Arabic is known to present unique challengesfor Automatic Speech Recognition (ASR). Onone hand, its rich linguistic diversity andwide range of dialects complicate the de-velopment of robust, inclusive models. Onthe other, current multilingual ASR modelsare compute-intensive and lack proper com-prehensive evaluations. In light of thesechallenges, we distill knowledge from largeteacher models into smaller student variantsthat more efficient. We also introduce a novelhuman-annotated dataset covering five under-represented Arabic dialects for evaluation. Wefurther evaluate both our models and existingSoTA multilingual models on both standardavailable benchmarks and our new dialectaldata. Our best-distilled model’s overall perfor-mance (45.0% WER) surpasses that of a SoTAmodel twice its size (SeamlessM4T-large-v2,WER=47.0%) and its teacher model (Whisper-large-v2, WER=55.1%), and its average perfor-mance on our new dialectal data (56.9% WER)outperforms all other models. To gain more in-sight into the poor performance of these modelson dialectal data, we conduct an error analysisand report the main types of errors the differentmodels tend to make. The GitHub repositoryfor the project is available at https://github.com/UBC-NLP/distill-whisper-ar.
论文检索
输入标题、作者或关键词，从 9,460 篇学术成果中精准定位
Conversation agents fueled by Large Language Models (LLMs) are providing a new way to interact with visual data. While there have been initial attempts for image-based conversation models, this work addresses the under-explored field of video-based conversation by introducing Video-ChatGPT. It is a multimodal model that merges a video-adapted visual encoder with an LLM. The resulting model is capable of understanding and generating detailed conversations about videos. We introduce a new dataset of 100,000 video-instruction pairs used to train Video-ChatGPT acquired via manual and semi-automated pipeline that is easily scalable and robust to label noise. We also develop a quantitative evaluation framework for video-based dialogue models to objectively analyze the strengths and weaknesses of video-based dialogue models. Code: https://github.com/mbzuai-oryx/Video-ChatGPT.
Large language models (LLMs) are central to modern natural language processing, delivering exceptional performance in various tasks. However, their substantial computational and memory requirements present challenges, especially for devices with limited DRAM capacity. This paper tackles the challenge of efficiently running LLMs that exceed the available DRAM capacity by storing the model parameters in flash memory, but bringing them on demand to DRAM. Our method involves constructing an inference cost model that takes into account the characteristics of flash memory, guiding us to optimize in two critical areas: reducing the volume of data transferred from flash and reading data in larger, more contiguous chunks. Within this hardware-informed framework, we introduce two principal techniques. First, “windowing” strategically reduces data transfer by reusing previously activated neurons, and second, “row-column bundling”, tailored to the sequential data access strengths of flash memory, increases the size of data chunks read from flash memory. These methods collectively enable running models up to twice the size of the available DRAM, with a 4-5x and 20-25x increase in inference speed compared to naive loading approaches in CPU and GPU, respectively. Our integration of sparsity awareness, context-adaptive loading, and a hardware-oriented design paves the way for effective inference of LLMs on devices with limited memory.
Analyzing LLM Behavior in Dialogue Summarization: Unveiling Circumstantial Hallucination Trends
PDF ↗Recent advancements in large language models (LLMs) have significantly advanced the capabilities of summarization systems.However, they continue to face a persistent challenge: hallucination. While prior work has extensively examined LLMs in news domains, evaluation of dialogue summarization has primarily focused on BART-based models, resulting in a notable gap in understanding LLM effectiveness.Our work seeks to address this gap by benchmarking LLMs for dialogue summarization faithfulness using human annotations,focusing on identifying and categorizing span-level inconsistencies.Specifically, we evaluate two prominent LLMs: GPT-4 and Alpaca-13B.Our evaluation reveals that LLMs often generate plausible, but not fully supported inferences based on conversation contextual cues, a trait absent in older models. As a result, we propose a refined taxonomy of errors, introducing a novel category termed “Contextual Inference” to address this aspect of LLM behavior. Using our taxonomy, we compare the behavioral differences between LLMs and older fine-tuned models. Additionally, we systematically assess the efficacy of automatic error detection methods on LLM summaries and find that they struggle to detect these nuanced errors effectively. To address this, we introduce two prompt-based approaches for fine-grained error detection. Our methods outperform existing metrics, particularly in identifying the novel “Contextual Inference” error type.
The performance of modern language models (LMs) has been improved by chain-of-thought (CoT) reasoning, i.e., the process of generating intermediate results that guide the model towards a final answer. A possible explanation for this improvement is that CoT reasoning extends an LM’s computational power, as RNNs and transformers with additional scratch space are known to be Turing complete. Comparing LMs to Turing machines, however, introduces a category error—Turing machines decide language membership, whereas LMs define distributions over strings. To bridge this gap, we formalize CoT reasoning in a probabilistic setting. We present several results on the representational capacity of recurrent and transformer LMs with CoT reasoning, showing that they can represent the same family of distributions over strings as probabilistic Turing machines.
A dog whistle is a form of coded communication that carries a secondary meaning to specific audiences and is often weaponized for racial and socioeconomic discrimination. Dog whistling historically originated from United States politics, but in recent years has taken root in social media as a means of evading hate speech detection systems and maintaining plausible deniability. In this paper, we present an approach for word-sense disambiguation of dog whistles from standard speech using Large Language Models (LLMs), and leverage this technique to create a dataset of 16,550 high-confidence coded examples of dog whistles used in formal and informal communication. Silent Signals is the largest dataset of disambiguated dog whistle usage, created for applications in hate speech detection, neology, and political science.
Many commercial and open-source models claim to detect machine-generated text with extremely high accuracy (99% or more). However, very few of these detectors are evaluated on shared benchmark datasets and even when they are, the datasets used for evaluation are insufficiently challenging—lacking variations in sampling strategy, adversarial attacks, and open-source generative models. In this work we present RAID: the largest and most challenging benchmark dataset for machine-generated text detection. RAID includes over 6 million generations spanning 11 models, 8 domains, 11 adversarial attacks and 4 decoding strategies. Using RAID, we evaluate the out-of-domain and adversarial robustness of 8 open- and 4 closed-source detectors and find that current detectors are easily fooled by adversarial attacks, variations in sampling strategies, repetition penalties, and unseen generative models. We release our data along with a leaderboard to encourage future research.
We introduce VoiceCraft, a token infilling neural codec language model, that achieves state-of-the-art performance on both speech editing and zero-shot text-to-speech (TTS) on audiobooks, internet videos, and podcasts. VoiceCraft employs a Transformer decoder architecture and introduces a token rearrangement procedure that combines causal masking and delayed stacking to enable generation within an existing sequence. On speech editing tasks, VoiceCraft produces edited speech that is nearly indistinguishable from unedited recordings in terms of naturalness, as evaluated by humans; for zero-shot TTS, our model outperforms prior SotA models including VALL-E and the popular commercial model XTTS v2. Crucially, the models are evaluated on challenging and realistic datasets, that consist of diverse accents, speaking styles, recording conditions, and background noise and music, and our model performs consistently well compared to other models and real recordings. In particular, for speech editing evaluation, we introduce a high quality, challenging, and realistic dataset named . We encourage readers to listen to the demos at https://jasonppy.github.io/VoiceCraft_web. Data, code, and model weights are available at https://github.com/jasonppy/VoiceCraft
The boardgame Diplomacy is a challenging setting for communicative and cooperative artificial intelligence. The most prominent communicative Diplomacy AI, Cicero, has excellent strategic abilities, exceeding human players. However, the best Diplomacy players master communication, not just tactics, which is why the game has received attention as an AI challenge. This work seeks to understand the degree to which Cicero succeeds at communication. First, we annotate in-game communication with abstract meaning representation to separate in-game tactics from general language. Second, we run two dozen games with humans and Cicero, totaling over 200 human-player hours of competition. While AI can consistently outplay human players, AI-Human communication is still limited because of AI’s difficulty with deception and persuasion. This shows that Cicero relies on strategy and has not yet reached the full promise of communicative and cooperative AI.
The intricate relationship between language and culture has long been a subject of exploration within the realm of linguistic anthropology. Large Language Models (LLMs), promoted as repositories of collective human knowledge, raise a pivotal question: do these models genuinely encapsulate the diverse knowledge adopted by different cultures? Our study reveals that these models demonstrate greater cultural alignment along two dimensions—firstly, when prompted with the dominant language of a specific culture, and secondly, when pretrained with a refined mixture of languages employed by that culture. We quantify cultural alignment by simulating sociological surveys, comparing model responses to those of actual survey participants as references. Specifically, we replicate a survey conducted in various regions of Egypt and the United States through prompting LLMs with different pretraining data mixtures in both Arabic and English with the personas of the real respondents and the survey questions. Further analysis reveals that misalignment becomes more pronounced for underrepresented personas and for culturally sensitive topics, such as those probing social values. Finally, we introduce Anthropological Prompting, a novel method leveraging anthropological reasoning to enhance cultural alignment. Our study emphasizes the necessity for a more balanced multilingual pretraining dataset to better represent the diversity of human experience and the plurality of different cultures with many implications on the topic of cross-lingual transfer.
Closed-source agents suffer from several issues such as a lack of affordability, transparency, and reproducibility, particularly on complex interactive tasks. This motivates the development of open-source alternatives. We introduce Lumos, one of the first frameworks for training open-source LLM-based agents. Lumos features a learnable, unified and modular architecture with a planning module that learns high-level subgoal generation, and a grounding module trained to translate these into the actions using various tools in the execution module. The design allows for modular upgrades and wider applicability to diverse interactive tasks. To foster generalizable agent learning, we collect large-scale, unified, and high-quality training annotations derived from diverse ground-truth reasoning rationales across various complex interactive tasks. On 9 datasets, Lumos exhibits several key advantages: (1) Lumos excels multiple larger open-source agents on the held-out datasets (unused for training) for each task type. Lumos even surpasses GPT agents on QA and web tasks; (2) Lumos outperforms open-source agents produced by chain-of-thoughts and unmodularized integrated training; and (3) Lumos effectively generalizes to unseen tasks, outperforming 33B-scale agents and domain-specific agents. Code and data will be released.
Large language models (LLMs) exhibit an intriguing ability to learn a novel task from in-context examples presented in a demonstration, termed in-context learning (ICL). Understandably, a swath of research has been dedicated to uncovering the theories underpinning ICL. One popular hypothesis explains ICL by task selection. LLMs identify the task based on the demonstration and generalize it to the prompt. Another popular hypothesis is that ICL is a form of meta-learning, i.e., the models learn a learning algorithm at pre-training time and apply it to the demonstration. Finally, a third hypothesis argues that LLMs use the demonstration to select a composition of tasks learned during pre-training to perform ICL. In this paper, we empirically explore these three hypotheses that explain LLMs’ ability to learn in context with a suite of experiments derived from common text classification tasks. We invalidate the first two hypotheses with counterexamples and provide evidence in support of the last hypothesis. Our results suggest an LLM could learn a novel task in context via composing tasks learned during pre-training.
Transferable and Efficient Non-Factual Content Detection via Probe Training with Offline Consistency Checking
PDF ↗This paper proposes PiNose, which trains a probing model on offline self-consistency checking results, thereby circumventing the need for human-annotated data and achieving transferability across diverse data distributions. As the consistency check process is offline, PiNose reduces the computational burden of generating multiple responses by online consistency verification. Additionally, it examines various aspects of internal states prior to response decoding, contributing to more effective detection of factual inaccuracies. Experiment results on both factuality detection and question answering benchmarks show that PiNose achieves surpassing results than existing factuality detection methods.
Recent studies have shown that Text-to-Image (T2I) model generations can reflect social stereotypes present in the real world. However, existing approaches for evaluating stereotypes have a noticeable lack of coverage of global identity groups and their associated stereotypes. To address this gap, we introduce the ViSAGe (Visual Stereotypes Around the Globe) dataset to enable the evaluation of known nationality-based stereotypes in T2I models, across 135 nationalities. We enrich an existing textual stereotype resource by distinguishing between stereotypical associations that are more likely to have visual depictions, such as ‘sombrero’, from those that are less visually concrete, such as ‘attractive’. We demonstrate ViSAGe’s utility through a multi-faceted evaluation of T2I generations. First, we show that stereotypical attributes in ViSAGe are thrice as likely to be present in generated images of corresponding identities as compared to other attributes, and that the offensiveness of these depictions is especially higher for identities from Africa, South America, and South East Asia. Second, we assess the ‘stereotypical pull’ of visual depictions of identity groups, which reveals how the ‘default’ representations of all identity groups in ViSAGe have a pull towards stereotypical depictions, and that this pull is even more prominent for identity groups from the Global South. CONTENT WARNING: Some examples contain offensive stereotypes.
We present the structured average intersection-over-union ratio (STRUCT-IOU), an evaluation metric that compares a constituency parse tree over automatically recognized spoken word boundaries with the ground-truth parse tree over written words. To compute the metric, we (1) project the ground-truth parse tree to the speech domain by forced alignment, (2) align the projected ground-truth constituents with the predicted ones under certain structured constraints, and (3) calculate the average IOU score across all aligned constituent pairs. STRUCT-IOU takes word boundaries into account and overcomes the challenge that the predicted words and ground truth may not have perfect one-to-one correspondence. Extending to the evaluation of text constituency parsing, we demonstrate that STRUCT-IOU shows higher tolerance to syntactically plausible parses than PARSEVAL (Black et al., 1991).
Tree-of-Traversals: A Zero-Shot Reasoning Algorithm for Augmenting Black-box Language Models with Knowledge Graphs
PDF ↗Knowledge graphs (KGs) complement Large Language Models (LLMs) by providing reliable, structured, domain-specific, and up-to-date external knowledge. However, KGs and LLMs are often developed separately and must be integrated after training. We introduce Tree-of-Traversals, a novel zero-shot reasoning algorithm that enables augmentation of black-box LLMs with one or more KGs. The algorithm equips a LLM with actions for interfacing a KG and enables the LLM to perform tree search over possible thoughts and actions to find high confidence reasoning paths. Tree-of-Traversals significantly improves performance on question answering and KG question answering tasks. Code is available at https://github.com/amazon-science/tree-of-traversals
Tree Transformer’s Disambiguation Ability of Prepositional Phrase Attachment and Garden Path Effects
PDF ↗This work studies two types of ambiguity in natural language: prepositional phrase (PP) attachment ambiguity, and garden path constructions. Due to the different nature of these ambiguities – one being structural, the other incremental in nature – we pretrain and evaluate the Tree Transformer of Wang et al. (2019), an unsupervised Transformer model that induces tree representations internally. To assess PP attachment ambiguity we inspect the model’s induced parse trees against a newly prepared dataset derived from the PP attachment corpus (Ratnaparkhi et al., 1994). Measuring garden path effects is done by considering surprisal rates of the underlying language model on a number of dedicated test suites, following Futrell et al. (2019). For comparison we evaluate a pretrained supervised BiLSTM-based model trained on constituency parsing as sequence labelling (Gómez-Rodríguez and Vilares, 2018). Results show that the unsupervised Tree Transformer does exhibit garden path effects, but its parsing ability is far inferior to the supervised BiLSTM, and it is not as sensitive to lexical cues as other large LSTM models, suggesting that supervised parsers based on a pre-Transformer architecture may be the better choice in the presence of ambiguity.
In the rapidly advancing field of conditional image generation research, challenges such as limited explainability lie in effectively evaluating the performance and capabilities of various models. This paper introduces VIEScore, a Visual Instruction-guided Explainable metric for evaluating any conditional image generation tasks. VIEScore leverages general knowledge from Multimodal Large Language Models (MLLMs) as the backbone and does not require training or fine-tuning. We evaluate VIEScore on seven prominent tasks in conditional image tasks and found: (1) VIEScore (GPT4-o) achieves a high Spearman correlation of 0.4 with human evaluations, while the human-to-human correlation is 0.45. (2) VIEScore (with open-source MLLM) is significantly weaker than GPT-4o and GPT-4v in evaluating synthetic images. (3) VIEScore achieves a correlation on par with human ratings in the generation tasks but struggles in editing tasks. With these results, we believe VIEScore shows its great potential to replace human judges in evaluating image synthesis tasks.
Back to Basics: Revisiting REINFORCE-Style Optimization for Learning from Human Feedback in LLMs
PDF ↗AI alignment in the shape of Reinforcement Learning from Human Feedback (RLHF) is increasingly treated as a crucial ingredient for high performance large language models. Proximal Policy Optimization (PPO) has been installed by the seminal literature as the standard method for the RL part of RLHF. However, it involves both high computational cost and sensitive hyperparameter tuning. We posit that most of the motivational principles that led to the development of PPO are less of a practical concern in RLHF and advocate for a less computationally expensive method that preserves and even increases performance. We revisit how alignment from human preferences is formulated in the context of RL. Keeping simplicity as a guiding principle, we show that many components of PPO are unnecessary in an RLHF context and that far simpler REINFORCE-style optimization variants outperform both PPO and newly proposed “RL-free” methods such as DPO and RAFT. Our work suggests that careful adaptation to LLMs alignment characteristics allows benefiting from online RL optimization at low cost.
Embeddings extracted by pre-trained Large Language Models (LLMs) have significant potential to improve information retrieval and search. Beyond the zero-shot setup in which they are being conventionally used, being able to take advantage of the information from the relevant query-corpus paired data can further boost the LLM capabilities. In this paper, we propose a novel method, Search-Adaptor, for customizing LLMs for information retrieval in an efficient and robust way. Search-Adaptor modifies the embeddings generated by pre-trained LLMs, and can be integrated with any LLM, including those only available via prediction APIs. On multiple English, multilingual, and multimodal retrieval datasets, we show consistent and significant performance benefits for Search-Adaptor – e.g., more than 5% improvements for Google Embedding APIs in nDCG@10 averaged over 14 BEIR datasets.