Unsupervised multitask pre-training has been the critical method behind the recent success of language models (LMs). However, supervised multitask learning still holds significant promise, as scaling it in the post-training stage trends towards better generalization. In this paper, we explore supervised multitask pre-training by proposing Instruction Pre-training, a framework that scalably augments massive raw corpora with instruction-response pairs to pre-train LMs. The instruction-response pairs are generated by an efficient instruction synthesizer built on open-source models. In our experiments, we synthesize 200M instruction response pairs covering 40+ task categories to verify the effectiveness of Instruction Pre-training. In pre-training from scratch, Instruction Pre-training not only consistently enhances pre-trained base models but also benefits more from further instruction tuning. In continual pre-training, Instruction Pre-training enables Llama3-8B to be comparable to or even outperform Llama3-70B. Our model, code, and data are available at https://github.com/microsoft/LMOps.
The safety defense methods of Large language models (LLMs) stays limited because the dangerous prompts are manually curated to just few known attack types, which fails to keep pace with emerging varieties. Recent studies found that attaching suffixes to harmful instructions can hack the defense of LLMs and lead to dangerous outputs. However, similar to traditional text adversarial attacks, this approach, while effective, is limited by the challenge of the discrete tokens. This gradient based discrete optimization attack requires over 100,000 LLM calls, and due to the unreadable of adversarial suffixes, it can be relatively easily penetrated by common defense methods such as perplexity filters.To cope with this challenge, in this paper, we propose an Adversarial Suffix Embedding Translation Framework (ASETF), aimed at transforming continuous adversarial suffix embeddings into coherent and understandable text. This method greatly reduces the computational overhead during the attack process and helps to automatically generate multiple adversarial samples, which can be used as data to strengthen LLM’s security defense. Experimental evaluations were conducted on Llama2, Vicuna, and other prominent LLMs, employing harmful directives sourced from the Advbench dataset.The results indicate that our method significantly reduces the computation time of adversarial suffixes and achieves a much better attack success rate than existing techniques, while significantly enhancing the textual fluency of the prompts. In addition, our approach can be generalized into a broader method for generating transferable adversarial suffixes that can successfully attack multiple LLMs, even black-box LLMs, such as ChatGPT and Gemini.
In the realm of multi-intent spoken language understanding, recent advancements have leveraged the potential of prompt learning frameworks. However, critical gaps exist in these frameworks: the lack of explicit modeling of dual-task dependencies and the oversight of task-specific semantic differences among utterances. To address these shortcomings, we propose DC-Instruct, a novel generative framework based on Dual-task Inter-dependent Instructions (DII) and Supervised Contrastive Instructions (SCI). Specifically, DII guides large language models (LLMs) to generate labels for one task based on the other task’s labels, thereby explicitly capturing dual-task inter-dependencies. Moreover, SCI leverages utterance semantics differences by guiding LLMs to determine whether a pair of utterances share the same or similar labels. This can improve LLMs on extracting and discriminating task-specific semantics, thus enhancing their SLU reasoning abilities. Extensive experiments on public benchmark datasets show that DC-Instruct markedly outperforms current generative models and state-of-the-art methods, demonstrating its effectiveness in enhancing dialogue language understanding and reasoning.
Is it always necessary to compute tokens from shallow to deep layers in Transformers? The continued success of vanilla Transformers and their variants suggests an undoubted “yes”. In this work, however, we attempt to break the depth-ordered convention by proposing a novel architecture dubbed mixture-of-modules (MoM), which is motivated by an intuition that any layer, regardless of its position, can be used to compute a token as long as it possesses the needed processing capabilities. The construction of MoM starts from a finite set of modules defined by multi-head attention and feed-forward networks, each distinguished by its unique parameterization. Two routers then iteratively select attention modules and feed-forward modules from the set to process a token. The selection dynamically expands the computation graph in the forward pass of the token, culminating in an assembly of modules. We show that MoM provides not only a unified framework for Transformers and their numerous variants but also a flexible and learnable approach for reducing redundancy in Transformer parameterization. We pre-train various MoMs using OpenWebText. Empirical results demonstrate that MoMs, of different sizes, consistently outperform vanilla transformers. More interestingly, after removing 50% of the multi-head attention modules and 25% of the feed-forward modules, an MoM model still holds comparable performance. Additionally, by properly adjusting the number of modules and compressing the model depth, one can have an MoM that achieves comparable performance to GPT-2 (774M) while saving 16% TFLOPs and 42% memory usage during forward computation.
In conversational AI, large-scale models excel in various tasks but struggle with target-driven conversation planning. Current methods, such as chain-of-thought reasoning and tree-search policy learning techniques, either neglect plan rationality or require extensive human simulation procedures. Addressing this, we propose a novel two-stage framework, named EnPL, to improve the LLMs’ capability in planning conversations towards designated targets, including (1) distilling natural language plans from target-driven conversation corpus and (2) generating new plans with demonstration-guided in-context learning. Specifically, we first propose a filter approach to distill a high-quality plan dataset, ConvPlan (Resources of this paper can be found at https://github.com/pandazzh2020/ConvPlan). With the aid of corresponding conversational data and support from relevant knowledge bases, we validate the quality and rationality of these plans. Then, these plans are leveraged to help guide LLMs to further plan for new targets. Empirical results demonstrate that our method significantly improves the planning ability of LLMs, especially in target-driven conversations. Furthermore, EnPL is demonstrated to be quite effective in collecting target-driven conversation datasets and enhancing response generation, paving the way for constructing extensive target-driven conversational models.
Character-based dialogue (CharacterDial) has become essential in the industry (e.g., Character.AI), enabling users to freely customize social characters for social interactions. However, the generalizability and adaptability across various conversational scenarios inherent in customizing social characters still lack public industrial solutions. To address these challenges, by dissecting well-rounded social characters composed of both inherent social profiles and external social behaviors, we manually collect a large-scale Chinese corpus featuring characters with diverse categories and behaviors, and develop CharacterGLM models alongside well-designed refinement methods. Extensive experiments show that CharacterGLM outperforms most popular open- and closed-source LLMs and performs comparably to GPT-4. We will release our data and models for local development and deployment.
Although Large Language Models (LLMs) are becoming increasingly powerful, they still exhibit significant but subtle weaknesses, such as mistakes in instruction-following or coding tasks.As these unexpected errors could lead to severe consequences in practical deployments, it is crucial to investigate the limitations within LLMs systematically.Traditional benchmarking approaches cannot thoroughly pinpoint specific model deficiencies, while manual inspections are costly and not scalable. In this paper, we introduce a unified framework, AutoDetect, to automatically expose weaknesses in LLMs across various tasks. Inspired by the educational assessment process that measures students’ learning outcomes, AutoDetect consists of three LLM-powered agents: Examiner, Questioner, and Assessor.The collaboration among these three agents is designed to realize comprehensive and in-depth weakness identification. Our framework demonstrates significant success in uncovering flaws, with an identification success rate exceeding 30% in prominent models such as ChatGPT and Claude.More importantly, these identified weaknesses can guide specific model improvements, proving more effective than untargeted data augmentation methods like Self-Instruct. Our approach has led to substantial enhancements in popular LLMs, including the Llama series and Mistral-7b, boosting their performance by over 10% across several benchmarks.Code and data are publicly available at https://github.com/thu-coai/AutoDetect.
The safety of Large Language Models (LLMs) has gained increasing attention in recent years, but there still lacks a comprehensive approach for detecting safety issues within LLMs’ responses in an aligned, customizable and explainable manner. In this paper, we propose ShieldLM, an LLM-based safety detector, which aligns with common safety standards, supports customizable detection rules, and provides explanations for its decisions. To train ShieldLM, we compile a large bilingual dataset comprising 14,387 query-response pairs, annotating the safety of responses based on various safety standards. Through extensive experiments, we demonstrate that ShieldLM surpasses strong baselines across four test sets, showcasing remarkable customizability and explainability. Besides performing well on standard detection datasets, ShieldLM has also been shown to be effective as a safety evaluator for advanced LLMs. ShieldLM is released at https://github.com/thu-coai/ShieldLM to support accurate and explainable safety detection under various safety standards.
Recent progress in natural language processing (NLP) owes much to remarkable advances in large language models (LLMs). Nevertheless, LLMs frequently “hallucinate,” resulting in non-factual outputs. Our carefully-designed human evaluation substantiates the serious hallucination issue, revealing that even GPT-3.5 produces factual outputs less than 25% of the time. This underscores the importance of fact verifiers in order to measure and incentivize progress. Our systematic investigation affirms that LLMs can be repurposed as effective fact verifiers with strong correlations with human judgments. Surprisingly, FLAN-T5-11B , the least factual generator in our study, performs the best as a fact verifier, even outperforming more capable LLMs like GPT3.5 and ChatGPT. Delving deeper, we analyze the reliance of these LLMs on high-quality evidence, as well as their deficiencies in robustness and generalization ability. Our study presents insights for developing trustworthy generation models.
Depression is a widespread mental health disorder affecting millions globally. Clinical interviews are the gold standard for assessing depression, but they heavily rely on scarce professional clinicians, highlighting the need for automated detection systems. However, existing methods only capture part of the relevant elements in clinical interviews, unable to incorporate all depressive cues. Moreover, the scarcity of participant data, due to privacy concerns and collection challenges, intrinsically constrains interview modeling. To address these limitations, in this paper, we propose a structural element graph (SEGA), which transforms the clinical interview into an expertise-inspired directed acyclic graph for comprehensive modeling. Additionally, we further empower SEGA by devising novel principle-guided data augmentation with large language models (LLMs) to supplement high-quality synthetic data and enable graph contrastive learning. Extensive evaluations on two real-world clinical datasets, in both English and Chinese, show that SEGA significantly outperforms baseline methods and powerful LLMs like GPT-3.5 and GPT-4.
Large language models (LLMs) have shown impressive success in various applications. However, these models are often not well aligned with human intents, which calls for additional treatments on them; that is, the alignment problem. To make LLMs better follow user instructions, existing alignment methods primarily focus on further training them. However, the extra training of LLMs is usually expensive in terms of GPU computing; even worse, some LLMs are not accessible for user-demanded training, such as GPTs. In this work, we take a different perspective—Black-Box Prompt Optimization (BPO)—to perform alignments. The idea is to optimize user prompts to suit LLMs’ input understanding, so as to best realize users’ intents without updating LLMs’ parameters. BPO leverages human preferences to optimize prompts, thus making it superior to LLM (e.g., ChatGPT) as a prompt engineer. Moreover, BPO is model-agnostic, and the empirical results demonstrate that the BPO-aligned ChatGPT yields a 22% increase in the win rate against its original version and 10% for GPT-4. Notably, the BPO-aligned LLMs can outperform the same models aligned by PPO and DPO, and it also brings additional performance gains when combining BPO with PPO or DPO. Code and datasets are released at https://github.com/thu-coai/BPO.
Recent advances in Large Language Models (LLMs) have highlighted the need for robust, comprehensive, and challenging benchmarks. Yet, research on evaluating their Emotional Intelligence (EI) is considerably limited. Existing benchmarks have two major shortcomings: first, they mainly focus on emotion recognition, neglecting essential EI capabilities such as emotion management and thought facilitation through emotion understanding; second, they are primarily constructed from existing datasets, which include frequent patterns, explicit information, and annotation errors, leading to unreliable evaluation. We propose EmoBench, a benchmark that draws upon established psychological theories and proposes a comprehensive definition for machine EI, including Emotional Understanding and Emotional Application. EmoBench includes a set of 400 hand-crafted questions in English and Chinese, which are meticulously designed to require thorough reasoning and understanding. Our findings reveal a considerable gap between the EI of existing LLMs and the average human, highlighting a promising direction for future research. Our code and data are publicly available at https://github.com/Sahandfer/EmoBench.
While significant attention has been dedicated to exploiting weaknesses in LLMs through jailbreaking attacks, there remains a paucity of effort in defending against these attacks. We point out a pivotal factor contributing to the success of jailbreaks: the intrinsic conflict between the goals of being helpful and ensuring safety. Accordingly, we propose to integrate goal prioritization at both training and inference stages to counteract. Implementing goal prioritization during inference substantially diminishes the Attack Success Rate (ASR) of jailbreaking from 66.4% to 3.6% for ChatGPT. And integrating goal prioritization into model training reduces the ASR from 71.0% to 6.6% for Llama2-13B. Remarkably, even in scenarios where no jailbreaking samples are included during training, our approach slashes the ASR by half. Additionally, our findings reveal that while stronger LLMs face greater safety risks, they also possess a greater capacity to be steered towards defending against such attacks, both because of their stronger ability in instruction following. Our work thus contributes to the comprehension of jailbreaking attacks and defenses, and sheds light on the relationship between LLMs’ capability and safety. Our code is available at https://github.com/thu-coai/JailbreakDefense_GoalPriority.
Alignment has become a critical step for instruction-tuned Large Language Models (LLMs) to become helpful assistants. However, effective evaluation of alignment for emerging Chinese LLMs is still significantly lacking, calling for real-scenario grounded, open-ended, challenging and automatic evaluations tailored for alignment. To fill in this gap, we introduce AlignBench, a comprehensive multi-dimensional benchmark for evaluating LLMs’ alignment in Chinese. We tailor a human-in-the-loop data curation pipeline, containing 8 main categories, 683 real-scenario rooted queries and corresponding human verified references.To ensure references’ correctness, each knowledge-intensive query is accompanied with evidences collected from reliable webpages (including the url and quotation) by our annotators.For automatic evaluation, our benchmark employs a rule-calibrated multi-dimensional LLM-as-Judge (CITATION) with Chain-of-Thought to generate explanations and final ratings as evaluations, ensuring high reliability and interpretability.All evaluation codes and data are publicly available at https://github.com/THUDM/AlignBench
When using language models (LMs) to solve complex problems, humans might struggle to understand the LM-generated solutions and repair the flawed ones. To assist humans in repairing them, we propose to automatically decompose complex solutions into multiple simpler pieces that correspond to specific subtasks. We introduce a novel objective for learning task decomposition, termed assistive value (AssistV), which measures the feasibility and speed for humans to repair the decomposed solution. We collect a dataset of human repair experiences on different decomposed solutions. Utilizing the collected data as in-context examples, we then learn to critique, refine, and rank decomposed solutions to improve AssistV. We validate our method under competitive programming problems: under 177 hours of human study, our method enables non-experts to solve 33.3% more problems, speeds them up by 3.3x, and empowers them to match unassisted experts.
Since the natural language processing (NLP) community started to make large language models (LLMs) act as a critic to evaluate the quality of generated texts, most of the existing works train a critique generation model on the evaluation data labeled by GPT-4’s direct prompting. We observe that these models lack the ability to generate informative critiques in both pointwise grading and pairwise comparison especially without references. As a result, their generated critiques cannot provide fine-grained distinguishability on generated texts, causing unsatisfactory evaluation performance. In this paper, we propose a simple yet effective method called Eval-Instruct, which can first acquire pointwise grading critiques with pseudo references and then revise these critiques via multi-path prompting to obtain informative evaluation data in different tasks and settings, including pointwise grading and pairwise comparison with / without references. After fine-tuning on these data, the resulting model CritiqueLLM is empirically shown to outperform ChatGPT and all the open-source baselines and even achieve comparable evaluation performance to GPT-4 in system-level correlations of pointwise grading. We also demonstrate that our generated critiques can act as scalable feedback to further improve the generation quality of strong LLMs like ChatGPT.
With the rapid development of Large Language Models (LLMs), increasing attention has been paid to their safety concerns. Consequently, evaluating the safety of LLMs has become an essential task for facilitating the broad applications of LLMs. Nevertheless, the absence of comprehensive safety evaluation benchmarks poses a significant impediment to effectively assess and enhance the safety of LLMs. In this work, we present SafetyBench, a comprehensive benchmark for evaluating the safety of LLMs, which comprises 11,435 diverse multiple choice questions spanning across 7 distinct categories of safety concerns. Notably, SafetyBench also incorporates both Chinese and English data, facilitating the evaluation in both languages. Our extensive tests over 25 popular Chinese and English LLMs in both zero-shot and few-shot settings reveal a substantial performance advantage for GPT-4 over its counterparts, and there is still significant room for improving the safety of current LLMs. We also demonstrate that the measured safety understanding abilities in SafetyBench are correlated with safety generation abilities. Data and evaluation guidelines are available at https://github.com/thu-coai/SafetyBench. Submission entrance and leaderboard are available at https://llmbench.ai/safety.
Theory of Mind (ToM) is the cognitive capability to perceive and ascribe mental states to oneself and others. Recent research has sparked a debate over whether large language models (LLMs) exhibit a form of ToM. However, existing ToM evaluations are hindered by challenges such as constrained scope, subjective judgment, and unintended contamination, yielding inadequate assessments. To address this gap, we introduce ToMBench with three key characteristics: a systematic evaluation framework encompassing 8 tasks and 31 abilities in social cognition, a multiple-choice question format to support automated and unbiased evaluation, and a build-from-scratch bilingual inventory to strictly avoid data leakage. Based on ToMBench, we conduct extensive experiments to evaluate the ToM performance of 10 popular LLMs across tasks and abilities. We find that even the most advanced LLMs like GPT-4 lag behind human performance by over 10% points, indicating that LLMs have not achieved a human-level theory of mind yet. Our aim with ToMBench is to enable an efficient and effective evaluation of LLMs’ ToM capabilities, thereby facilitating the development of LLMs with inherent social intelligence.
Theory of mind (ToM) refers to humans’ ability to understand and infer the desires, beliefs, and intentions of others. The acquisition of ToM plays a key role in humans’ social cognition and interpersonal relations. Though indispensable for social intelligence, ToM is still lacking for modern AI and NLP systems since they cannot access the human mental state and cognitive process beneath the training corpus. To empower AI systems with the ToM ability and narrow the gap between them and humans, in this paper, we propose COKE: the first cognitive knowledge graph for machine theory of mind. Specifically, COKE formalizes ToM as a collection of 45k+ manually verified cognitive chains that characterize human mental activities and subsequent behavioral/affective responses when facing specific social circumstances. In addition, we further generalize COKE using LLMs and build a powerful generation model COLM tailored for cognitive reasoning. Experimental results in both automatic and human evaluation demonstrate the high quality of COKE, the superior ToM ability of COLM, and its potential to significantly enhance social applications.
Counterspeech is an effective way to combat online hate speech. Considering the multifaceted nature of online hate speech, counterspeech with varying intents (e.g., denouncing or empathy) has significant potential to mitigate hate speech effectively. Recently, controlled approaches based on large language models (LLMs) have been explored to generate intent-specific counterspeech. Due to the lack of attention to intent-specific information by LLMs during the decoding process, those methods cater more to the semantic information rather than matching with the desired intents. Further, there are still limitations in quantitatively evaluating the effectiveness of counterspeech with different intents in mitigating hate speech. In this paper, to address the above issues, we propose DART, an LLMs-based DuAl-discRiminaTor guided framework for counterspeech generation. We employ an intent-aware discriminator and hate-mitigating discriminator to jointly guide the decoding preferences of LLMs, which facilitates the model towards generating counterspeech catering to specific intent and hate mitigation. We apply a maximum-margin relative objective for training discriminators. This objective leverages the distance between counterspeech aligned with the desired target (such as specific intent or effectiveness in hate mitigation) and undesired as an effective learning signal. Extensive experiments show that DART achieves excellent performances in matching the desired intent and mitigating hate.
Emotional support conversation (ESC) aims to provide emotional support (ES) to improve one’s mental state. Existing works stay at fitting grounded responses and responding strategies (e.g., question), which ignore the effect on ES and lack explicit goals to guide emotional positive transition. To this end, we introduce a new paradigm to formalize multi-turn ESC as a process of positive emotion elicitation. Addressing this task requires finely adjusting the elicitation intensity in ES as the conversation progresses while maintaining conversational goals like coherence. In this paper, we propose Supporter, a mixture-of-expert-based reinforcement learning model, and well design ES and dialogue coherence rewards to guide policy’s learning for responding. Experiments verify the superiority of Supporter in achieving positive emotion elicitation during responding while maintaining conversational goals including coherence.
Morality in dialogue systems has raised great attention in research recently. A moral dialogue system aligned with users’ values could enhance conversation engagement and user connections. In this paper, we propose a framework, MoralDial to train and evaluate moral dialogue systems. In our framework, we first explore the communication mechanisms of morality and resolve expressed morality into three parts, which indicate the roadmap for building a moral dialogue system. Based on that, we design a simple yet effective method: constructing moral discussions between simulated specific users and the dialogue system. The constructed discussions consist of expressing, explaining, revising, and inferring moral views in dialogue exchanges, which makes conversational models learn morality well in a natural manner. Furthermore, we propose a novel evaluation method under the framework. We evaluate the multiple aspects of morality by judging the relation between dialogue responses and human values in discussions, where the multifaceted nature of morality is particularly considered. Automatic and manual experiments demonstrate that our framework is promising to train and evaluate moral dialogue systems.
In-context learning, where pre-trained language models learn to perform tasks from task examples and instructions in their contexts, has attracted much attention in the NLP community. However, the ability of in-context learning is not fully exploited because language models are not explicitly trained to learn in context. To this end, we propose PICL (Pre-training for In-Context Learning), a framework to enhance the language models’ in-context learning ability by pre-training the model on a large collection of “intrinsic tasks” in the general plain-text corpus using the simple language modeling objective. PICL encourages the model to infer and perform tasks by conditioning on the contexts while maintaining task generalization of pre-trained models. We evaluate the in-context learning performance of the model trained with PICL on seven widely-used text classification datasets and the Super-NaturalInstrctions benchmark, which contains 100+ NLP tasks formulated to text generation. Our experiments show that PICL is more effective and task-generalizable than a range of baselines, outperforming larger language models with nearly 4x parameters. The code is publicly available at https://github.com/thu-coai/PICL.
Empathetic conversation is psychologically supposed to be the result of conscious alignment and interaction between the cognition and affection of empathy. However, existing empathetic dialogue models usually consider only the affective aspect or treat cognition and affection in isolation, which limits the capability of empathetic response generation. In this work, we propose the CASE model for empathetic dialogue generation. It first builds upon a commonsense cognition graph and an emotional concept graph and then aligns the user’s cognition and affection at both the coarse-grained and fine-grained levels. Through automatic and manual evaluation, we demonstrate that CASE outperforms state-of-the-art baselines of empathetic dialogues and can generate more empathetic and informative responses.
Existing evaluation metrics for natural language generation (NLG) tasks face the challenges on generalization ability and interpretability. Specifically, most of the well-performed metrics are required to train on evaluation datasets of specific NLG tasks and evaluation dimensions, which may cause over-fitting to task-specific datasets. Furthermore, existing metrics only provide an evaluation score for each dimension without revealing the evidence to interpret how this score is obtained. To deal with these challenges, we propose a simple yet effective metric called DecompEval. This metric formulates NLG evaluation as an instruction-style question answering task and utilizes instruction-tuned pre-trained language models (PLMs) without training on evaluation datasets, aiming to enhance the generalization ability. To make the evaluation process more interpretable, we decompose our devised instruction-style question about the quality of generated texts into the subquestions that measure the quality of each sentence. The subquestions with their answers generated by PLMs are then recomposed as evidence to obtain the evaluation result. Experimental results show that DecompEval achieves state-of-the-art performance in untrained metrics for evaluating text summarization and dialogue generation, which also exhibits strong dimension-level / task-level generalization ability and interpretability.
Large pre-trained language models achieve impressive results across many tasks. However, recent works point out that pre-trained language models may memorize a considerable fraction of their training data, leading to the privacy risk of information leakage. In this paper, we propose a method named Ethicist for targeted training data extraction through loss smoothed soft prompting and calibrated confidence estimation, investigating how to recover the suffix in the training data when given a prefix. To elicit memorization in the attacked model, we tune soft prompt embeddings while keeping the model fixed. We further propose a smoothing loss that smooths the loss distribution of the suffix tokens to make it easier to sample the correct suffix. In order to select the most probable suffix from a collection of sampled suffixes and estimate the prediction confidence, we propose a calibrated confidence estimation method, which normalizes the confidence of the generated suffixes with a local estimation. We show that Ethicist significantly improves the extraction performance on a recently proposed public benchmark. We also investigate several factors influencing the data extraction performance, including decoding strategy, model scale, prefix length, and suffix length. Our code is availabel at https://github.com/thu-coai/Targeted-Data-Extraction.
Non-parallel text style transfer is an important task in natural language generation. However, previous studies concentrate on the token or sentence level, such as sentence sentiment and formality transfer, but neglect long style transfer at the discourse level. Long texts usually involve more complicated author linguistic preferences such as discourse structures than sentences. In this paper, we formulate the task of non-parallel story author-style transfer, which requires transferring an input story into a specified author style while maintaining source semantics. To tackle this problem, we propose a generation model, named StoryTrans, which leverages discourse representations to capture source content information and transfer them to target styles with learnable style embeddings. We use an additional training objective to disentangle stylistic features from the learned discourse representation to prevent the model from degenerating to an auto-encoder. Moreover, to enhance content preservation, we design a mask-and-fill framework to explicitly fuse style-specific keywords of source texts into generation. Furthermore, we constructed new datasets for this task in Chinese and English, respectively. Extensive experiments show that our model outperforms strong baselines in overall performance of style transfer and content preservation.
Teachers often guide students to improve their essays by adding engaging modifiers to polish the sentences. In this work, we present the first study on automatic sentence polishing by adding modifiers. Since there is no available dataset for the new task, we first automatically construct a large number of parallel data by removing modifiers in the engaging sentences collected from public resources. Then we fine-tune LongLM to reconstruct the original sentences from the corrupted ones. Considering that much overlap between inputs and outputs may bias the model to completely copy the inputs, we split each source sentence into sub-sentences and only require the model to generate the modified sub-sentences. Furthermore, we design a retrieval augmentation algorithm to prompt the model to add suitable modifiers. Automatic and manual evaluation on the auto-constructed test set and real human texts show that our model can generate more engaging sentences with suitable modifiers than strong baselines while keeping fluency. We deploy the model at http://coai.cs.tsinghua.edu.cn/static/polishSent/. A demo video is available at https://youtu.be/Y6gFHOgSv8Y.
Conversational systems are envisioned to provide social support or functional service to human users via natural language interactions. Conventional conversation researches mainly focus on the responseability of the system, such as dialogue context understanding and response generation, but overlooks the design of an essential property in intelligent conversations, i.e., goal awareness. The awareness of goals means the state of not only being responsive to the users but also aware of the target conversational goal and capable of leading the conversation towards the goal, which is a significant step towards higher-level intelligence and artificial consciousness. It can not only largely improve user engagement and service efficiency in the conversation, but also empower the system to handle more complicated conversation tasks that involve strategical and motivational interactions. In this tutorial, we will introduce the recent advances on the design of agent’s awareness of goals in a wide range of conversational systems.
Due to the lack of human resources for mental health support, there is an increasing demand for employing conversational agents for support. Recent work has demonstrated the effectiveness of dialogue models in providing emotional support. As previous studies have demonstrated that seekers’ persona is an important factor for effective support, we investigate whether there are benefits to modeling such information in dialogue models for support. In this paper, our empirical analysis verifies that persona has an important impact on emotional support. Therefore, we propose a framework for dynamically inferring and modeling seekers’ persona. We first train a model for inferring the seeker’s persona from the conversation history. Accordingly, we propose PAL, a model that leverages persona information and, in conjunction with our strategy-based controllable generation method, provides personalized emotional support. Automatic and manual evaluations demonstrate that PAL achieves state-of-the-art results, outperforming the baselines on the studied benchmark. Our code and data are publicly available at https://github.com/chengjl19/PAL.
It has always been an important yet challenging problem to control language models to avoid generating texts with undesirable attributes, such as toxic language and unnatural repetition. We introduce Leo for controllable text generation, which needs no modification to the model architecture and facilitates out-of-the-box use of trained models. It employs a contrastive loss on sequence likelihood, which fundamentally decreases the generation probability of negative samples (i.e., generations with undesirable attributes). It also adopts a novel likelihood ranking-based strategy to construct contrastive samples from model generations. On the tasks of language detoxification, sentiment steering, and repetition reduction, we show that Leo outperforms strong baselines of controllable text generation and demonstrate the superiority of Leo’s sample construction strategy.
Crowdsourced dialogue corpora are usually limited in scale and topic coverage due to the expensive cost of data curation. This would hinder the generalization of downstream dialogue models to open-domain topics. In this work, we leverage large language models for dialogue augmentation in the task of emotional support conversation (ESC). By treating dialogue augmentation as a dialogue completion task, we prompt a fine-tuned language model to complete full dialogues from available dialogue posts of various topics, which are then postprocessed based on heuristics. Applying this approach, we construct AugESC, an augmented dataset for the ESC task, which largely extends the scale and topic coverage of the crowdsourced ESConv corpus. Through comprehensive human evaluation, we demonstrate that our approach is superior to strong baselines of dialogue augmentation and that AugESC has comparable dialogue quality to the crowdsourced corpus. We also conduct human interactive evaluation and prove that post-training on AugESC improves downstream dialogue models’ generalization ability to open-domain topics. These results suggest the utility of AugESC and highlight the potential of large language models in improving data-scarce dialogue generation tasks.
Most named entity recognition (NER) systems focus on improving model performance, ignoring the need to quantify model uncertainty, which is critical to the reliability of NER systems in open environments. Evidential deep learning (EDL) has recently been proposed as a promising solution to explicitly model predictive uncertainty for classification tasks. However, directly applying EDL to NER applications faces two challenges, i.e., the problems of sparse entities and OOV/OOD entities in NER tasks. To address these challenges, we propose a trustworthy NER framework named E-NER by introducing two uncertainty-guided loss terms to the conventional EDL, along with a series of uncertainty-guided training strategies. Experiments show that E-NER can be applied to multiple NER paradigms to obtain accurate uncertainty estimation. Furthermore, compared to state-of-the-art baselines, the proposed method achieves a better OOV/OOD detection performance and better generalization ability on OOV entities.
Despite the huge progress in myriad generation tasks, pretrained language models (LMs) such as GPT2 still tend to generate repetitive texts with maximization-based decoding algorithms for open-ended generation. We attribute their overestimation of token-level repetition probabilities to the learning bias: LMs capture simple repetitive patterns faster with the MLE loss. We propose self-contrastive training to penalize the output of a premature checkpoint of the same model when it incorrectly predicts repetition, which is shown to mitigate repetition effectively while maintaining fluency on two datasets. Furthermore, we find that LMs use longer-range dependencies to predict repetitive tokens than non-repetitive ones, which may be the cause of sentence-level repetition loops.
Recently, aspect sentiment quad prediction has received widespread attention in the field of aspect-based sentiment analysis. Existing studies extract quadruplets via pre-trained generative language models to paraphrase the original sentence into a templated target sequence. However, previous works only focus on what to generate but ignore what not to generate. We argue that considering the negative samples also leads to potential benefits. In this work, we propose a template-agnostic method to control the token-level generation, which boosts original learning and reduces mistakes simultaneously. Specifically, we introduce Monte Carlo dropout to understand the built-in uncertainty of pre-trained language models, acquiring the noises and errors. We further propose marginalized unlikelihood learning to suppress the uncertainty-aware mistake tokens. Finally, we introduce minimization entropy to balance the effects of marginalized unlikelihood learning. Extensive experiments on four public datasets demonstrate the effectiveness of our approach on various generation templates.
Retrieval-augmented generation has raise extensive attention as it is promising to address the limitations of large language models including outdated knowledge and hallucinations. However, retrievers struggle to capture relevance, especially for queries with complex information needs. Recent work has proposed to improve relevance modeling by having large language models actively involved in retrieval, i.e., to guide retrieval with generation. In this paper, we show that strong performance can be achieved by a method we call Iter-RetGen, which synergizes retrieval and generation in an iterative manner: a model’s response to a task input shows what might be needed to finish the task, and thus can serve as an informative context for retrieving more relevant knowledge which in turn helps generate a better response in another iteration. Compared with recent work which interleaves retrieval with generation when completing a single output, Iter-RetGen processes all retrieved knowledge as a whole and largely preserves the flexibility in generation without structural constraints. We evaluate Iter-RetGen on multi-hop question answering, fact verification, and commonsense reasoning, and show that it can flexibly leverage parametric knowledge and non-parametric knowledge, and is superior to or competitive with state-of-the-art retrieval-augmented baselines while causing fewer overheads of retrieval and generation. We can further improve performance via generation-augmented retrieval adaptation.
Safety detection has been an increasingly important topic in recent years and it has become even more necessary to develop reliable safety detection systems with the rapid development of large language models. However, currently available safety detection systems have limitations in terms of their versatility and interpretability. In this paper, we first introduce InstructSafety, a safety detection framework that unifies 7 common sub-tasks for safety detection. These tasks are unified into a similar form through different instructions. We then conduct a comprehensive survey of existing safety detection datasets and process 39 human-annotated datasets for instruction tuning. We also construct adversarial samples to enhance the model’s robustness. After fine-tuning Flan-T5 on the collected data, we have developed Safety-Flan-T5, a multidimensional and explainable safety detector. We conduct comprehensive experiments on a variety of datasets and tasks, and demonstrate the strong performance of Safety-Flan-T5 in comparison to supervised baselines and served APIs (Perspective API, ChatGPT and InstructGPT). We will release the processed data, fine-tuned Safety-Flan-T5 and related code for public use.
The open-endedness of large language models (LLMs) combined with their impressive capabilities may lead to new safety issues when being exploited for malicious use. While recent studies primarily focus on probing toxic outputs that can be easily detected with existing toxicity classifiers, we show that LLMs can generate diverse implicit toxic outputs that are exceptionally difficult to detect via simply zero-shot prompting. Moreover, we propose a reinforcement learning (RL) based attacking method to further induce the implicit toxicity in LLMs. Specifically, we optimize the language model with a reward that prefers implicit toxic outputs to explicit toxic and non-toxic ones. Experiments on five widely-adopted toxicity classifiers demonstrate that the attack success rate can be significantly improved through RL fine-tuning. For instance, the RL-finetuned LLaMA-13B model achieves an attack success rate of 90.04% on BAD and 62.85% on Davinci003. Our findings suggest that LLMs pose a significant threat in generating undetectable implicit toxic outputs. We further show that fine-tuning toxicity classifiers on the annotated examples from our attacking method can effectively enhance their ability to detect LLM-generated implicit toxic language.
Pre-training on large-scale open-domain dialogue data can substantially improve the performance of dialogue models. However, the pre-trained dialogue model’s ability to utilize long-range context is limited due to the scarcity of long-turn dialogue sessions. Most dialogues in existing pre-training corpora contain fewer than three turns of dialogue. To alleviate this issue, we propose the Retrieve, Reorganize and Rescale framework (Re3Dial), which can automatically construct billion-scale long-turn dialogues by reorganizing existing short-turn ones. Given a short-turn session, Re3Dial first employs a session retriever to retrieve coherent consecutive sessions. To this end, we train the retriever to capture semantic and discourse relations within multi-turn dialogues through contrastive training. Next, Re3Dial samples a session from retrieved results following a diversity sampling strategy, which is designed to penalize repetitive or generic sessions. A longer session is then derived by concatenating the original session and the sampled session. By repeating the above process, Re3Dial can yield a coherent long-turn dialogue. Extensive experiments on multiple multi-turn dialogue benchmarks demonstrate that Re3Dial significantly improves the dialogue model’s ability to utilize long-range context and thus generate more sensible and informative responses. Finally, we build a toolkit for efficiently rescaling conversations with Re3Dial, which enables us to construct a corpus containing 1B Chinese dialogue sessions with 11.3 turns on average (5X longer than the original corpus). We will release our retriever model, toolkit, and data for public use.
Dialogue comprehension and generation are vital to the success of open-domain dialogue systems. Although pre-trained generative conversation models have made significant progress in generating fluent responses, people have difficulty judging whether they understand and efficiently model the contextual information of the conversation. In this study, we propose a Multi-Source Probing (MSP) method to probe the dialogue comprehension abilities of open-domain dialogue models. MSP aggregates features from multiple sources to accomplish diverse task goals and conducts downstream tasks in a generative manner that is consistent with dialogue model pre-training to leverage model capabilities. We conduct probing experiments on seven tasks that require various dialogue comprehension skills, based on the internal representations encoded by dialogue models. Experimental results show that open-domain dialogue models can encode semantic information in the intermediate hidden states, which facilitates dialogue comprehension tasks. Models of different scales and structures possess different conversational understanding capabilities. Our findings encourage a comprehensive evaluation and design of open-domain dialogue models.
We propose task-adaptive tokenization as a way to adapt the generation pipeline to the specifics of a downstream task and enhance long-form generation in mental health. Inspired by insights from cognitive science, our task-adaptive tokenizer samples variable segmentations from multiple outcomes, with sampling probabilities optimized based on task-specific data. We introduce a strategy for building a specialized vocabulary and introduce a vocabulary merging protocol that allows for the integration of task-specific tokens into the pre-trained model’s tokenization step. Through extensive experiments on psychological question-answering tasks in both Chinese and English, we find that our task-adaptive tokenization approach brings a significant improvement in generation performance while using up to 60% fewer tokens. Preliminary experiments point to promising results when using our tokenization approach with very large language models.
Existing multi-domain dialog state tracking (DST) models are developed based on multi-domain dialogs, which require significant manual effort to define domain relations and collect data. This process can be challenging and expensive, particularly when numerous domains are involved. In this paper, we propose a divide-and-conquer (DAC) DST paradigm and a multi-domain dialog synthesis framework, which makes building multi-domain DST models from single-domain dialogs possible. The DAC paradigm segments a multi-domain dialog into multiple single-domain dialogs for DST, which makes models generalize better on dialogs involving unseen domain combinations. The multi-domain dialog synthesis framework merges several potentially related single-domain dialogs into one multi-domain dialog and modifies the dialog to simulate domain relations. The synthesized dialogs can help DST models capture the value transfer between domains. Experiments with three representative DST models on two datasets demonstrate the effectiveness of our proposed DAC paradigm and data synthesis framework.
Task-oriented dialogue (TOD) systems function as digital assistants, guiding users through various tasks such as booking flights or finding restaurants. Existing toolkits for building TOD systems often fall short in delivering comprehensive arrays of data, model, and experimental environments with a user-friendly experience. We introduce ConvLab-3: a multifaceted dialogue system toolkit crafted to bridge this gap. Our unified data format simplifies the integration of diverse datasets and models, significantly reducing complexity and cost for studying generalization and transfer. Enhanced with robust reinforcement learning (RL) tools, featuring a streamlined training process, in-depth evaluation tools, and a selection of user simulators, ConvLab-3 supports the rapid development and evaluation of robust dialogue policies. Through an extensive study, we demonstrate the efficacy of transfer learning and RL and showcase that ConvLab-3 is not only a powerful tool for seasoned researchers but also an accessible platform for newcomers.
Semantic parsing maps natural language questions into logical forms, which can be executed against a knowledge base for answers. In real-world applications, the performance of a parser is often limited by the lack of training data. To facilitate zero-shot learning, data synthesis has been widely studied to automatically generate paired questions and logical forms. However, data synthesis methods can hardly cover the diverse structures in natural languages, leading to a large gap in sentence structure between synthetic and natural questions. In this paper, we propose a decomposition-based method to unify the sentence structures of questions, which benefits the generalization to natural questions. Experiments demonstrate that our method significantly improves the semantic parser trained on synthetic data (+7.9% on KQA and +8.9% on ComplexWebQuestions in terms of exact match accuracy). Extensive analysis demonstrates that our method can better generalize to natural questions with novel text expressions compared with baselines. Besides semantic parsing, our idea potentially benefits other semantic understanding tasks by mitigating the distracting structure features. To illustrate this, we extend our method to the task of sentence embedding learning, and observe substantial improvements on sentence retrieval (+13.1% for Hit@1).
Non-AutoRegressive (NAR) text generation models have drawn much attention because of their significantly faster decoding speed and good generation quality in machine translation. However, in a wider range of text generation tasks, existing NAR models lack proper pre-training, making them still far behind the pre-trained autoregressive models. In this paper, we propose Pre-trained Directed Acyclic Transformer (PreDAT) and a novel pre-training task to promote prediction consistency in NAR generation. Experiments on five text generation tasks show that our PreDAT remarkably outperforms existing pre-trained NAR models (+4.2 score on average) and even achieves better results than pre-trained autoregressive baselines in n-gram-based metrics, along with 17 times speedup in throughput. Further analysis shows that PreDAT benefits from the unbiased prediction order that alleviates the error accumulation problem in autoregressive generation, which provides new insights into the advantages of NAR generation.1
Endowing the protagonist with a specific personality is essential for writing an engaging story. In this paper, we aim to control the protagonist’s persona in story generation, i.e., generating a story from a leading context and a persona description, where the protagonist should exhibit the specified personality through a coherent event sequence. Considering that personas are usually embodied implicitly and sparsely in stories, we propose a planning-based generation model named ConPer to explicitly model the relationship between personas and events. ConPer first plans events of the protagonist’s behavior which are motivated by the specified persona through predicting one target sentence, then plans the plot as a sequence of keywords with the guidance of the predicted persona-related events and commonsense knowledge, and finally generates the whole story. Both automatic and manual evaluation results demonstrate that ConPer outperforms state-of-the-art baselines for generating more coherent and persona-controllable stories. Our code is available at https://github.com/thu-coai/ConPer.
Teaching morals is one of the most important purposes of storytelling. An essential ability for understanding and writing moral stories is bridging story plots and implied morals. Its challenges mainly lie in: (1) grasping knowledge about abstract concepts in morals, (2) capturing inter-event discourse relations in stories, and (3) aligning value preferences of stories and morals concerning good or bad behavior. In this paper, we propose two understanding tasks and two generation tasks to assess these abilities of machines. We present STORAL, a new dataset of Chinese and English human-written moral stories. We show the difficulty of the proposed tasks by testing various models with automatic and manual evaluation on STORAL. Furthermore, we present a retrieval-augmented algorithm that effectively exploits related concepts or events in training sets as additional guidance to improve performance on these tasks.
Although Transformers with fully connected self-attentions are powerful to model long-term dependencies, they are struggling to scale to long texts with thousands of words in language modeling. One of the solutions is to equip the model with a recurrence memory. However, existing approaches directly reuse hidden states from the previous segment that encodes contexts in a uni-directional way. As a result, this prohibits the memory to dynamically interact with the current context that provides up-to-date information for token prediction. To remedy this issue, we propose Look-Ahead Memory (LaMemo) that enhances the recurrence memory by incrementally attending to the right-side tokens and interpolating with the old memory states to maintain long-term information in the history. LaMemo embraces bi-directional attention and segment recurrence with an additional computation overhead only linearly proportional to the memory length. Experiments on widely used language modeling benchmarks demonstrate its superiority over the baselines equipped with different types of memory mechanisms.
A desirable dialog system should be able to continually learn new skills without forgetting old ones, and thereby adapt to new domains or tasks in its life cycle. However, continually training a model often leads to a well-known catastrophic forgetting issue. In this paper, we present Continual Prompt Tuning, a parameter-efficient framework that not only avoids forgetting but also enables knowledge transfer between tasks. To avoid forgetting, we only learn and store a few prompt tokens’ embeddings for each task while freezing the backbone pre-trained model. To achieve bi-directional knowledge transfer among tasks, we propose several techniques (continual prompt initialization, query fusion, and memory replay) to transfer knowledge from preceding tasks and a memory-guided technique to transfer knowledge from subsequent tasks. Extensive experiments demonstrate the effectiveness and efficiency of our proposed method on continual learning for dialog state tracking, compared with state-of-the-art baselines.
Open-domain questions are likely to be open-ended and ambiguous, leading to multiple valid answers. Existing approaches typically adopt the rerank-then-read framework, where a reader reads top-ranking evidence to predict answers. According to our empirical analysis, this framework faces three problems: first, to leverage a large reader under a memory constraint, the reranker should select only a few relevant passages to cover diverse answers, while balancing relevance and diversity is non-trivial; second, the small reading budget prevents the reader from accessing valuable retrieved evidence filtered out by the reranker; third, when using a generative reader to predict answers all at once based on all selected evidence, whether a valid answer will be predicted also pathologically depends on evidence of some other valid answer(s). To address these issues, we propose to answer open-domain multi-answer questions with a recall-then-verify framework, which separates the reasoning process of each answer so that we can make better use of retrieved evidence while also leveraging large models under the same memory constraint. Our framework achieves state-of-the-art results on two multi-answer datasets, and predicts significantly more gold answers than a rerank-then-read system that uses an oracle reranker.
Existing reference-free metrics have obvious limitations for evaluating controlled text generation models. Unsupervised metrics can only provide a task-agnostic evaluation result which correlates weakly with human judgments, whereas supervised ones may overfit task-specific data with poor generalization ability to other datasets. In this paper, we propose an unsupervised reference-free metric called CTRLEval, which evaluates controlled text generation from different aspects by formulating each aspect into multiple text infilling tasks. On top of these tasks, the metric assembles the generation probabilities from a pre-trained language model without any model training. Experimental results show that our metric has higher correlations with human judgments than other baselines, while obtaining better generalization of evaluating generated texts from different models and with different qualities.
Prompts for pre-trained language models (PLMs) have shown remarkable performance by bridging the gap between pre-training tasks and various downstream tasks. Among these methods, prompt tuning, which freezes PLMs and only tunes soft prompts, provides an efficient and effective solution for adapting large-scale PLMs to downstream tasks. However, prompt tuning is yet to be fully explored. In our pilot experiments, we find that prompt tuning performs comparably with conventional full-model tuning when downstream data are sufficient, whereas it is much worse under few-shot learning settings, which may hinder the application of prompt tuning. We attribute this low performance to the manner of initializing soft prompts. Therefore, in this work, we propose to pre-train prompts by adding soft prompts into the pre-training stage to obtain a better initialization. We name this Pre-trained Prompt Tuning framework “PPT”. To ensure the generalization of PPT, we formulate similar classification tasks into a unified task form and pre-train soft prompts for this unified task. Extensive experiments show that tuning pre-trained prompts for downstream tasks can reach or even outperform full-model fine-tuning under both full-data and few-shot settings. Our approach is effective and efficient for using large-scale PLMs in practice.
Distinct is a widely used automatic metric for evaluating diversity in language generation tasks. However, we observed that the original approach to calculating distinct scores has evident biases that tend to assign higher penalties to longer sequences. We refine the calculation of distinct scores by scaling the number of distinct tokens based on their expectations. We provide both empirical and theoretical evidence to show that our method effectively removes the biases existing in the original distinct score. Our experiments show that our proposed metric, Expectation-Adjusted Distinct (EAD), correlates better with human judgment in evaluating response diversity.To assist future research, we provide an example implementation at https://github.com/lsy641/Expectation-Adjusted-Distinct.
Dialogue contradiction is a critical issue in open-domain dialogue systems. The contextualization nature of conversations makes dialogue contradiction detection rather challenging. In this work, we propose a benchmark for Contradiction Detection in Chinese Conversations, namely CDConv. It contains 12K multi-turn conversations annotated with three typical contradiction categories: Intra-sentence Contradiction, Role Confusion, and History Contradiction. To efficiently construct the CDConv conversations, we devise a series of methods for automatic conversation generation, which simulate common user behaviors that trigger chatbots to make contradictions. We conduct careful manual quality screening of the constructed conversations and show that state-of-the-art Chinese chatbots can be easily goaded into making contradictions. Experiments on CDConv show that properly modeling contextual information is critical for dialogue contradiction detection, but there are still unresolved challenges that require future research.
Human conversations of recommendation naturally involve the shift of interests which can align the recommendation actions and conversation process to make accurate recommendations with rich explanations. However, existing conversational recommendation systems (CRS) ignore the advantage of user interest shift in connecting recommendation and conversation, which leads to an ineffective loose coupling structure of CRS. To address this issue, by modeling the recommendation actions as recommendation paths in a knowledge graph (KG), we propose DICR (Dual Imitation for Conversational Recommendation), which designs a dual imitation to explicitly align the recommendation paths and user interest shift paths in a recommendation module and a conversation module, respectively. By exchanging alignment signals, DICR achieves bidirectional promotion between recommendation and conversation modules and generates high-quality responses with accurate recommendations and coherent explanations. Experiments demonstrate that DICR outperforms the state-of-the-art models on recommendation and conversation performance with automatic, human, and novel explainability metrics.
Training language models to learn from human instructions for zero-shot cross-task generalization has attracted much attention in NLP communities. Recently, instruction tuning (IT), which fine-tunes a pre-trained language model on a massive collection of tasks described via human-craft instructions, has been shown effective in instruction learning for unseen tasks. However, IT relies on a large amount of human-annotated samples, which restricts its generalization. Unlike labeled data, unlabeled data are often massive and cheap to obtain. In this work, we study how IT can be improved with unlabeled data. We first empirically explore the IT performance trends versus the number of labeled data, instructions, and training tasks. We find it critical to enlarge the number of training instructions, and the instructions can be underutilized due to the scarcity of labeled data. Then, we propose Unlabeled Data Augmented Instruction Tuning (UDIT) to take better advantage of the instructions during IT by constructing pseudo-labeled data from unlabeled plain texts. We conduct extensive experiments to show UDIT’s effectiveness in various scenarios of tasks and datasets. We also comprehensively analyze the key factors of UDIT to investigate how to better improve IT with unlabeled data. The code is publicly available at https://github.com/thu-coai/UDIT.
Offensive language detection is increasingly crucial for maintaining a civilized social media platform and deploying pre-trained language models. However, this task in Chinese is still under exploration due to the scarcity of reliable datasets. To this end, we propose a benchmark –COLD for Chinese offensive language analysis, including a Chinese Offensive Language Dataset –COLDATASET and a baseline detector –COLDETECTOR which is trained on the dataset. We show that the COLD benchmark contributes to Chinese offensive language detection which is challenging for existing resources. We then deploy the COLDETECTOR and conduct detailed analyses on popular Chinese pre-trained language models. We first analyze the offensiveness of existing generative models and show that these models inevitably expose varying degrees of offensive issues. Furthermore, we investigate the factors that influence the offensive generations, and we find that anti-bias contents and keywords referring to certain groups or revealing negative attitudes trigger offensive outputs easier.
Automatic essay evaluation can help reduce teachers’ workload and enable students to refine their works rapidly. Previous studies focus mainly on giving discrete scores for either the holistic quality orseveral distinct traits. However, real-world teachers usually provide detailed comments in natural language, which are more informative than single scores. In this paper, we present the comment generation task, which aims to generate commentsfor specified segments from given student narrative essays. To tackle this task, we propose a planning-based generation model, which first plans a sequence of keywords, and then expands these keywords into a complete comment. To improve the correctness and informativeness of generated comments, we adopt two following techniques: (1) training an error correction module to filter out incorrect keywords, and (2) recognizing fine-grained structured features from source essays to enrich the keywords. To support the evaluation of the task, we collect a human-written Chinese dataset, which contains 22,399 essay-comment pairs. Extensive experiments show that our model outperforms strong baselines significantly. Moreover, we exert explicit control on our model to generate comments to describe the strengths or weaknesses of inputs with a 91% success rate. We deploy the model at http://coai.cs.tsinghua.edu.cn/static/essayComment/. A demo video is available at https://youtu.be/IuFVk8dUxbI. Our code and data are available at https://github.com/thu-coai/EssayCommentGen.
Dialogue safety problems severely limit the real-world deployment of neural conversational models and have attracted great research interests recently. However, dialogue safety problems remain under-defined and the corresponding dataset is scarce. We propose a taxonomy for dialogue safety specifically designed to capture unsafe behaviors in human-bot dialogue settings, with focuses on context-sensitive unsafety, which is under-explored in prior works. To spur research in this direction, we compile DiaSafety, a dataset with rich context-sensitive unsafe examples. Experiments show that existing safety guarding tools fail severely on our dataset. As a remedy, we train a dialogue safety classifier to provide a strong baseline for context-sensitive dialogue unsafety detection. With our classifier, we perform safety evaluations on popular conversational models and show that existing dialogue systems still exhibit concerning context-sensitive safety problems.
Chinese spelling check (CSC) detects and corrects spelling errors in Chinese texts. Previous approaches have combined character-level phonetic and graphic information, ignoring the importance of segment-level information. According to our pilot study, spelling errors are always associated with incorrect word segmentation. When appropriate word boundaries are provided, CSC performance is greatly enhanced. Based on these findings, we present WSpeller, a CSC model that takes into account word segmentation. A fundamental component of WSpeller is a W-MLM, which is trained by predicting visually and phonetically similar words. Through modification of the embedding layer’s input, word segmentation information can be incorporated. Additionally, a robust module is trained to assist the W-MLM-based correction module by predicting the correct word segmentations from sentences containing spelling errors. We evaluate WSpeller on the widely used benchmark datasets SIGHAN13, SIGHAN14, and SIGHAN15. Our model is superior to state-of-the-art baselines on SIGHAN13 and SIGHAN15 and maintains equal performance on SIGHAN14.
Recent studies have shown the impressive efficacy of counterfactually augmented data (CAD) for reducing NLU models’ reliance on spurious features and improving their generalizability. However, current methods still heavily rely on human efforts or task-specific designs to generate counterfactuals, thereby impeding CAD’s applicability to a broad range of NLU tasks. In this paper, we present AutoCAD, a fully automatic and task-agnostic CAD generation framework. AutoCAD first leverages a classifier to unsupervisedly identify rationales as spans to be intervened, which disentangles spurious and causal features. Then, AutoCAD performs controllable generation enhanced by unlikelihood training to produce diverse counterfactuals. Extensive evaluations on multiple out-of-domain and challenge benchmarks demonstrate that AutoCAD consistently and significantly boosts the out-of-distribution performance of powerful pre-trained models across different NLU tasks, which is comparable or even better than previous state-of-the-art human-in-the-loop or task-specific CAD methods.
Given that rich information is hidden behind ubiquitous numbers in text, numerical reasoning over text should be an essential skill of AI systems. To derive precise equations to solve numerical reasoning problems, previous work focused on modeling the structures of equations, and has proposed various structured decoders. Though structure modeling proves to be effective, these structured decoders construct a single equation in a pre-defined autoregressive order, potentially placing an unnecessary restriction on how a model should grasp the reasoning process. Intuitively, humans may have numerous pieces of thoughts popping up in no pre-defined order; thoughts are not limited to the problem at hand, and can even be concerned with other related problems. By comparing diverse thoughts and chaining relevant pieces, humans are less prone to errors. In this paper, we take this inspiration and propose CANTOR, a numerical reasoner that models reasoning steps using a directed acyclic graph where we produce diverse reasoning steps simultaneously without pre-defined decoding dependencies, and compare and chain relevant ones to reach a solution. Extensive experiments demonstrated the effectiveness of CANTOR under both fully-supervised and weakly-supervised settings.
Among all the safety concerns that hinder the deployment of open-domain dialog systems (e.g., offensive languages, biases, and toxic behaviors), social bias presents an insidious challenge. Addressing this challenge requires rigorous analyses and normative reasoning. In this paper, we focus our investigation on social bias measurement to facilitate the development of unbiased dialog systems. We first propose a novel Dial-Bias Framework for analyzing the social bias in conversations using a holistic method beyond bias lexicons or dichotomous annotations. Leveraging the proposed framework, we further introduce the CDial-Bias Dataset which is, to the best of our knowledge, the first annotated Chinese social bias dialog dataset. We also establish a fine-grained dialog bias measurement benchmark and conduct in-depth ablation studies to shed light on the utility of the detailed annotations in the proposed dataset. Finally, we evaluate representative Chinese generative models with our classifiers to unveil the presence of social bias in these systems.
Large pretrained language models can easily produce toxic or biased content, which is prohibitive for practical use. In order to detect such toxic generations, existing methods rely on templates, real-world data extraction, crowdsourcing workers or automatic generation to construct adversarial contexts that are likely to induce toxic generations. However, what type of context is more likely to induce unsafe responses is still under-explored. In this paper, we identify that context toxicity and context category (e.g., profanity, insult, drugs, etc.) are two important factors to cause safety issues in response generation. Hence, we propose a method called reverse generation to construct adversarial contexts conditioned on a given response, with the flexibility to control category, toxicity level and inductivity of the generated contexts. Via reverse generation, we augment the existing BAD dataset and construct a new dataset BAD+ which contains more than 120K diverse and highly inductive contexts in 12 categories. We test three popular pretrained dialogue models (Blender, DialoGPT and Plato2) and find that BAD+ can largely expose their safety problems. Furthermore, we show that BAD+ can greatly enhance the safety of generation, and we reveal the key factors of safety improvement. Our code and dataset is available at https://github.com/thu-coai/Reverse_Generation.
Pre-trained language models have shown superior performance in task-oriented dialogues. However, existing datasets are on limited scales, which cannot support large-scale pre-training. Fortunately, various data augmentation methods have been developed to augment large-scale task-oriented dialogue corpora. However, they heavily rely on annotated data in the target domain, which require a tremendous amount of data collection and human labeling work. In this paper, we build a unified dialogue user simulation model by pre-training on several publicly available datasets. The model can then be tuned on a target domain with few-shot data. The experiments on a target dataset across multiple domains show that our proposed model brings remarkable performance increases through data augmentation.
Standard multi-task benchmarks are essential for developing pretraining models that can generalize to various downstream tasks. Existing benchmarks for natural language processing (NLP) usually focus only on understanding or generating short texts. However, long text modeling requires many distinct abilities in contrast to short texts, such as the modeling of long-range discourse and commonsense relations, and the coherence and controllability of generation. The lack of standardized benchmarks makes it difficult to assess these abilities of a model and fairly compare different models, especially Chinese models. Therefore, we propose a story-centric benchmark named LOT for evaluating Chinese long text modeling, which aggregates two understanding tasks and two generation tasks. We construct new datasets for these tasks based on human-written Chinese stories with hundreds of words. Furthermore, we release an encoder-decoder-based Chinese long text pretraining model named LongLM with up to 1 billion parameters. We pretrain LongLM on 120G Chinese novels with two generative tasks including text infilling and conditional continuation. Extensive experiments show that LongLM outperforms similar-sized pretraining models substantially on both the understanding and generation tasks in LOT.
Most language understanding models in task-oriented dialog systems are trained on a small amount of annotated training data, and evaluated in a small set from the same distribution. However, these models can lead to system failure or undesirable output when being exposed to natural language perturbation or variation in practice. In this paper, we conduct comprehensive evaluation and analysis with respect to the robustness of natural language understanding models, and introduce three important aspects related to language understanding in real-world dialog systems, namely, language variety, speech characteristics, and noise perturbation. We propose a model-agnostic toolkit LAUG to approximate natural language perturbations for testing the robustness issues in task-oriented dialog. Four data augmentation approaches covering the three aspects are assembled in LAUG, which reveals critical robustness issues in state-of-the-art models. The augmented dataset through LAUG can be used to facilitate future research on the robustness testing of language understanding in task-oriented dialog.
Unsupervised commonsense question answering is appealing since it does not rely on any labeled task data. Among existing work, a popular solution is to use pre-trained language models to score candidate choices directly conditioned on the question or context. However, such scores from language models can be easily affected by irrelevant factors, such as word frequencies, sentence structures, etc. These distracting factors may not only mislead the model to choose a wrong answer but also make it oversensitive to lexical perturbations in candidate answers. In this paper, we present a novel SEmantic-based Question Answering method (SEQA) for unsupervised commonsense question answering. Instead of directly scoring each answer choice, our method first generates a set of plausible answers with generative models (e.g., GPT-2), and then uses these plausible answers to select the correct choice by considering the semantic similarity between each plausible answer and each choice. We devise a simple, yet sound formalism for this idea and verify its effectiveness and robustness with extensive experiments. We evaluate the proposed method on four benchmark datasets, and our method achieves the best results in unsupervised settings. Moreover, when attacked by TextFooler with synonym replacement, SEQA demonstrates much less performance drops than baselines, thereby indicating stronger robustness.
Pre-trained Language Models (PLMs) have shown superior performance on various downstream Natural Language Processing (NLP) tasks. However, conventional pre-training objectives do not explicitly model relational facts in text, which are crucial for textual understanding. To address this issue, we propose a novel contrastive learning framework ERICA to obtain a deep understanding of the entities and their relations in text. Specifically, we define two novel pre-training tasks to better understand entities and relations: (1) the entity discrimination task to distinguish which tail entity can be inferred by the given head entity and relation; (2) the relation discrimination task to distinguish whether two relations are close or not semantically, which involves complex relational reasoning. Experimental results demonstrate that ERICA can improve typical PLMs (BERT and RoBERTa) on several language understanding tasks, including relation extraction, entity typing and question answering, especially under low-resource settings.
Emotional support is a crucial ability for many conversation scenarios, including social interactions, mental health support, and customer service chats. Following reasonable procedures and using various support skills can help to effectively provide support. However, due to the lack of a well-designed task and corpora of effective emotional support conversations, research on building emotional support into dialog systems remains lacking. In this paper, we define the Emotional Support Conversation (ESC) task and propose an ESC Framework, which is grounded on the Helping Skills Theory. We construct an Emotion Support Conversation dataset (ESConv) with rich annotation (especially support strategy) in a help-seeker and supporter mode. To ensure a corpus of high-quality conversations that provide examples of effective emotional support, we take extensive effort to design training tutorials for supporters and several mechanisms for quality control during data collection. Finally, we evaluate state-of-the-art dialog models with respect to the ability to provide emotional support. Our results show the importance of support strategies in providing effective emotional support and the utility of ESConv in training more emotional support systems.
Neural dialogue generation models trained with the one-hot target distribution suffer from the over-confidence issue, which leads to poor generation diversity as widely reported in the literature. Although existing approaches such as label smoothing can alleviate this issue, they fail to adapt to diverse dialog contexts. In this paper, we propose an Adaptive Label Smoothing (AdaLabel) approach that can adaptively estimate a target label distribution at each time step for different contexts. The maximum probability in the predicted distribution is used to modify the soft target distribution produced by a novel light-weight bi-directional decoder module. The resulting target distribution is aware of both previous and future contexts and is adjusted to avoid over-training the dialogue model. Our model can be trained in an endto-end manner. Extensive experiments on two benchmark datasets show that our approach outperforms various competitive baselines in producing diverse responses.
Weakly supervised question answering usually has only the final answers as supervision signals while the correct solutions to derive the answers are not provided. This setting gives rise to the spurious solution problem: there may exist many spurious solutions that coincidentally derive the correct answer, but training on such solutions can hurt model performance (e.g., producing wrong solutions or answers). For example, for discrete reasoning tasks as on DROP, there may exist many equations to derive a numeric answer, and typically only one of them is correct. Previous learning methods mostly filter out spurious solutions with heuristics or using model confidence, but do not explicitly exploit the semantic correlations between a question and its solution. In this paper, to alleviate the spurious solution problem, we propose to explicitly exploit such semantic correlations by maximizing the mutual information between question-answer pairs and predicted solutions. Extensive experiments on four question answering datasets show that our method significantly outperforms previous learning methods in terms of task performance and is more effective in training models to produce correct solutions.
Generating long and coherent text is an important but challenging task, particularly for open-ended language generation tasks such as story generation. Despite the success in modeling intra-sentence coherence, existing generation models (e.g., BART) still struggle to maintain a coherent event sequence throughout the generated text. We conjecture that this is because of the difficulty for the decoder to capture the high-level semantics and discourse structures in the context beyond token-level co-occurrence. In this paper, we propose a long text generation model, which can represent the prefix sentences at sentence level and discourse level in the decoding process. To this end, we propose two pretraining objectives to learn the representations by predicting inter-sentence semantic similarity and distinguishing between normal and shuffled sentence orders. Extensive experiments show that our model can generate more coherent texts than state-of-the-art baselines.
Automatic metrics are essential for developing natural language generation (NLG) models, particularly for open-ended language generation tasks such as story generation. However, existing automatic metrics are observed to correlate poorly with human evaluation. The lack of standardized benchmark datasets makes it difficult to fully evaluate the capabilities of a metric and fairly compare different metrics. Therefore, we propose OpenMEVA, a benchmark for evaluating open-ended story generation metrics. OpenMEVA provides a comprehensive test suite to assess the capabilities of metrics, including (a) the correlation with human judgments, (b) the generalization to different model outputs and datasets, (c) the ability to judge story coherence, and (d) the robustness to perturbations. To this end, OpenMEVA includes both manually annotated stories and auto-constructed test examples. We evaluate existing metrics on OpenMEVA and observe that they have poor correlation with human judgments, fail to recognize discourse-level incoherence, and lack inferential knowledge (e.g., causal order between events), the generalization ability and robustness. Our study presents insights for developing NLG models and metrics in further research.
There is a long history of research related to automated story generation, dating back as far as the 1970s. Recently, the rapid development of pre-trained language models has spurred great progresses in this field. Equipped with GPT-2 and the latest GPT-3, AI Dungeon has been seen as a famous example of the powerful text generation capabilities of large-scale pre-trained language models, and a possibility for future games. However, as a game, AI Dungeon lacks incentives to players and relies entirely on players to explore on their own. This makes players’ enthusiasm decline rapidly. In this paper, we present an open-ended text adventure game in Chinese, named as KuiLeiXi. In KuiLeiXi, players need to interact with the AI until the pre-determined plot goals are reached. By introducing the plot goals, players have a stronger incentive to explore ways to reach plot goals, while the AI’s abilities are not abused to generate harmful contents. This limited freedom allows this game to be integrated as a part of a romance simulation mobile game, Yu Jian Love. Since KuiLeiXi was launched, it has received a lot of positive feedbacks from more than 100,000 players. A demo video is available at https://youtu.be/DyYZhxMRrkk.
Further pre-training language models on in-domain data (domain-adaptive pre-training, DAPT) or task-relevant data (task-adaptive pre-training, TAPT) before fine-tuning has been shown to improve downstream tasks’ performances. However, in task-oriented dialog modeling, we observe that further pre-training MLM does not always boost the performance on a downstream task. We find that DAPT is beneficial in the low-resource setting, but as the fine-tuning data size grows, DAPT becomes less beneficial or even useless, and scaling the size of DAPT data does not help. Through Representational Similarity Analysis, we conclude that more data for fine-tuning yields greater change of the model’s representations and thus reduces the influence of initialization.
User satisfaction estimation in the dialogue-based customer service is critical not only for helping developers find the system defects, but also making it possible to get timely human intervention for dissatisfied customers. In this paper, we investigate the problem of user satisfaction estimation in E-commerce customer service. In order to apply the estimator to online services for timely human intervention, we need to estimate the satisfaction score at each turn. However, in actual scenario we can only collect the satisfaction labels for the whole dialogue sessions via user feedback. To this end, we formalize the turn-level satisfaction estimation as a reinforcement learning problem, in which the model can be optimized with only session-level satisfaction labels. We conduct experiments on the dataset collected from a commercial customer service system, and compare our model with the supervised learning models. Extensive experiments show that the proposed method outperforms all the baseline models.
Growing interests have been attracted in Conversational Recommender Systems (CRS), which explore user preference through conversational interactions in order to make appropriate recommendation. However, there is still a lack of ability in existing CRS to (1) traverse multiple reasoning paths over background knowledge to introduce relevant items and attributes, and (2) arrange selected entities appropriately under current system intents to control response generation. To address these issues, we propose CR-Walker in this paper, a model that performs tree-structured reasoning on a knowledge graph, and generates informative dialog acts to guide language generation. The unique scheme of tree-structured reasoning views the traversed entity at each hop as part of dialog acts to facilitate language generation, which links how entities are selected and expressed. Automatic and human evaluations show that CR-Walker can arrive at more accurate recommendation, and generate more informative and engaging responses.
As the labeling cost for different modules in task-oriented dialog (ToD) systems is expensive, a major challenge is to train different modules with the least amount of labeled data. Recently, large-scale pre-trained language models, have shown promising results for few-shot learning in ToD. In this paper, we devise a self-training approach to utilize the abundant unlabeled dialog data to further improve state-of-the-art pre-trained models in few-shot learning scenarios for ToD systems. Specifically, we propose a self-training approach that iteratively labels the most confident unlabeled data to train a stronger Student model. Moreover, a new text augmentation technique (GradAug) is proposed to better train the Student by replacing non-crucial tokens using a masked language model. We conduct extensive experiments and present analyses on four downstream tasks in ToD, including intent classification, dialog state tracking, dialog act prediction, and response selection. Empirical results demonstrate that the proposed self-training approach consistently improves state-of-the-art pre-trained models (BERT, ToD-BERT) when only a small number of labeled data are available.
Grounded dialogue models generate responses that are grounded on certain concepts. Limited by the distribution of grounded dialogue data, models trained on such data face the transferability challenges in terms of the data distribution and the type of grounded concepts. To address the challenges, we propose the grounded minimal editing framework, which minimally edits existing responses to be grounded on the given concept. Focusing on personas, we propose Grounded Minimal Editor (GME), which learns to edit by disentangling and recombining persona-related and persona-agnostic parts of the response. To evaluate persona-grounded minimal editing, we present the PersonaMi-nEdit dataset, and experimental results show that GME outperforms competitive baselines by a large margin. To evaluate the transferability, we experiment on the test set of BlendedSkillTalk and show that GME can edit dialogue models’ responses to largely improve their persona consistency while preserving the use of knowledge and empathy.
Generating informative and appropriate responses is challenging but important for building human-like dialogue systems. Although various knowledge-grounded conversation models have been proposed, these models have limitations in utilizing knowledge that infrequently occurs in the training data, not to mention integrating unseen knowledge into conversation generation. In this paper, we propose an Entity-Agnostic Representation Learning (EARL) method to introduce knowledge graphs to informative conversation generation. Unlike traditional approaches that parameterize the specific representation for each entity, EARL utilizes the context of conversations and the relational structure of knowledge graphs to learn the category representation for entities, which is generalized to incorporating unseen entities in knowledge graphs into conversation generation. Automatic and manual evaluations demonstrate that our model can generate more informative, coherent, and natural responses than baseline models.
Despite the recent advances in applying pre-trained language models to generate high-quality texts, generating long passages that maintain long-range coherence is yet challenging for these models. In this paper, we propose DiscoDVT, a discourse-aware discrete variational Transformer to tackle the incoherence issue. DiscoDVT learns a discrete variable sequence that summarizes the global structure of the text and then applies it to guide the generation process at each decoding step. To further embed discourse-aware information into the discrete latent representations, we introduce an auxiliary objective to model the discourse relations within the text. We conduct extensive experiments on two open story generation datasets and demonstrate that the latent codes learn meaningful correspondence to the discourse structures that guide the model to generate long texts with better long-range coherence.
Most existing approaches for goal-oriented dialogue policy learning used reinforcement learning, which focuses on the target agent policy and simply treats the opposite agent policy as part of the environment. While in real-world scenarios, the behavior of an opposite agent often exhibits certain patterns or underlies hidden policies, which can be inferred and utilized by the target agent to facilitate its own decision making. This strategy is common in human mental simulation by first imaging a specific action and the probable results before really acting it. We therefore propose an opposite behavior aware framework for policy learning in goal-oriented dialogues. We estimate the opposite agent’s policy from its behavior and use this estimation to improve the target agent by regarding it as part of the target policy. We evaluate our model on both cooperative and competitive dialogue tasks, showing superior performance over state-of-the-art baselines.
Commonsense explanation generation aims to empower the machine’s sense-making capability by generating plausible explanations to statements against commonsense. While this task is easy to human, the machine still struggles to generate reasonable and informative explanations. In this work, we propose a method that first extracts the underlying concepts which are served as bridges in the reasoning chain and then integrates these concepts to generate the final explanation. To facilitate the reasoning process, we utilize external commonsense knowledge to build the connection between a statement and the bridge concepts by extracting and pruning multi-hop paths to build a subgraph. We design a bridge concept extraction model that first scores the triples, routes the paths in the subgraph, and further selects bridge concepts with weak supervision at both the triple level and the concept level. We conduct experiments on the commonsense explanation generation task and our model outperforms the state-of-the-art baselines in both automatic and human evaluation.
Within the prosperity of Massive Open Online Courses (MOOCs), the education applications that automatically provide extracurricular knowledge for MOOC users become rising research topics. However, MOOC courses’ diversity and rapid updates make it more challenging to find suitable new knowledge for students. In this paper, we present ExpanRL, an end-to-end hierarchical reinforcement learning (HRL) model for concept expansion in MOOCs. Employing a two-level HRL mechanism of seed selection and concept expansion, ExpanRL is more feasible to adjust the expansion strategy to find new concepts based on the students’ feedback on expansion results. Our experiments on nine novel datasets from real MOOCs show that ExpanRL achieves significant improvements over existing methods and maintain competitive performance under different settings.
Many studies have applied reinforcement learning to train a dialog policy and show great promise these years. One common approach is to employ a user simulator to obtain a large number of simulated user experiences for reinforcement learning algorithms. However, modeling a realistic user simulator is challenging. A rule-based simulator requires heavy domain expertise for complex tasks, and a data-driven simulator requires considerable data and it is even unclear how to evaluate a simulator. To avoid explicitly building a user simulator beforehand, we propose Multi-Agent Dialog Policy Learning, which regards both the system and the user as the dialog agents. Two agents interact with each other and are jointly learned simultaneously. The method uses the actor-critic framework to facilitate pretraining and improve scalability. We also propose Hybrid Value Network for the role-aware reward decomposition to integrate role-specific domain knowledge of each agent in the task-oriented dialog. Results show that our method can successfully build a system policy and a user policy simultaneously, and two agents can achieve a high task success rate through conversational interaction.
Neural models have achieved great success on machine reading comprehension (MRC), many of which typically consist of two components: an evidence extractor and an answer predictor. The former seeks the most relevant information from a reference text, while the latter is to locate or generate answers from the extracted evidence. Despite the importance of evidence labels for training the evidence extractor, they are not cheaply accessible, particularly in many non-extractive MRC tasks such as YES/NO question answering and multi-choice MRC. To address this problem, we present a Self-Training method (STM), which supervises the evidence extractor with auto-generated evidence labels in an iterative process. At each iteration, a base MRC model is trained with golden answers and noisy evidence labels. The trained model will predict pseudo evidence labels as extra supervision in the next iteration. We evaluate STM on seven datasets over three MRC tasks. Experimental results demonstrate the improvement on existing MRC models, and we also analyze how and why such a self-training method works in MRC.
The research of knowledge-driven conversational systems is largely limited due to the lack of dialog data which consists of multi-turn conversations on multiple topics and with knowledge annotations. In this paper, we propose a Chinese multi-domain knowledge-driven conversation dataset, KdConv, which grounds the topics in multi-turn conversations to knowledge graphs. Our corpus contains 4.5K conversations from three domains (film, music, and travel), and 86K utterances with an average turn number of 19.0. These conversations contain in-depth discussions on related topics and natural transition between multiple topics. To facilitate the following research on this corpus, we provide several benchmark models. Comparative results show that the models can be enhanced by introducing background knowledge, yet there is still a large space for leveraging knowledge to model multi-turn conversations for further research. Results also show that there are obvious performance differences between different domains, indicating that it is worth further explore transfer learning and domain adaptation. The corpus and benchmark models are publicly available.
We present ConvLab-2, an open-source toolkit that enables researchers to build task-oriented dialogue systems with state-of-the-art models, perform an end-to-end evaluation, and diagnose the weakness of systems. As the successor of ConvLab, ConvLab-2 inherits ConvLab’s framework but integrates more powerful dialogue models and supports more datasets. Besides, we have developed an analysis tool and an interactive tool to assist researchers in diagnosing dialogue systems. The analysis tool presents rich statistics and summarizes common mistakes from simulated dialogues, which facilitates error analysis and system improvement. The interactive tool provides an user interface that allows developers to diagnose an assembled dialogue system by interacting with the system and modifying the output of each system component.
There is a growing interest in developing goal-oriented dialog systems which serve users in accomplishing complex tasks through multi-turn conversations. Although many methods are devised to evaluate and improve the performance of individual dialog components, there is a lack of comprehensive empirical study on how different components contribute to the overall performance of a dialog system. In this paper, we perform a system-wise evaluation and present an empirical analysis on different types of dialog systems which are composed of different modules in different settings. Our results show that (1) a pipeline dialog system trained using fine-grained supervision signals at different component levels often obtains better performance than the systems that use joint or end-to-end models trained on coarse-grained labels, (2) component-wise, single-turn evaluation results are not always consistent with the overall performance of a dialog system, and (3) despite the discrepancy between simulators and human users, simulated evaluation is still a valid alternative to the costly human evaluation especially in the early stage of development.
Story generation, namely, generating a reasonable story from a leading context, is an important but challenging task. In spite of the success in modeling fluency and local coherence, existing neural language generation models (e.g., GPT-2) still suffer from repetition, logic conflicts, and lack of long-range coherence in generated stories. We conjecture that this is because of the difficulty of associating relevant commonsense knowledge, understanding the causal relationships, and planning entities and events with proper temporal order. In this paper, we devise a knowledge-enhanced pretraining model for commonsense story generation. We propose to utilize commonsense knowledge from external knowledge bases to generate reasonable stories. To further capture the causal and temporal dependencies between the sentences in a reasonable story, we use multi-task learning, which combines a discriminative objective to distinguish true and fake stories during fine-tuning. Automatic and manual evaluation shows that our model can generate more reasonable stories than state-of-the-art baselines, particularly in terms of logic and global coherence.
To advance multi-domain (cross-domain) dialogue modeling as well as alleviate the shortage of Chinese task-oriented datasets, we propose CrossWOZ, the first large-scale Chinese Cross-Domain Wizard-of-Oz task-oriented dataset. It contains 6K dialogue sessions and 102K utterances for 5 domains, including hotel, restaurant, attraction, metro, and taxi. Moreover, the corpus contains rich annotation of dialogue states and dialogue acts on both user and system sides. About 60% of the dialogues have cross-domain user goals that favor inter-domain dependency and encourage natural transition across domains in conversation. We also provide a user simulator and several benchmark models for pipelined task-oriented dialogue systems, which will facilitate researchers to compare and evaluate their models on this corpus. The large size and rich annotation of CrossWOZ make it suitable to investigate a variety of tasks in cross-domain dialogue modeling, such as dialogue state tracking, policy learning, user simulation, etc.
In a multi-turn knowledge-grounded dialog, the difference between the knowledge selected at different turns usually provides potential clues to knowledge selection, which has been largely neglected in previous research. In this paper, we propose a difference-aware knowledge selection method. It first computes the difference between the candidate knowledge sentences provided at the current turn and those chosen in the previous turns. Then, the differential information is fused with or disentangled from the contextual information to facilitate final knowledge selection. Automatic, human observational, and interactive evaluation shows that our method is able to select knowledge more accurately and generate more informative responses, significantly outperforming the state-of-the-art baselines.
Revealing the robustness issues of natural language processing models and improving their robustness is important to their performance under difficult situations. In this paper, we study the robustness of paraphrase identification models from a new perspective – via modification with shared words, and we show that the models have significant robustness issues when facing such modifications. To modify an example consisting of a sentence pair, we either replace some words shared by both sentences or introduce new shared words. We aim to construct a valid new example such that a target model makes a wrong prediction. To find a modification solution, we use beam search constrained by heuristic rules, and we leverage a BERT masked language model for generating substitution words compatible with the context. Experiments show that the performance of the target models has a dramatic drop on the modified examples, thereby revealing the robustness issue. We also show that adversarial training can mitigate this issue.
Natural language generation (NLG) is an essential component of task-oriented dialog systems. Despite the recent success of neural approaches for NLG, they are typically developed in an offline manner for particular domains. To better fit real-life applications where new data come in a stream, we study NLG in a “continual learning” setting to expand its knowledge to new domains or functionalities incrementally. The major challenge towards this goal is catastrophic forgetting, meaning that a continually trained model tends to forget the knowledge it has learned before. To this end, we propose a method called ARPER (Adaptively Regularized Prioritized Exemplar Replay) by replaying prioritized historical exemplars, together with an adaptive regularization technique based on Elastic Weight Consolidation. Extensive experiments to continually learn new domains and intents are conducted on MultiWoZ-2.0 to benchmark ARPER with a wide range of techniques. Empirical results demonstrate that ARPER significantly outperforms other methods by effectively mitigating the detrimental catastrophic forgetting issue.
Despite the success of generative pre-trained language models on a series of text generation tasks, they still suffer in cases where reasoning over underlying commonsense knowledge is required during generation. Existing approaches that integrate commonsense knowledge into generative pre-trained language models simply transfer relational knowledge by post-training on individual knowledge triples while ignoring rich connections within the knowledge graph. We argue that exploiting both the structural and semantic information of the knowledge graph facilitates commonsense-aware text generation. In this paper, we propose Generation with Multi-Hop Reasoning Flow (GRF) that enables pre-trained models with dynamic multi-hop reasoning on multi-relational paths extracted from the external commonsense knowledge graph. We empirically show that our model outperforms existing baselines on three text generation tasks that require reasoning over commonsense knowledge. We also demonstrate the effectiveness of the dynamic multi-hop reasoning module with reasoning paths inferred by the model that provide rationale to the generation.
Recent advances in open-domain dialogue systems rely on the success of neural models that are trained on large-scale data. However, collecting large-scale dialogue data is usually time-consuming and labor-intensive. To address this data dilemma, we propose a novel data augmentation method for training open-domain dialogue models by utilizing unpaired data. Specifically, a data-level distillation process is first proposed to construct augmented dialogues where both post and response are retrieved from the unpaired data. A ranking module is employed to filter out low-quality dialogues. Further, a model-level distillation process is employed to distill a teacher model trained on high-quality paired data to augmented dialogue pairs, thereby preventing dialogue models from being affected by the noise in the augmented data. Automatic and manual evaluation indicates that our method can produce high-quality dialogue pairs with diverse contents, and the proposed data-level and model-level dialogue distillation can improve the performance of competitive baselines.
Most of the existing pre-trained language representation models neglect to consider the linguistic knowledge of texts, which can promote language understanding in NLP tasks. To benefit the downstream tasks in sentiment analysis, we propose a novel language representation model called SentiLARE, which introduces word-level linguistic knowledge including part-of-speech tag and sentiment polarity (inferred from SentiWordNet) into pre-trained models. We first propose a context-aware sentiment attention mechanism to acquire the sentiment polarity of each word with its part-of-speech tag by querying SentiWordNet. Then, we devise a new pre-training task called label-aware masked language model to construct knowledge-aware language representation. Experiments show that SentiLARE obtains new state-of-the-art performance on a variety of sentiment analysis tasks.
Despite the success of existing referenced metrics (e.g., BLEU and MoverScore), they correlate poorly with human judgments for open-ended text generation including story or dialog generation because of the notorious one-to-many issue: there are many plausible outputs for the same input, which may differ substantially in literal or semantics from the limited number of given references. To alleviate this issue, we propose UNION, a learnable UNreferenced metrIc for evaluating Open-eNded story generation, which measures the quality of a generated story without any reference. Built on top of BERT, UNION is trained to distinguish human-written stories from negative samples and recover the perturbation in negative stories. We propose an approach of constructing negative samples by mimicking the errors commonly observed in existing NLG models, including repeated plots, conflicting logic, and long-range incoherence. Experiments on two story datasets demonstrate that UNION is a reliable measure for evaluating the quality of generated stories, which correlates better with human judgments and is more generalizable than existing state-of-the-art metrics.
Recently, a variety of neural models have been proposed for lyrics generation. However, most previous work completes the generation process in a single pass with little human intervention. We believe that lyrics creation is a creative process with human intelligence centered. AI should play a role as an assistant in the lyrics creation process, where human interactions are crucial for high-quality creation. This paper demonstrates Youling, an AI-assisted lyrics creation system, designed to collaborate with music creators. In the lyrics generation process, Youling supports traditional one pass full-text generation mode as well as an interactive generation mode, which allows users to select the satisfactory sentences from generated candidates conditioned on preceding context. The system also provides a revision module which enables users to revise undesired sentences or words of lyrics repeatedly. Besides, Youling allows users to use multifaceted attributes to control the content and format of generated lyrics. The demo video of the system is available at https://youtu.be/DFeNpHk0pm4.
Cloze-style reading comprehension in Chinese is still limited due to the lack of various corpora. In this paper we propose a large-scale Chinese cloze test dataset ChID, which studies the comprehension of idiom, a unique language phenomenon in Chinese. In this corpus, the idioms in a passage are replaced by blank symbols and the correct answer needs to be chosen from well-designed candidate idioms. We carefully study how the design of candidate idioms and the representation of idioms affect the performance of state-of-the-art models. Results show that the machine accuracy is substantially worse than that of human, indicating a large space for further research.
We present ConvLab, an open-source multi-domain end-to-end dialog system platform, that enables researchers to quickly set up experiments with reusable components and compare a large set of different approaches, ranging from conventional pipeline systems to end-to-end neural models, in common environments. ConvLab offers a set of fully annotated datasets and associated pre-trained reference models. As a showcase, we extend the MultiWOZ dataset with user dialog act annotations to train all component models and demonstrate how ConvLab makes it easy and effortless to conduct complicated experiments in multi-domain end-to-end dialog settings.
Dialog policy decides what and how a task-oriented dialog system will respond, and plays a vital role in delivering effective conversations. Many studies apply Reinforcement Learning to learn a dialog policy with the reward function which requires elaborate design and pre-specified user goals. With the growing needs to handle complex goals across multiple domains, such manually designed reward functions are not affordable to deal with the complexity of real-world tasks. To this end, we propose Guided Dialog Policy Learning, a novel algorithm based on Adversarial Inverse Reinforcement Learning for joint reward estimation and policy optimization in multi-domain task-oriented dialog. The proposed approach estimates the reward signal and infers the user goal in the dialog sessions. The reward estimator evaluates the state-action pairs so that it can guide the dialog policy at each dialog turn. Extensive experiments on a multi-domain dialog dataset show that the dialog policy guided by the learned reward function achieves remarkably higher task success than state-of-the-art baselines.
Existing neural methods for data-to-text generation are still struggling to produce long and diverse texts: they are insufficient to model input data dynamically during generation, to capture inter-sentence coherence, or to generate diversified expressions. To address these issues, we propose a Planning-based Hierarchical Variational Model (PHVM). Our model first plans a sequence of groups (each group is a subset of input items to be covered by a sentence) and then realizes each sentence conditioned on the planning result and the previously generated context, thereby decomposing long text generation into dependent sentence generation sub-tasks. To capture expression diversity, we devise a hierarchical latent structure where a global planning latent variable models the diversity of reasonable planning and a sequence of local latent variables controls sentence realization. Experiments show that our model outperforms state-of-the-art baselines in long and diverse text generation.
Most of the existing generative adversarial networks (GAN) for text generation suffer from the instability of reinforcement learning training algorithms such as policy gradient, leading to unstable performance. To tackle this problem, we propose a novel framework called Adversarial Reward Augmented Maximum Likelihood (ARAML). During adversarial training, the discriminator assigns rewards to samples which are acquired from a stationary distribution near the data rather than the generator’s distribution. The generator is optimized with maximum likelihood estimation augmented by the discriminator’s rewards instead of policy gradient. Experiments show that our model can outperform state-of-the-art text GANs with a more stable training process.
Sentence function is a significant factor to achieve the purpose of the speaker, which, however, has not been touched in large-scale conversation generation so far. In this paper, we present a model to generate informative responses with controlled sentence function. Our model utilizes a continuous latent variable to capture various word patterns that realize the expected sentence function, and introduces a type controller to deal with the compatibility of controlling sentence function and generating informative content. Conditioned on the latent variable, the type controller determines the type (i.e., function-related, topic, and ordinary word) of a word to be generated at each decoding position. Experiments show that our model outperforms state-of-the-art baselines, and it has the ability to generate responses with both controlled sentence function and informative content.
Asking good questions in open-domain conversational systems is quite significant but rather untouched. This task, substantially different from traditional question generation, requires to question not only with various patterns but also on diverse and relevant topics. We observe that a good question is a natural composition of interrogatives, topic words, and ordinary words. Interrogatives lexicalize the pattern of questioning, topic words address the key information for topic transition in dialogue, and ordinary words play syntactical and grammatical roles in making a natural sentence. We devise two typed decoders (soft typed decoder and hard typed decoder) in which a type distribution over the three types is estimated and the type distribution is used to modulate the final generation distribution. Extensive experiments show that the typed decoders outperform state-of-the-art baselines and can generate more meaningful questions.
Sentence compression condenses a sentence while preserving its most important contents. Delete-based models have the strong ability to delete undesired words, while generate-based models are able to reorder or rephrase the words, which are more coherent to human sentence compression. In this paper, we propose Operation Network, a neural network approach for abstractive sentence compression, which combines the advantages of both delete-based and generate-based sentence compression models. The central idea of Operation Network is to model the sentence compression process as an editing procedure. First, unnecessary words are deleted from the source sentence, then new words are either generated from a large vocabulary or copied directly from the source sentence. A compressed sentence can be obtained by a series of such edit operations (delete, copy and generate). Experiments show that Operation Network outperforms state-of-the-art baselines.
Multi-relation Question Answering is a challenging task, due to the requirement of elaborated analysis on questions and reasoning over multiple fact triples in knowledge base. In this paper, we present a novel model called Interpretable Reasoning Network that employs an interpretable, hop-by-hop reasoning process for question answering. The model dynamically decides which part of an input question should be analyzed at each hop; predicts a relation that corresponds to the current parsed results; utilizes the predicted relation to update the question representation and the state of the reasoning process; and then drives the next-hop reasoning. Experiments show that our model yields state-of-the-art results on two datasets. More interestingly, the model can offer traceable and observable intermediate predictions for reasoning analysis and failure diagnosis, thereby allowing manual manipulation in predicting the final answer.
This paper deals with sentence-level sentiment classification. Though a variety of neural network models have been proposed recently, however, previous models either depend on expensive phrase-level annotation, most of which has remarkably degraded performance when trained with only sentence-level annotation; or do not fully employ linguistic resources (e.g., sentiment lexicons, negation words, intensity words). In this paper, we propose simple models trained with sentence-level annotation, but also attempt to model the linguistic role of sentiment lexicons, negation words, and intensity words. Results show that our models are able to capture the linguistic role of sentiment words, negation words, and intensity words in sentiment expression.
Knowledge embedding, which projects triples in a given knowledge base to d-dimensional vectors, has attracted considerable research efforts recently. Most existing approaches treat the given knowledge base as a set of triplets, each of whose representation is then learned separately. However, as a fact, triples are connected and depend on each other. In this paper, we propose a graph aware knowledge embedding method (GAKE), which formulates knowledge base as a directed graph, and learns representations for any vertices or edges by leveraging the graph’s structural information. We introduce three types of graph context for embedding: neighbor context, path context, and edge context, each reflects properties of knowledge from different perspectives. We also design an attention mechanism to learn representative power of different vertices or edges. To validate our method, we conduct several experiments on two tasks. Experimental results suggest that our method outperforms several state-of-art knowledge embedding models.
We propose a phrase-based approach for generating product review summaries. The main idea of our method is to leverage phrase properties to choose a subset of optimal phrases for generating the final summary. Specifically, we exploit two phrase properties, popularity and specificity. Popularity describes how popular the phrase is in the original reviews. Specificity describes how descriptive a phrase is in comparison to generic comments. We formalize the phrase selection procedure as an optimization problem and solve it using integer linear programming (ILP). An aspect-based bigram language model is used for generating the final summary with the selected phrases. Experiments show that our summarizer outperforms the other baselines.
Natural language generation (NLG) is an important component of question answering(QA) systems which has a significant impact on system quality. Most tranditional QA systems based on templates or rules tend to generate rigid and stylised responses without the natural variation of human language. Furthermore, such methods need an amount of work to generate the templates or rules. To address this problem, we propose a Context-Aware LSTM model for NLG. The model is completely driven by data without manual designed templates or rules. In addition, the context information, including the question to be answered, semantic values to be addressed in the response, and the dialogue act type during interaction, are well approached in the neural network model, which enables the model to produce variant and informative responses. The quantitative evaluation and human evaluation show that CA-LSTM obtains state-of-the-art performance.