We study the problem of entropy calibration, which asks whether a language model's entropy over generations matches its log loss on human text. Past work found that models are miscalibrated, with entropy per step increasing (and text quality decreasing) as generations grow longer. This error accumulation is a fundamental problem in autoregressive models, and the standard solution is to truncate the distribution, which improves text quality at the cost of diversity. In this paper, we ask: is miscalibration likely to improve with scale, and is it theoretically possible to calibrate without tradeoffs? To build intuition, we first study a simplified theoretical setting to characterize the scaling behavior of miscalibration with respect to dataset size. We find that the scaling behavior depends on the power law exponent of the data distribution -- in particular, for a power law exponent close to 1, the scaling exponent is close to 0, meaning that miscalibration improves very slowly with scale. Next, we measure miscalibration empirically in language models ranging from 0.5B to 70B parameters. We find that the observed scaling behavior is similar to what is predicted by the simplified setting: our fitted scaling exponents for text are close to 0, meaning that larger models accumulate error at a similar rate as smaller ones. This scaling (or, lack thereof) provides one explanation for why we sample from larger models with similar amounts of truncation as smaller models, even though the larger models are of higher quality. However, truncation is not a satisfying solution because it comes at the cost of increased log loss. In theory, is it even possible to reduce entropy while preserving log loss? We prove that it is possible, if we assume access to a black box which can fit models to predict the future entropy of text.

API calls for each position identified in a piece of text. Question Answering We use the Atlas model of Izacard et al. (2022) finetuned on Natural Questions Calculator Our calculator is based on a simple Python script and only supports the operators " It does not return any result for syntactically invalid equations. "=", "equals", "equal to", "total of", "average of" followed by a number, or (iii) contain at least three English text before generating API calls. Below, we list the prompts used to sample API calls for each tool considered. Your task is to add calls to a Question Answering API to a piece of text. Input: Joe Biden was born in Scranton, Pennsylvania. Output: Joe Biden was born in [QA("Where was Joe Biden born?")] Scranton, [QA("In Output: Coca-Cola, or [QA("What other name is Coca-Cola known by?")] Coke, is Your task is to add calls to a Calculator API to a piece of text.

Inductive knowledge graph completion (KGC) aims to predict missing triples with unseen entities. Recent works focus on modeling reasoning paths between the head and tail entity as direct supporting evidence. However, these methods depend heavily on the existence and quality of reasoning paths, which limits their general applicability in different scenarios. In addition, we observe that latent type constraints and neighboring facts inherent in KGs are also vital in inferring missing triples. To effectively utilize all useful information in KGs, we introduce CATS, a novel context-aware inductive KGC solution. With sufficient guidance from proper prompts and supervised fine-tuning, CATS activates the strong semantic understanding and reasoning capabilities of large language models to assess the existence of query triples, which consist of two modules. First, the type-aware reasoning module evaluates whether the candidate entity matches the latent entity type as required by the query relation. Then, the subgraph reasoning module selects relevant reasoning paths and neighboring facts, and evaluates their correlation to the query triple. Experiment results on three widely used datasets demonstrate that CATS significantly outperforms state-of-the-art methods in 16 out of 18 transductive, inductive, and few-shot settings with an average absolute MRR improvement of 7.2%.

Our study delves into Multimodal Entity Linking, aligning the mention in multimodal information with entities in knowledge base. Existing methods are still facing challenges like ambiguous entity representations and limited image information utilization. Thus, we propose dynamic entity extraction using ChatGPT, which dynamically extracts entities and enhances datasets. We also propose a method: Dynamically Integrate Multimodal information with knowledge base (DIM), employing the capability of the Large Language Model (LLM) for visual understanding. The LLM, such as BLIP-2, extracts information relevant to entities in the image, which can facilitate improved extraction of entity features and linking them with the dynamic entity representations provided by ChatGPT. The experiments demonstrate that our proposed DIM method outperforms the majority of existing methods on the three original datasets, and achieves state-of-the-art (SOTA) on the dynamically enhanced datasets (Wiki+, Rich+, Diverse+).

In this work, we explicitly show that modern LLMs tend to generate correct facts first, then "drift away" and generate incorrect facts later: this was occasionally observed but never properly measured. We develop a semantic drift score that measures the degree of separation between correct and incorrect facts in generated texts and confirm our hypothesis when generating Wikipedia-style biographies. This correct-then-incorrect generation pattern suggests that factual accuracy can be improved by knowing when to stop generation. Therefore, we explore the trade-off between information quantity and factual accuracy for several early stopping methods and manage to improve factuality by a large margin. We further show that reranking with semantic similarity can further improve these results, both compared to the baseline and when combined with early stopping. Finally, we try calling external API to bring the model back to the right generation path, but do not get positive results. Overall, our methods generalize and can be applied to any long-form text generation to produce more reliable information, by balancing trade-offs between factual accuracy, information quantity and computational cost.

Large language models (LLMs) can make predictions using parametric knowledge--knowledge encoded in the model weights--or contextual knowledge--knowledge presented in the context. In many scenarios, a desirable behavior is that LLMs give precedence to contextual knowledge when it conflicts with the parametric knowledge, and fall back to using their parametric knowledge when the context is irrelevant. This enables updating and correcting the model's knowledge by in-context editing instead of retraining. Previous works have shown that LLMs are inclined to ignore contextual knowledge and fail to reliably fall back to parametric knowledge when presented with irrelevant context. In this work, we discover that, with proper prompting methods, instruction-finetuned LLMs can be highly controllable by contextual knowledge and robust to irrelevant context. Utilizing this feature, we propose EREN (Edit models by REading Notes) to improve the scalability and robustness of LLM editing. To better evaluate the robustness of model editors, we collect a new dataset, that contains irrelevant questions that are more challenging than the ones in existing datasets. Empirical results show that our method outperforms current state-of-the-art methods by a large margin. Unlike existing techniques, it can integrate knowledge from multiple edits, and correctly respond to syntactically similar but semantically unrelated inputs (and vice versa). The source code can be found at https://github.com/thunlp/EREN.

Neural knowledge-to-text generation models often struggle to faithfully generate descriptions for the input facts: they may produce hallucinations that contradict the given facts, or describe facts not present in the input. To reduce hallucinations, we propose a novel decoding method, TWEAK (Think While Effectively Articulating Knowledge). TWEAK treats the generated sequences at each decoding step and its future sequences as hypotheses, and ranks each generation candidate based on how well their corresponding hypotheses support the input facts using a Hypothesis Verification Model (HVM). We first demonstrate the effectiveness of TWEAK by using a Natural Language Inference (NLI) model as the HVM and report improved faithfulness with minimal impact on the quality. We then replace the NLI model with our task-specific HVM trained with a first-of-a-kind dataset, FATE (Fact-Aligned Textual Entailment), which pairs input facts with their faithful and hallucinated descriptions with the hallucinated spans marked. The new HVM improves the faithfulness and the quality further and runs faster. Overall the best TWEAK variants improve on average 2.22/7.17 points on faithfulness measured by FactKB over WebNLG and TekGen/GenWiki, respectively, with only 0.14/0.32 points degradation on quality measured by BERTScore over the same datasets. Since TWEAK is a decoding-only approach, it can be integrated with any neural generative model without retraining.

One challenge in speech translation is that plenty of spoken content is long-form, but short units are necessary for obtaining high-quality translations. To address this mismatch, we adapt large language models (LLMs) to split long ASR transcripts into segments that can be independently translated so as to maximize the overall translation quality. We overcome the tendency of hallucination in LLMs by incorporating finite-state constraints during decoding; these eliminate invalid outputs without requiring additional training. We discover that LLMs are adaptable to transcripts containing ASR errors through prompt-tuning or fine-tuning. Relative to a state-of-the-art automatic punctuation baseline, our best LLM improves the average BLEU by 2.9 points for English-German, English-Spanish, and English-Arabic TED talk translation in 9 test sets, just by improving segmentation.

Memes are a powerful tool for communication over social media. Their affinity for evolving across politics, history, and sociocultural phenomena makes them an ideal communication vehicle. To comprehend the subtle message conveyed within a meme, one must understand the background that facilitates its holistic assimilation. Besides digital archiving of memes and their metadata by a few websites like knowyourmeme.com, currently, there is no efficient way to deduce a meme's context dynamically. In this work, we propose a novel task, MEMEX - given a meme and a related document, the aim is to mine the context that succinctly explains the background of the meme. At first, we develop MCC (Meme Context Corpus), a novel dataset for MEMEX. Further, to benchmark MCC, we propose MIME (MultImodal Meme Explainer), a multimodal neural framework that uses common sense enriched meme representation and a layered approach to capture the cross-modal semantic dependencies between the meme and the context. MIME surpasses several unimodal and multimodal systems and yields an absolute improvement of ~ 4% F1-score over the best baseline. Lastly, we conduct detailed analyses of MIME's performance, highlighting the aspects that could lead to optimal modeling of cross-modal contextual associations.