EMNLP 2024 Papers

Zero-shot inference, where pre-trained models perform tasks without specific training data, is an exciting emergent ability of large models like CLIP. Although there has been considerable exploration into enhancing zero-shot abilities in image captioning (IC) for popular datasets such as MSCOCO and Flickr8k, these approaches fall short with fine-grained datasets like CUB, FLO, UCM-Captions, and Sydney-Captions. These datasets require captions to discern between visually and semantically similar classes, focusing on detailed object parts and their attributes. To overcome this challenge, we introduce TRaining-Free Object-Part Enhancement (TROPE). TROPE enriches a base caption with additional object-part details using object detector proposals and natural language processing techniques. It complements rather than alters the base caption, allowing seamless integration with other captioning methods and offering users enhanced flexibility. Our evaluations show that TROPE consistently boosts performance across all tested zero-shot IC approaches and achieves state-of-the-art results on fine-grained IC datasets.

A sophisticated negotiation dialogue system for tourism should engage in negotiations beyond mere price considerations, encompassing various other aspects and amenities inherent in the tourism package. To ensure such tailored interaction, it is imperative to understand the intricacies of traveler preferences, constraints, and expectations. Incorporating these personality facets allows for customizing negotiation strategies, resulting in a more personalized and integrative experience. With this aim, we take a pivotal step in advancing automated dialogue systems for personalized integrative negotiation tasks. We develop DEAL, a pioneering Dialogue datasEt for personALized integrative negotiation task in the tourism domain. Further, we propose TRIP NEGOTIATOR, a novel Travel persona-aware Reinforced dIalogue generation model for Personalized iNtegrative nEGOTIATion within the tOuRism domain. TRIP NEGOTIATOR is built to discern the traveler’s persona and intent, systematically adjusts negotiation strategies, and directs the negotiation toward a pertinent phase to ensure effective negotiation. Through reinforcement learning with Proximal Policy Optimization (PPO), we guide TRIP NEGOTIATOR to generate coherent and diverse responses consistent with the traveler’s personality. Extensive qualitative and quantitative analyses demonstrate the effectiveness of TRIP NEGOTIATOR in generating personalized responses during negotiation.

Retrieval-augmented generation (RAG) offers an effective approach for addressing question answering (QA) tasks. However, the imperfections of the retrievers in RAG models often result in the retrieval of irrelevant information, which could introduce noise and degrade the performance, especially when handling multi-hop questions that require multiple steps of reasoning. To enhance the multi-hop reasoning ability of RAG models, we propose TRACE. TRACE constructs knowledge-grounded reasoning chains, which are a series of logically connected knowledge triples, to identify and integrate supporting evidence from the retrieved documents for answering questions. Specifically, TRACE employs a KG Generator to create a knowledge graph (KG) from the retrieved documents, and then uses a novel Autoregressive Reasoning Chain Constructor to build reasoning chains. Experimental results on three multi-hop QA datasets show that TRACE achieves an average performance improvement of up to 14.03% compared to using all the retrieved documents. Moreover, the results indicate that using reasoning chains as context, rather than the entire documents, is often sufficient to correctly answer questions.

Large Language Models (LLMs) have demonstrated proficiency in addressing tasks that necessitate a combination of task planning and the usage of external tools, such as weather and calculator APIs. However, real-world industrial systems present prevalent challenges in task planning and tool usage: numerous APIs in the real system make it intricate to invoke the appropriate one, while the inherent limitations of LLMs pose challenges in orchestrating an accurate sub-task sequence and API-calling order. This paper introduces a comprehensive framework aimed at enhancing the Task Planning and Tool Usage (TPTU) abilities of LLM-based agents in industry. Our framework comprises three key components designed to address these challenges: (1) the API Retriever selects the most pertinent APIs among the extensive API set; (2) the Demo Selector retrieves task-level demonstrations, which is further used for in-context learning to aid LLMs in accurately decomposing subtasks and effectively invoking hard-to-distinguish APIs; (3) LLM Finetuner tunes a base LLM to enhance its capability for task planning and API calling. We validate our methods using a real-world industry system and an open-sourced academic dataset, demonstrating the efficacy of each individual component as well as the integrated framework. The code is available at here.

Evaluating the ability of large language models (LLMs) to follow complex human-written instructions is essential for their deployment in real-world applications. While benchmarks like Chatbot Arena use human judges to assess model performance, they are resource-intensive and time-consuming. Alternative methods using LLMs as judges, such as AlpacaEval, MT Bench, WildBench, and InFoBench offer improvements but still do not capture that certain complex instruction aspects are more important than others to follow.To address this gap, we propose a novel evaluation metric, TOWER, that incorporates human-judged importance into the assessment of complex instruction following. We show that human annotators agree with tree-based representations of these complex instructions nearly as much as they agree with other human annotators. We release tree-based annotations of the InFoBench dataset and the corresponding evaluation code to facilitate future research.

Teaching language models to use tools is an important milestone towards building general assistants, but remains an open problem. While there has been significant progress on learning to use specific tools via fine-tuning, language models still struggle with learning how to robustly use new tools from only a few demonstrations. In this work we introduce a self-verification method which distinguishes between close candidates by self-asking contrastive questions during (1) tool selection; and parameter generation. We construct synthetic, high-quality, self-generated data for this goal using Llama-2 70B, which we intend to release publicly. Extensive experiments on 4 tasks from the ToolBench benchmark, consisting of 17 unseen tools, demonstrate an average improvement of 22% over few-shot baselines, even in scenarios where the distinctions between candidate tools are finely nuanced.

This paper introduces and investigates the problem of Task and Language Incremental Continual Learning (TLCL), wherein a multilingual model is systematically updated to accommodate new tasks in previously learned languages or new languages for established tasks. This significant yet previously unexplored area holds substantial practical relevance as it mirrors the dynamic requirements of real-world applications. We benchmark a representative set of continual learning (CL) algorithms for TLCL. Furthermore, we propose Task and Language-Specific Adapters (TLSA), an adapter-based parameter-efficient fine-tuning strategy. TLSA facilitates cross-lingual and cross-task transfer and outperforms other parameter-efficient fine-tuning techniques. Crucially, TLSA reduces parameter growth stemming from saving adapters to linear complexity from polynomial complexity as it was with parameter isolation-based adapter tuning. We conducted experiments on several NLP tasks arising across several languages. We observed that TLSA outperforms all other parameter-efficient approaches without requiring access to historical data for replay.

The task of text-to-table receives widespread attention, yet its importance and difficulty are underestimated. Existing works use simple datasets similar to table-to-text tasks and employ methods that ignore domain structures. As a bridge between raw text and statistical analysis, the text-to-table task often deals with complex semi-structured texts that refer to specific domain topics in the real world with entities and events, especially from those of social sciences. In this paper, we analyze the limitations of benchmark datasets and methods used in the text-to-table literature and redefine the text-to-table task to improve its compatibility with long text-processing tasks. Based on this redefinition, we propose a new dataset called CPL (Chinese Private Lending), which consists of judgments from China and is derived from a real-world legal academic project. We further propose TKGT (Text-KG-Table), a two stages domain-aware pipeline, which firstly generates domain knowledge graphs (KGs) classes semi-automatically from raw text with the mixed information extraction (Mixed-IE) method, then adopts the hybrid retrieval augmented generation (Hybird-RAG) method to transform it to tables for downstream needs under the guidance of KGs classes. Experiment results show that TKGT achieves state-of-the-art (SOTA) performance on both traditional datasets and the CPL. Our data and main code are available at https://github.com/jiangpw41/TKGT.

The objective of the research we present is to remedy the problem of the low quality of language models for low-resource languages. We introduce an algorithm, the Token Embedding Mapping Algorithm (TEMA), that maps the token embeddings of a richly pre-trained model L1 to a poorly trained model L2, thus creating a richer L2’ model. Our experiments show that the L2’ model reduces perplexity with respect to the original monolingual model L2, and that for downstream tasks, including SuperGLUE, the results are state-of-the-art or better for the most semantic tasks. The models obtained with TEMA are also competitive or better than multilingual or extended models proposed as solutions for mitigating the low-resource language problems.

Zero-shot singing voice synthesis (SVS) with style transfer and style control aims to generate high-quality singing voices with unseen timbres and styles (including singing method, emotion, rhythm, technique, and pronunciation) from audio and text prompts. However, the multifaceted nature of singing styles poses a significant challenge for effective modeling, transfer, and control. Furthermore, current SVS models often fail to generate singing voices rich in stylistic nuances for unseen singers. To address these challenges, we introduce TCSinger, the first zero-shot SVS model for style transfer across cross-lingual speech and singing styles, along with multi-level style control. Specifically, TCSinger proposes three primary modules: 1) the clustering style encoder employs a clustering vector quantization model to stably condense style information into a compact latent space; 2) the Style and Duration Language Model (S&D-LM) concurrently predicts style information and phoneme duration, which benefits both; 3) the style adaptive decoder uses a novel mel-style adaptive normalization method to generate singing voices with enhanced details. Experimental results show that TCSinger outperforms all baseline models in synthesis quality, singer similarity, and style controllability across various tasks, including zero-shot style transfer, multi-level style control, cross-lingual style transfer, and speech-to-singing style transfer.

Table reasoning tasks have shown remarkable progress with the development of large language models (LLMs), which involve interpreting and drawing conclusions from tabular data based on natural language (NL) questions. Existing solutions mainly tested on smaller tables face scalability issues and struggle with complex queries due to incomplete or dispersed data across different table sections. To alleviate these challenges, we propose TAP4LLM as a versatile pre-processor suite for leveraging LLMs in table-based tasks effectively. It covers several distinct components: (1) table sampling to decompose large tables into manageable sub-tables based on query semantics, (2) table augmentation to enhance tables with additional knowledge from external sources or models, and (3) table packing & serialization to convert tables into various formats suitable for LLMs’ understanding. In each module, we design and compare several common methods for usage in various scenarios, aiming to shed light on the best practices for leveraging LLMs for table-reasoning tasks. Our experiments show that our method improves LLMs’ reasoning capabilities in various tabular tasks and enhances the interaction between LLMs and tabular data by employing effective pre-processing.

As long-context large language models (LLMs) are attracting increasing attention for their ability to handle context windows exceeding 128k tokens, the need for effective evaluation methods for these models becomes critical.Existing evaluation methods, however, fall short: needle-in-a-haystack (NIAH) and its variants are overly simplistic, while creating realistic benchmarks is prohibitively expensive due to extensive human annotation requirements. To bridge this gap, we propose TAIL, an automatic toolkit for creating realistic evaluation benchmarks and assessing the performance of long-context LLMs.With TAIL, users can customize the building of a long-context, document-grounded QA benchmark and obtain visualized performance metrics of evaluated models.TAIL has the advantage of requiring minimal human annotation and generating natural questions based on user-provided long-context documents. We apply TAIL to construct a benchmark encompassing multiple expert domains, such as finance, law, patent, and scientific literature. We then evaluate four state-of-the-art long-context LLMs using this benchmark. Results show that all LLMs experience varyingdegrees of performance degradation as contextlengths increase.

Tokenizers are crucial for encoding information in Large Language Models, but their development has recently stagnated, and they contain inherent weaknesses. Major limitations include computational overhead, ineffective vocabulary use, and unnecessarily large embedding and head layers. Additionally, their performance is biased towards a reference corpus, leading to reduced effectiveness for underrepresented languages.To remedy these issues, we propose T-Free, which directly embeds words through sparse activation patterns over character triplets and does not require a reference corpus. T-Free inherently exploits morphological similarities and allows for strong compression of embedding layers. In our exhaustive experimental evaluation, we achieve competitive downstream performance with a parameter reduction of more than 85% on these layers. Further, T-Free shows significant improvements in cross-lingual transfer learning.

Long-context large language models (LC LLMs) promise to increase reliability of LLMs in real-world tasks requiring processing and understanding of long input documents. However, this ability of LC LLMs to reliably utilize their growing context windows remains under investigation. In this work, we evaluate the performance of state-of-the-art GPT-4 suite of LC LLMs in solving a series of progressively challenging tasks, as a function of factors such as context length, task difficulty, and position of key information by creating a real world financial news dataset. Our findings indicate that LC LLMs exhibit brittleness at longer context lengths even for simple tasks, with performance deteriorating sharply as task complexity increases. At longer context lengths, these state-of-the-art models experience catastrophic failures in instruction following resulting in degenerate outputs. Our prompt ablations also reveal unfortunate continued sensitivity to both the placement of the task instruction in the context window as well as minor markdown formatting. Finally, we advocate for more rigorous evaluation of LC LLMs by employing holistic metrics such as F1 (rather than recall) and reporting confidence intervals, thereby ensuring robust and conclusive findings.

The emergence of Large Language Models (LLMs), has opened exciting possibilities for constructing computational simulations designed to replicate human behavior accurately. Current research suggests that LLM-based agents become increasingly human-like in their performance, sparking interest in using these AI agents as substitutes for human participants in behavioral studies. However, LLMs are complex statistical learners without straightforward deductive rules, making them prone to unexpected behaviors. Hence, it is crucial to study and pinpoint the key behavioral distinctions between humans and LLM-based agents. In this study, we highlight the limitations of LLMs in simulating human interactions, particularly focusing on LLMs’ ability to simulate political debates on topics that are important aspects of people’s day-to-day lives and decision-making processes. Our findings indicate a tendency for LLM agents to conform to the model’s inherent social biases despite being directed to debate from certain political perspectives. This tendency results in behavioral patterns that seem to deviate from well-established social dynamics among humans. We reinforce these observations using an automatic self-fine-tuning method, which enables us to manipulate the biases within the LLM and demonstrate that agents subsequently align with the altered biases. These results underscore the need for further research to develop methods that help agents overcome these biases, a critical step toward creating more realistic simulations.

Multimodal Retrieval Augmented Generation (MMRAG) is a powerful approach to question-answering over multimodal documents. A key challenge with evaluating MMRAG is the paucity of high-quality datasets matching the question styles and modalities of interest. In light of this, we propose SMMQG, a synthetic data generation framework. SMMQG leverages interplay between a retriever, large language model (LLM) and large multimodal model (LMM) to generate question and answer pairs directly from multimodal documents, with the questions conforming to specified styles and modalities. We use SMMQG to generate an MMRAG dataset of 1024 questions over Wikipedia documents and evaluate state-of-the-art models using it, revealing insights into model performance that are attainable only through style- and modality-specific evaluation data. Next, we measure the quality of data produced by SMMQG via a human study. We find that the quality of SMMQG-generated synthetic data is on par with the quality of the crowdsourced benchmark MMQA and that downstream evaluation results using both datasets strongly concur.

Large language models (LLMs) are proficient in capturing factual knowledge across various domains. However, refining their capabilities on previously seen knowledge or integrating new knowledge from external sources remains a significant challenge. In this work, we propose a novel synthetic knowledge ingestion method called , which leverages fine-grained synthesis, interleaved generation, and assemble augmentation strategies to construct high-quality data representations from raw knowledge sources. We then integrate and its variations with three knowledge injection techniques: Retrieval Augmented Generation (RAG), Supervised Fine-tuning (SFT), and Continual Pre-training (CPT) to inject and refine knowledge in language models. Extensive empirical experiments are conducted on various question-answering tasks spanning finance, biomedicine, and open-generation domains to demonstrate that significantly outperforms baseline methods by facilitating effective knowledge injection. We believe that our work is an important step towards enhancing the factual accuracy of LLM outputs by refining knowledge representation and injection capabilities.

It is often desirable to distill the capabilities of large language models (LLMs) into smaller student models due to compute and memory constraints. One way to do this for classification tasks is via dataset synthesis, which can be accomplished by generating examples of each label from the LLM. Prior approaches to synthesis use few-shot prompting, which relies on the LLM’s parametric knowledge to generate usable examples. However, this leads to issues of repetition, bias towards popular entities, and stylistic differences from human text. In this work, we propose Synthesize by Retrieval and Refinement (SynthesizRR), which uses retrieval augmentation to introduce variety into the dataset synthesis process: as retrieved passages vary, the LLM is seeded with different content to generate its examples. We empirically study the synthesis of six datasets, covering topic classification, sentiment analysis, tone detection, and humor, requiring complex synthesis strategies. We find SynthesizRR greatly improves lexical and semantic diversity, similarity to human-written text, and distillation performance, when compared to 32-shot prompting and four prior approaches.

Traditional benchmarking in NLP typically involves using static, held-out test sets and calculating aggregated statistics based on diverse examples. However, this approach often results in an overestimation of performance and lacks the ability to offer comprehensive, interpretable, and dynamic assessments of NLP models. Recently, works like DynaBench and Checklist have addressed these limitations through behavioral testing of NLP models with test types generated by a multi-step human-annotated pipeline. Unfortunately, manually creating a variety of test types requires significant human labor, thus weakening efficiency. In this work, we propose SynthEval, a hybrid behavioral testing framework that leverages large language models (LLMs) to generate a wide range of test types for a comprehensive evaluation of NLP models. The SynthEval framework first generates sentences via LLMs using controlled generation, and then identifies challenging examples by comparing the predictions made by LLMs with task-specific NLP models. In the last stage, human experts investigate the challenging examples, manually design templates, and identify the types of failures the task-specific models consistently exhibit. We apply SynthEval to two classification tasks and show that our framework is effective in identifying weaknesses of strong models on these tasks.