Text-to-Image diffusion models have shown remarkable capabilities in generating high-quality images. However, current models often struggle to adhere to the complete set of conditions specified in the input text and return unfaithful generations. Existing works address this problem by either fine-tuning the base model or modifying the latent representations during the inference stage with gradient-based updates. Not only are these approaches computationally expensive, but also they usually only improve limited kinds of errors (e.g., the count of objects). In this work, we propose an intervention-based mechanism to enhance the faithfulness of diffusion models by controlling the denoising process. Starting with layout-conditional diffusion models, our approach first detects incorrectly-generated/missing objects during denoising steps. Next, a layout is constructed from the erroneous objects (feedback). Finally, we return to an earlier denoising step. The new layout is fed to the diffusion model to obtain its latent representation. Correction is applied by composing the new latents with the original ones and continuing the generation process, thereby driving the generation away from erroneous directions. As additional feedback and correction strategy, we also explore retrieval-augmented generation to help the model recover missing objects. We conduct experiments on VPEval and HRS-Bench datasets and measure faithfulness across four dimensions; presence of objects, object counts, scale of objects and spatial relations between objects. Compared to GLIGEN, the state-of-the-art model on the VPEval dataset, our approach significantly improves on all metrics (+6.7% average accuracy increase). On HRS-Bench dataset, it also outperforms existing models in count and scale metrics.

Abstract Decoding methods for large language models (LLMs) usually struggle with the tradeoff between ensuring factuality and maintaining diversity. In this paper, we propose REAL (Residual Entropy from Asymptotic Line) sampling,1 which predicts the step-wise hallucination likelihood of an LLM. When an LLM is likely to hallucinate, REAL lowers the p threshold in nucleus sampling. Otherwise, REAL sampling increases the p threshold to boost the diversity. To predict the step-wise hallucination likelihood without supervision, we construct a THF (Token-level Hallucination Forecasting) model, which predicts the asymptotic entropy (i.e., inherent uncertainty) of the next token by extrapolating the next-token entropies of an infinitely large language model from a series of LLMs with different sizes. If an LLM’s entropy is higher than the asymptotic entropy (i.e., the LLM is more uncertain than it should be), the THF model predicts a high hallucination hazard, which leads to a lower p threshold in REAL sampling. In the FactualityPrompts benchmark (Lee et al., 2022), we demonstrate that REAL sampling based on a 70M THF model can substantially improve the factuality and diversity of 7B LLMs simultaneously. After combined with contrastive decoding, REAL sampling outperforms 13 sampling methods, and generates texts that are more factual than the greedy sampling and more diverse than the nucleus sampling with p = 0.5.

Contrastive decoding (CD) (Li et al., 2023) improves the next-token distribution of a large expert language model (LM) using a small amateur LM.Although CD is applied to various LMs and domains to enhance open-ended text generation, it is still unclear why CD often works well, when it could fail, and how we can make it better.To deepen our understanding of CD, we first theoretically prove that CD could be viewed as linearly extrapolating the next-token logits from a huge and hypothetical LM.We also highlight that the linear extrapolation could make CD unable to output the most obvious answers that have already been assigned high probabilities by the amateur LM.To overcome CD's limitation, we propose a new unsupervised decoding method called Asymptotic Probability Decoding (APD). 1 APD explicitly extrapolates the probability curves from the LMs of different sizes to infer the asymptotic probabilities from an infinitely large LM without inducing more inference costs than CD.In FACTUALITYPROMPTS, an open-ended text generation benchmark, sampling using APD significantly boosts factuality in comparison to the CD sampling and its variants, and achieves state-of-the-art results for Pythia 6.9B and OPT 6.7B.Furthermore, in five commonsense QA datasets, APD is often significantly better than CD and achieves a similar effect of using a larger LLM.For example, the perplexity of APD on top of Pythia 6.9B is even lower than the perplexity of Pythia 12B in CommonsenseQA and LAMBADA.* The work was mostly done at Amazon.

NeurAlly-Decomposed Oracle (NADO) is a powerful approach for controllable generation with large language models. It is designed to avoid catastrophic forgetting while achieving guaranteed convergence to an entropy-maximized closed-form optimal solution with reasonable modeling capacity. Despite the success, several challenges arise when apply NADO to a wide range of scenarios. Vanilla NADO suffers from gradient vanishing for low-probability control signals and is highly reliant on a regularization to satisfy the stochastic version of Bellman equation. In addition, the vanilla implementation of NADO introduces a few additional transformer layers, suffering from a limited capacity especially compared to other finetune-based model adaptation methods like LoRA. In this paper, we propose a improved version of the NADO algorithm, namely DiNADO (norm-Disentangled NeurAlly-Decomposed Oracles), which improves the performance of the NADO algorithm through disentangling the step-wise global norm over the approximated oracle $R$-value for all potential next-tokens, allowing DiNADO to be combined with finetuning methods like LoRA. We discuss in depth how DiNADO achieves better capacity, stability and flexibility with both empirical and theoretical results. Experiments on formality control in machine translation and the lexically constrained generation task CommonGen demonstrates the significance of the improvements.

Mingyu Ma, Jiun-Yu Kao, Arpit Gupta, Yu-Hsiang Lin, Wenbo Zhao, Tagyoung Chung, Wei Wang, Kai-Wei Chang, Nanyun Peng. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Yufei Tian, Anjali Narayan-Chen, Shereen Oraby, Alessandra Cervone, Gunnar Sigurdsson, Chenyang Tao, Wenbo Zhao, Tagyoung Chung, Jing Huang, Nanyun Peng. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Goal-oriented dialogue agents can comfortably utilize the conversational context and understand its users' goals. However, in visually driven user experiences, these conversational agents are also required to make sense of the screen context in order to provide a proper interactive experience. In this paper, we propose a novel multimodal conversational framework where the dialogue agent's next action and their arguments are derived jointly conditioned both on the conversational and the visual context. We demonstrate the proposed approach via a prototypical furniture shopping experience for a multimodal virtual assistant.

Learning from multimodal data has become a popular research topic in recent years. Multimodal coreference resolution (MCR) is an important task in this area. MCR involves resolving the references across different modalities, e.g., text and images, which is a crucial capability for building next-generation conversational agents. MCR is challenging as it requires encoding information from different modalities and modeling associations between them. Although significant progress has been made for visual-linguistic tasks such as visual grounding, most of the current works involve single turn utterances and focus on simple coreference resolutions. In this work, we propose an MCR model that resolves coreferences made in multi-turn dialogues with scene images. We present GRAVL-BERT, a unified MCR framework which combines visual relationships between objects, background scenes, dialogue, and metadata by integrating Graph Neural Networks with VL-BERT. We present results on the SIMMC 2.0 multimodal conversational dataset, achieving the rank-1 on the DSTC-10 SIMMC 2.0 MCR challenge with F1 score 0.783. Our code is available at https://github.com/alexa/gravl-bert.

Jiao Sun, Anjali Narayan-Chen, Shereen Oraby, Alessandra Cervone, Tagyoung Chung, Jing Huang, Yang Liu, Nanyun Peng. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. 2022.

Jiao Sun, Anjali Narayan-Chen, Shereen Oraby, Shuyang Gao, Tagyoung Chung, Jing Huang, Yang Liu, Nanyun Peng. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. 2022.

Saket Dingliwal, Shuyang Gao, Sanchit Agarwal, Chien-Wei Lin, Tagyoung Chung, Dilek Hakkani-Tur. Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume. 2021.

Machine Reading Comprehension (MRC) for question answering (QA), which aims to answer a question given the relevant context passages, is an important way to test the ability of intelligence systems to understand human language. Multiple-Choice QA (MCQA) is one of the most difficult tasks in MRC because it often requires more advanced reading comprehension skills such as logical reasoning, summarization, and arithmetic operations, compared to the extractive counterpart where answers are usually spans of text within given passages. Moreover, most existing MCQA datasets are small in size, making the task even harder. We introduce MMM, a Multi-stage Multi-task learning framework for Multi-choice reading comprehension. Our method involves two sequential stages: coarse-tuning stage using out-of-domain datasets and multi-task learning stage using a larger in-domain dataset to help model generalize better with limited data. Furthermore, we propose a novel multi-step attention network (MAN) as the top-level classifier for this task. We demonstrate MMM significantly advances the state-of-the-art on four representative MCQA datasets.

Yuheng Du, Shereen Oraby, Vittorio Perera, Minmin Shen, Anjali Narayan-Chen, Tagyoung Chung, Anushree Venkatesh, Dilek Hakkani-Tur. Proceedings of the 13th International Conference on Natural Language Generation. 2020.

Sanghyun Yi, Rahul Goel, Chandra Khatri, Alessandra Cervone, Tagyoung Chung, Behnam Hedayatnia, Anu Venkatesh, Raefer Gabriel, Dilek Hakkani-Tur. Proceedings of the 12th International Conference on Natural Language Generation. 2019.

Dialog state tracking is used to estimate the current belief state of a dialog given all the preceding conversation.Machine reading comprehension, on the other hand, focuses on building systems that read passages of text and answer questions that require some understanding of passages.We formulate dialog state tracking as a reading comprehension task to answer the question what is the state of the current dialog?after reading conversational context.In contrast to traditional state tracking methods where the dialog state is often predicted as a distribution over a closed set of all the possible slot values within an ontology, our method uses a simple attention-based neural network to point to the slot values within the conversation.Experiments on MultiWOZ-2.0 cross-domain dialog dataset show that our simple system can obtain similar accuracies compared to the previous more complex methods.By exploiting recent advances in contextual word embeddings, adding a model that explicitly tracks whether a slot value should be carried over to the next turn, and combining our method with a traditional joint state tracking method that relies on closed set vocabulary, we can obtain a joint-goal accuracy of 47.33% on the standard test split, exceeding current state-of-the-art by 11.75%**.

The Alexa Meaning Representation Language (AMRL) is a compositional graph-based semantic representation that includes fine-grained types, properties, actions, and roles and can represent a wide variety of spoken language. AMRL increases the ability of virtual assistants to represent more complex requests, including logical and conditional statements as well as ones with nested clauses. Due to this representational capacity, the acquisition of large scale data resources is challenging, which limits the accuracy of resulting models. This paper has two primary contributions. First, we develop a linearization of AMRL graphs along with a deep multi-task model that predicts fine-grained types, properties, and intents. Second, we show how to jointly train a model that predicts an existing representation for spoken language understanding (SLU) along with the linearized AMRL parse. The resulting model, which leverages learned embeddings from both tasks, is able to predict the AMRL representation more accurately than other approaches, decreasing the error rates in the full parse by 3.56% absolute and reducing the amount of natively annotated data needed to train accurate parsing models.

Typical spoken language understanding systems provide narrow semantic parses using a domain-specific ontology. The parses contain intents and slots that are directly consumed by downstream domain applications. In this work we discuss expanding such systems to handle compound entities and intents by introducing a domain-agnostic shallow parser that handles linguistic coordination. We show that our model for parsing coordination learns domain-independent and slot-independent features and is able to segment conjunct boundaries of many different phrasal categories. We also show that using adversarial training can be effective for improving generalization across different slot types for coordination parsing.

We study the problem of sampling trees from forests, in the setting where probabilities for each tree may be a function of arbitrarily large tree fragments. This setting extends recent work for sampling to learn Tree Substitution Grammars to the case where the tree structure (TSG derived tree) is not fixed. We develop a Markov chain Monte Carlo algorithm which corrects for the bias introduced by unbalanced forests, and we present experiments using the algorithm to learn Synchronous Context-Free Grammar rules for machine translation. In this application, the forests being sampled represent the set of Hiero-style rules that are consistent with fixed input word-level alignments. We demonstrate equivalent machine translation performance to standard techniques but with much smaller grammars.