leaving safety-irrelevant details unchanged. Using SafetyPairs, we construct a new safety benchmark, which serves as a powerful source of evaluation data that highlights weaknesses in vision-language models' abilities to distinguish between subtly different images. Beyond evaluation, we find our pipeline serves as an effective data augmentation strategy that improves the sample efficiency of training lightweight guard models. We release a benchmark containing over 3,020 SafetyPair images spanning a diverse taxonomy of 9 safety categories, providing the first systematic resource for studying fine-grained image safety distinctions.

Shruti Palaskar, Leon Gatys, Mona Abdelrahman et al.

Safety evaluation of multimodal foundation models often treats vision and language inputs separately, missing risks from joint interpretation where benign content becomes harmful in combination. Existing approaches also fail to distinguish clearly unsafe content from borderline cases, leading to problematic over-blocking or under-refusal of genuinely harmful content. We present Vision Language Safety Understanding (VLSU), a comprehensive framework to systematically evaluate multimodal safety through fine-grained severity classification and combinatorial analysis across 17 distinct safety patterns. Using a multi-stage pipeline with real-world images and human annotation, we construct a large-scale benchmark of 8,187 samples spanning 15 harm categories. Our evaluation of eleven state-of-the-art models reveals systematic joint understanding failures: while models achieve 90%-plus accuracy on clear unimodal safety signals, performance degrades substantially to 20-55% when joint image-text reasoning is required to determine the safety label. Most critically, 34% of errors in joint image-text safety classification occur despite correct classification of the individual modalities, further demonstrating absent compositional reasoning capabilities. Additionally, we find that models struggle to balance refusing unsafe content while still responding to borderline cases that deserve engagement. For example, we find that instruction framing can reduce the over-blocking rate on borderline content from 62.4% to 10.4% in Gemini-1.5, but only at the cost of under-refusing on unsafe content with refusal rate dropping from 90.8% to 53.9%. Overall, our framework exposes weaknesses in joint image-text understanding and alignment gaps in current models, and provides a critical test bed to enable the next milestones in research on robust vision-language safety.

Shruti Palaskar, Oggi Rudovic, Sameer Dharur et al.

Although Large Language Models (LLMs) have shown promise for human-like conversations, they are primarily pre-trained on text data. Incorporating audio or video improves performance, but collecting large-scale multimodal data and pre-training multimodal LLMs is challenging. To this end, we propose a Fusion Low Rank Adaptation (FLoRA) technique that efficiently adapts a pre-trained unimodal LLM to consume new, previously unseen modalities via low rank adaptation. For device-directed speech detection, using FLoRA, the multimodal LLM achieves 22% relative reduction in equal error rate (EER) over the text-only approach and attains performance parity with its full fine-tuning (FFT) counterpart while needing to tune only a fraction of its parameters. Furthermore, with the newly introduced adapter dropout, FLoRA is robust to missing data, improving over FFT by 20% lower EER and 56% lower false accept rate. The proposed approach scales well for model sizes from 16M to 3B parameters.

Roshan Sharma, Shruti Palaskar, Alan W Black et al.

Speech summarization is typically performed by using a cascade of speech recognition and text summarization models. End-to-end modeling of speech summarization models is challenging due to memory and compute constraints arising from long input audio sequences. Recent work in document summarization has inspired methods to reduce the complexity of self-attentions, which enables transformer models to handle long sequences. In this work, we introduce a single model optimized end-to-end for speech summarization. We apply the restricted self-attention technique from text-based models to speech models to address the memory and compute constraints. We demonstrate that the proposed model learns to directly summarize speech for the How-2 corpus of instructional videos. The proposed end-to-end model outperforms the previously proposed cascaded model by 3 points absolute on ROUGE. Further, we consider the spoken language understanding task of predicting concepts from speech inputs and show that the proposed end-to-end model outperforms the cascade model by 4 points absolute F-1.

Amanda Duarte, Shruti Palaskar, Lucas Ventura et al.

One of the factors that have hindered progress in the areas of sign language recognition, translation, and production is the absence of large annotated datasets. Towards this end, we introduce How2Sign, a multimodal and multiview continuous American Sign Language (ASL) dataset, consisting of a parallel corpus of more than 80 hours of sign language videos and a set of corresponding modalities including speech, English transcripts, and depth. A three-hour subset was further recorded in the Panoptic studio enabling detailed 3D pose estimation. To evaluate the potential of How2Sign for real-world impact, we conduct a study with ASL signers and show that synthesized videos using our dataset can indeed be understood. The study further gives insights on challenges that computer vision should address in order to make progress in this field. Dataset website: http://how2sign.github.io/

Shruti Palaskar, Jindrich Libovický, Spandana Gella et al.

In this paper, we study abstractive summarization for open-domain videos. Unlike the traditional text news summarization, the goal is less to "compress" text information but rather to provide a fluent textual summary of information that has been collected and fused from different source modalities, in our case video and audio transcripts (or text). We show how a multi-source sequence-to-sequence model with hierarchical attention can integrate information from different modalities into a coherent output, compare various models trained with different modalities and present pilot experiments on the How2 corpus of instructional videos. We also propose a new evaluation metric (Content F1) for abstractive summarization task that measures semantic adequacy rather than fluency of the summaries, which is covered by metrics like ROUGE and BLEU.

Shruti Palaskar, Vikas Raunak, Florian Metze

End-to-end acoustic-to-word speech recognition models have recently gained popularity because they are easy to train, scale well to large amounts of training data, and do not require a lexicon. In addition, word models may also be easier to integrate with downstream tasks such as spoken language understanding, because inference (search) is much simplified compared to phoneme, character or any other sort of sub-word units. In this paper, we describe methods to construct contextual acoustic word embeddings directly from a supervised sequence-to-sequence acoustic-to-word speech recognition model using the learned attention distribution. On a suite of 16 standard sentence evaluation tasks, our embeddings show competitive performance against a word2vec model trained on the speech transcriptions. In addition, we evaluate these embeddings on a spoken language understanding task, and observe that our embeddings match the performance of text-based embeddings in a pipeline of first performing speech recognition and then constructing word embeddings from transcriptions.

Nils Holzenberger, Shruti Palaskar, Pranava Madhyastha et al.

An increasing number of datasets contain multiple views, such as video, sound and automatic captions. A basic challenge in representation learning is how to leverage multiple views to learn better representations. This is further complicated by the existence of a latent alignment between views, such as between speech and its transcription, and by the multitude of choices for the learning objective. We explore an advanced, correlation-based representation learning method on a 4-way parallel, multimodal dataset, and assess the quality of the learned representations on retrieval-based tasks. We show that the proposed approach produces rich representations that capture most of the information shared across views. Our best models for speech and textual modalities achieve retrieval rates from 70.7% to 96.9% on open-domain, user-generated instructional videos. This shows it is possible to learn reliable representations across disparate, unaligned and noisy modalities, and encourages using the proposed approach on larger datasets.

Ozan Caglayan, Ramon Sanabria, Shruti Palaskar et al.

Humans are capable of processing speech by making use of multiple sensory modalities. For example, the environment where a conversation takes place generally provides semantic and/or acoustic context that helps us to resolve ambiguities or to recall named entities. Motivated by this, there have been many works studying the integration of visual information into the speech recognition pipeline. Specifically, in our previous work, we propose a multistep visual adaptive training approach which improves the accuracy of an audio-based Automatic Speech Recognition (ASR) system. This approach, however, is not end-to-end as it requires fine-tuning the whole model with an adaptation layer. In this paper, we propose novel end-to-end multimodal ASR systems and compare them to the adaptive approach by using a range of visual representations obtained from state-of-the-art convolutional neural networks. We show that adaptive training is effective for S2S models leading to an absolute improvement of 1.4% in word error rate. As for the end-to-end systems, although they perform better than baseline, the improvements are slightly less than adaptive training, 0.8 absolute WER reduction in single-best models. Using ensemble decoding, end-to-end models reach a WER of 15% which is the lowest score among all systems.

Ramon Sanabria, Ozan Caglayan, Shruti Palaskar et al.

In this paper, we introduce How2, a multimodal collection of instructional videos with English subtitles and crowdsourced Portuguese translations. We also present integrated sequence-to-sequence baselines for machine translation, automatic speech recognition, spoken language translation, and multimodal summarization. By making available data and code for several multimodal natural language tasks, we hope to stimulate more research on these and similar challenges, to obtain a deeper understanding of multimodality in language processing.

Shruti Palaskar, Florian Metze

Acoustic-to-Word recognition provides a straightforward solution to end-to-end speech recognition without needing external decoding, language model re-scoring or lexicon. While character-based models offer a natural solution to the out-of-vocabulary problem, word models can be simpler to decode and may also be able to directly recognize semantically meaningful units. We present effective methods to train Sequence-to-Sequence models for direct word-level recognition (and character-level recognition) and show an absolute improvement of 4.4-5.0\% in Word Error Rate on the Switchboard corpus compared to prior work. In addition to these promising results, word-based models are more interpretable than character models, which have to be composed into words using a separate decoding step. We analyze the encoder hidden states and the attention behavior, and show that location-aware attention naturally represents words as a single speech-word-vector, despite spanning multiple frames in the input. We finally show that the Acoustic-to-Word model also learns to segment speech into words with a mean standard deviation of 3 frames as compared with human annotated forced-alignments for the Switchboard corpus.

Shruti Palaskar, Ramon Sanabria, Florian Metze

Transcription or sub-titling of open-domain videos is still a challenging domain for Automatic Speech Recognition (ASR) due to the data's challenging acoustics, variable signal processing and the essentially unrestricted domain of the data. In previous work, we have shown that the visual channel -- specifically object and scene features -- can help to adapt the acoustic model (AM) and language model (LM) of a recognizer, and we are now expanding this work to end-to-end approaches. In the case of a Connectionist Temporal Classification (CTC)-based approach, we retain the separation of AM and LM, while for a sequence-to-sequence (S2S) approach, both information sources are adapted together, in a single model. This paper also analyzes the behavior of CTC and S2S models on noisy video data (How-To corpus), and compares it to results on the clean Wall Street Journal (WSJ) corpus, providing insight into the robustness of both approaches.

Odette Scharenborg, Laurent Besacier, Alan Black et al.

We summarize the accomplishments of a multi-disciplinary workshop exploring the computational and scientific issues surrounding the discovery of linguistic units (subwords and words) in a language without orthography. We study the replacement of orthographic transcriptions by images and/or translated text in a well-resourced language to help unsupervised discovery from raw speech.

Yohan Jo, Lisa Lee, Shruti Palaskar

There is a great need for technologies that can predict the mortality of patients in intensive care units with both high accuracy and accountability. We present joint end-to-end neural network architectures that combine long short-term memory (LSTM) and a latent topic model to simultaneously train a classifier for mortality prediction and learn latent topics indicative of mortality from textual clinical notes. For topic interpretability, the topic modeling layer has been carefully designed as a single-layer network with constraints inspired by LDA. Experiments on the MIMIC-III dataset show that our models significantly outperform prior models that are based on LDA topics in mortality prediction. However, we achieve limited success with our method for interpreting topics from the trained models by looking at the neural network weights.