The recent breakthroughs in Artificial Intelligence (AI) can be attributed to the remarkable performance of Large Language Models (LLMs) across a spectrum of research areas (e.g., machine translation, question-answering, automatic speech recognition, text-to-speech generation) and application domains (e.g., business, law, healthcare, education, and psychology). The success of these LLMs largely depends on specific training techniques, most notably instruction tuning, RLHF, and subsequent prompting to achieve the desired output. As the development of such LLMs continues to increase in both closed and open settings, evaluation has become crucial for understanding their generalization capabilities across different tasks, modalities, languages, and dialects. This evaluation process is tightly coupled with prompting, which plays a key role in obtaining better outputs. There has been attempts to evaluate such models focusing on diverse tasks, languages, and dialects, which suggests that the capabilities of LLMs are still limited for medium-to-low-resource languages due to the lack of representative datasets. The tutorial offers an overview of this emerging research area. We explore the capabilities of LLMs in terms of their performance, zero- and few-shot settings, fine-tuning, instructions tuning, and close vs. open models with a special emphasis on low-resource settings. In addition to LLMs for standard NLP tasks, we will focus on speech and multimodality.

LLMs for Low Resource Languages in Multilingual, Multimodal and Dialectal Settings

In this work, we focus on the task of determining the public attitude toward various social issues discussed on social media platforms. Platforms such as Twitter, however, are often used to spread misinformation, fake news through polarizing views. Existing literature suggests that higher levels of toxicity prevalent in Twitter conversations often spread negativity and delay addressing issues. Further, the embedded moral values and speech acts specifying the intention of the tweet correlate with public opinions expressed on various topics. However, previous works, which mainly focus on stance detection, either ignore the speech act, toxic, and moral features of these tweets that can collectively help capture public opinion or lack an efficient architecture that can detect the attitudes across targets. Therefore, in our work, we focus on the main task of stance detection by exploiting the toxicity, morality, and speech act as auxiliary tasks. We propose a multitasking model TWISTED that initially extracts the valence, arousal, and dominance aspects hidden in the tweets and injects the emotional sense into the embedded text followed by an efficient attention framework to correctly detect the tweet's stance by using the shared features of toxicity, morality, and speech acts present in the tweet. Extensive experiments conducted on 4 benchmark stance detection datasets (SemEval-2016, P-Stance, COVID19-Stance, and ClimateChange) comprising different domains demonstrate the effectiveness and generalizability of our approach.

Toxicity, Morality, and Speech Act Guided Stance Detection

We introduce a new cross-modal fusion technique designed for generative error correction in automatic speech recognition (ASR). Our methodology leverages both acoustic information and external linguistic representations to generate accurate speech transcription contexts. This marks a step towards a fresh paradigm in generative error correction within the realm of n-best hypotheses. Unlike the existing ranking-based rescoring methods, our approach adeptly uses distinct initialization techniques and parameter-efficient algorithms to boost ASR performance derived from pre-trained speech and text models. Through evaluation across diverse ASR datasets, we assess our fusion technique, demonstrating a 37.66\% improvement in word error rate (WER) relative performance compared to the n-best Oracle. To encourage future research, we have made our code and pre-trained models open source at [https://github.com/Srijith-rkr/Whispering-LLaMA](https://github.com/Srijith-rkr/Whispering-LLaMA)

Whispering LLaMA: A Cross-Modal Generative Error Correction Framework for Speech Recognition

Transformers have become a key architecture in speech processing, but our understanding of how they build up representations of acoustic and linguistic structure is limited. In this study, we address this gap by investigating how measures of 'context-mixing' developed for text models can be adapted and applied to models of spoken language. We identify a linguistic phenomenon that is ideal for such a case study: homophony in French (e.g. livre vs livres), where a speech recognition model has to attend to syntactic cues such as determiners and pronouns in order to disambiguate spoken words with identical pronunciations and transcribe them while respecting grammatical agreement. We perform a series of controlled experiments and probing analyses on Transformer-based speech models. Our findings reveal that representations in encoder-only models effectively incorporate these cues to identify the correct transcription, whereas encoders in encoder-decoder models mainly relegate the task of capturing contextual dependencies to decoder modules.

Homophone Disambiguation Reveals Patterns of Context Mixing in Speech Transformers

Significant improvements in end-to-end speech translation (ST) have been achieved through the application of multi-task learning. However, the extent to which auxiliary tasks are highly consistent with the ST task, and how much this approach truly helps, have not been thoroughly studied. In this paper, we investigate the consistency between different tasks, considering different times and modules. We find that the textual encoder primarily facilitates cross-modal conversion, but the presence of noise in speech impedes the consistency between text and speech representations. Furthermore, we propose an improved multi-task learning (IMTL) approach for the ST task, which bridges the modal gap by mitigating the difference in length and representation. We conduct experiments on the MuST-C dataset. The results demonstrate that our method attains state-of-the-art results. Moreover, when additional data is used, we achieve the new SOTA result on MuST-C English to Spanish task with 20.8\% of the training time required by the current SOTA method.

Rethinking and Improving Multi-task Learning for End-to-end Speech Translation | VIDEO

Counterspeech, i.e., responses to counteract potential harms of hateful speech, has become an increasingly popular solution to address online hate speech without censorship. However, properly countering hateful language requires countering and dispelling the underlying inaccurate stereotypes implied by such language. In this work, we draw from psychology and philosophy literature to craft six psychologically inspired strategies to challenge the underlying stereotypical implications of hateful language. We first examine the convincingness of each of these strategies through a user study, and then compare their usages in both human- and machine-generated counterspeech datasets. Our results show that human-written counterspeech uses countering strategies that are more specific to the implied stereotype (e.g., counter examples to the stereotype, external factors about the stereotype’s origins), whereas machine-generated counterspeech uses less specific strategies (e.g., generally denouncing the hatefulness of speech). Furthermore, machine generated counterspeech often employs strategies that humans deem less convincing compared to human-produced counterspeech. Our findings point to the importance of accounting for the underlying stereotypical implications of speech when generating counterspeech and for better machine reasoning about anti-stereotypical examples.

Beyond Denouncing Hate: Strategies for Countering Implied Biases and Stereotypes in Language

In this paper, we present a methodology for linguistic feature extraction, focusing particularly on automatically syllabifying words in multiple languages, with a design to be compatible with a forced-alignment tool, the Montreal Forced Aligner (MFA). In both the textual and phonetic domains, our method focuses on the extraction of phonetic transcriptions from text, stress marks, and a unified automatic syllabification (in text and phonetic domains). The system was built with open-source components and resources. Through an ablation study, we demonstrate the efficacy of our approach in automatically syllabifying words from several languages (English, French and Spanish). Additionally, we apply the technique to the transcriptions of the CMU ARCTIC dataset, generating valuable annotations available online (https://github.com/noetits/MUST_P-SRL) that are ideal for speech representation learning, speech unit discovery, and disentanglement of speech factors in several speech-related fields.

MUST&P-SRL: Multi-lingual and Unified Syllabification in Text and Phonetic Domains for Speech Representation Learning

Large Language Models (LLMs) exhibit impressive performance on a range of NLP tasks, due to the general-purpose linguistic knowledge acquired during pretraining. Existing model interpretability research (Tenney et al., 2019) suggests that a linguistic hierarchy emerges in the LLM layers, with lower layers better suited to solving syntactic tasks and higher layers employed for semantic processing. Yet, little is known about how encodings of different linguistic phenomena interact within the models and to what extent processing of linguistically-related categories relies on the same, shared model representations. In this paper, we propose a framework for testing the joint encoding of linguistic categories in LLMs. Focusing on syntax, we find evidence of joint encoding both at the same (related part-of-speech (POS) classes) and different (POS classes and related syntactic dependency relations) levels of linguistic hierarchy. Our cross-lingual experiments show that the same patterns hold across languages in multilingual LLMs.

Probing LLMs for Joint Encoding of Linguistic Categories

We present a novel approach to multilingual audio-visual speech recognition tasks by introducing a single model on a multilingual dataset. Motivated by a human cognitive system where humans can intuitively distinguish different languages without any conscious effort or guidance, we propose a model that can capture which language is given as an input speech by distinguishing the inherent similarities and differences between languages. To do so, we design a prompt fine-tuning technique into the largely pre-trained audio-visual representation model so that the network can recognize the language class as well as the speech with the corresponding language. Our work contributes to developing robust and efficient multilingual audio-visual speech recognition systems, reducing the need for language-specific models.

Intuitive Multilingual Audio-Visual Speech Recognition with a Single-Trained Model

Countering rather than censoring hate speech has emerged as a promising strategy to address hatred. There are many types of counterspeech in user-generated content: addressing the hateful content or its author, generic requests, well-reasoned counter arguments, insults, etc. The effectiveness of counterspeech, which we define as subsequent incivility, depends on these types. In this paper, we present a theoretically grounded taxonomy of replies to hate speech and a new corpus. We work with real, user-generated hate speech and all the replies it elicits rather than replies generated by a third party. Our analyses provide insights into the content real users reply with as well as which replies are empirically most effective. We also experiment with models to characterize the replies to hate speech, thereby opening the door to estimating whether a reply to hate speech will result in further incivility.

A Fine-Grained Taxonomy of Replies to Hate Speech

The tremendous growth of social media users interacting in online conversations has led to significant growth in hate speech affecting people from various demographics. Most of the prior works focus on detecting explicit hate speech, which is overt and leverages hateful phrases, with very little work focusing on detecting hate speech that is implicit or denotes hatred through indirect or coded language. In this paper, we present CoSyn, a context synergized neural network that explicitly incorporates user- and conversational-context for detecting implicit hate speech in online conversations. CoSyn introduces novel ways to encode these external contexts and employs a novel context interaction mechanism that clearly captures the interplay between them, making independent assessments of the amounts of information to be retrieved from these noisy contexts. Additionally, it carries out all these operations in the hyperbolic space to account for the scale-free dynamics of social media. We demonstrate the effectiveness of CoSyn on 6 hate speech datasets and show that CoSyn outperforms all our baselines in detecting implicit hate speech with absolute improvements in the range of 1.24% - 57.8%. We make our code available.

CoSyn: Detecting Implicit Hate Speech in Online Conversations Using a Context Synergized Hyperbolic Network

Keyphrase Generation (KPG) is a longstanding task in NLP with widespread applications. The advent of sequence-to-sequence (seq2seq) pre-trained language models (PLMs) has ushered in a transformative era for KPG, yielding promising performance improvements. However, many design decisions remain unexplored and are often made arbitrarily. This paper undertakes a systematic analysis of the influence of model selection and decoding strategies on PLM-based KPG. We begin by elucidating why seq2seq PLMs are apt for KPG, anchored by an attention-driven hypothesis. We then establish that conventional wisdom for selecting seq2seq PLMs lacks depth: (1) merely increasing model size or performing task-specific adaptation is not parameter-efficient; (2) although combining in-domain pre-training with task adaptation benefits KPG, it does partially hinder generalization. Regarding decoding, we demonstrate that while greedy search achieves strong F1 scores, it lags in recall compared with sampling-based methods. Based on these insights, we propose DeSel, a likelihood-based decode-select algorithm for seq2seq PLMs. DeSel improves greedy search by an average of 4.7% semantic F1 across five datasets. Our collective findings pave the way for deeper future investigations into PLM-based KPG.

Underline digital video library

LLMs for Low Resource Languages in Multilingual, Multimodal and Dialectal Settings

Toxicity, Morality, and Speech Act Guided Stance Detection

Whispering LLaMA: A Cross-Modal Generative Error Correction Framework for Speech Recognition

Homophone Disambiguation Reveals Patterns of Context Mixing in Speech Transformers

Rethinking and Improving Multi-task Learning for End-to-end Speech Translation | VIDEO

Beyond Denouncing Hate: Strategies for Countering Implied Biases and Stereotypes in Language

MUST&P-SRL: Multi-lingual and Unified Syllabification in Text and Phonetic Domains for Speech Representation Learning

Probing LLMs for Joint Encoding of Linguistic Categories

Intuitive Multilingual Audio-Visual Speech Recognition with a Single-Trained Model

A Fine-Grained Taxonomy of Replies to Hate Speech

CoSyn: Detecting Implicit Hate Speech in Online Conversations Using a Context Synergized Hyperbolic Network

Rethinking Model Selection and Decoding for Keyphrase Generation with Pre-trained Sequence-to-Sequence Models

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES