Personalizing dialogue agents is important for dialogue systems to generate more specific,consistent, and engaging responses. However, most current dialogue personalization approaches rely on explicit persona descriptions during inference, which severely restricts its application. In this paper, we propose a novel approach that learns to predict persona information based on the dialogue history to personalize the dialogue agent without relying on any explicit persona descriptions during inference. Experimental results on the PersonaChat dataset show that the proposed method can improve the consistency of generated responses when conditioning on the predicted profile of the dialogue agent (i.e. “self persona”), and improve the engagingness of the generated responses when conditioning on the predicted persona of the dialogue partner (i.e. “their persona”). We also find that a trained persona prediction model can be successfully transferred to other datasets and help generate more relevant responses.

Despite serving as the foundation models for a wide range of NLP benchmarks, pre-trained language models have shown limited capabilities of acquiring implicit commonsense knowledge from self-supervision alone, compared to learning linguistic and factual knowledge that appear more explicitly in the surface patterns in text. In this work, we introduce commonsense knowledge transfer, a framework to transfer the commonsense knowledge stored in a neural commonsense knowledge model to a general-purpose pre-trained language model. It first exploits general texts to form queries for extracting commonsense knowledge from the neural commonsense knowledge model and then refines the language model with two self-supervised objectives: commonsense mask infilling and commonsense relation prediction, which align human language with the underlying commonsense knowledge. Empirical results show that our approach consistently improves the model’s performance on downstream tasks that require commonsense reasoning. Moreover, we find that the improvement is more significant in the few-shot setting. This suggests that our approach helps language models better transfer to downstream tasks without extensive supervision by injecting commonsense knowledge into their parameters.

Pre-trained Transformer models like T5 and BART have advanced the state of the art on a wide range of text generation tasks. Compressing these models into smaller ones has become critically important for practical use. Common neural network compression techniques such as knowledge distillation or quantization are limited to static compression where the compression ratio is fixed. In this paper, we introduce Modular Transformers, a modularized encoder-decoder framework for flexible sequence-to-sequence model compression. Modular Transformers trains modularized layers that have the same function of two or more consecutive layers in the original model via module replacing and knowledge distillation. After training, the modularized layers can be flexibly assembled into sequence-to-sequence models that meet different performance-efficiency trade-offs. Experimental results show that after a single training phase, by simply varying the assemble strategy, Modular Transformers can achieve flexible compression ratios from 1.1x to 6x with little to moderate relative performance drop.

Pre-trained vision-language models (VLMs) have achieved impressive results in a range of vision-language tasks. However, popular VLMs usually consist of hundreds of millions of parameters which brings challenges for fine-tuning and deployment in real-world applications due to space, memory, and latency constraints. In this work, we introduce a distilling then pruning framework to compress large vision-language models into smaller, faster, and more accurate ones. We first shrink the size ofa pre-trained large VLM and apply knowledge distillation in the vision-language pre-training stage to obtain a task-agnostic compact VLM. Then we propose a modal-adaptive pruning algorithm to automatically infer the importance of vision and language modalities for different downstream tasks and adaptively remove redundant structures and neurons in different encoders with controllable target sparsity. We apply our framework to train EfficientVLM, a fast and accurate vision-language model consisting of 6 vision layers, 3 text layers, and 3 cross-modal fusion layers, accounting for only 93 million parameters in total, which is 44.3% of the teacher model. EfficientVLM retains 98.4% performance of the teacher model and accelerates its inference speed by 2.2×. EfficientVLM achieves a large absolute improvement over previous SoTA efficient VLMs of similar sizes by a large margin on various vision-language tasks, including VQAv2 (+4.9%), NLVR2 (+5.6%), ITR (R@1 on TR +17.2%, on IR + 15.6% ) and COCO caption generation (CIDEr +6.5), demonstrating a large potential on training lightweight VLMs.

*Data Synthesis* is a promising way to train a small model with very little labeled data. One approach for data synthesis is to leverage the rich knowledge from large language models to synthesize pseudo training examples for small models, making it possible to achieve both data and compute efficiency at the same time. However, a key challenge in data synthesis is that the synthesized dataset often suffers from a large distributional discrepancy from the *real task* data distribution. Thus, in this paper, we propose *Synthesis Step by Step* (**S3**), a data synthesis framework that shrinks this distribution gap by iteratively extrapolating the errors made by a small model trained on the synthesized dataset on a small real-world validation dataset using a large language model. Extensive experiments on multiple NLP tasks show that our approach improves the performance of a small model by reducing the gap between the synthetic dataset and the real data, resulting in significant improvement compared to several baselines: 9.48% improvement compared to ZeroGen and 2.73% compared to GoldGen, and at most 15.17% improvement compared to the small model trained on human-annotated data.

While recent studies have looked into the abilities of large language models in various benchmark tasks, including question generation, reading comprehension, multilingual and etc, there have been few studies looking into the controllability of large language models on generation tasks. We present an extensive analysis of various benchmarks including a sentence planning benchmark with different granularities. After comparing large language models against state-of-the-start finetuned smaller models, we present a spectrum showing large language models falling behind, are comparable, or exceed the ability of smaller models. We conclude that *large language models struggle at meeting fine-grained hard constraints*.

Recent work has shown that language models (LMs) have strong multi-step (i.e., procedural) reasoning capabilities. However, it is unclear whether LMs perform these tasks by cheating with answers memorized from pretraining corpus, or, via a multi-step reasoning mechanism. In this paper, we try to answer this question by exploring a mechanistic interpretation of LMs for multi-step reasoning tasks. Concretely, we hypothesize that the LM implicitly embeds a reasoning tree resembling the correct reasoning process within it. We test this hypothesis by introducing a new probing approach (called MechanisticProbe) that recovers the reasoning tree from the model’s attention patterns. We use our probe to analyze two LMs: GPT-2 on a synthetic task (k-th smallest element), and LLaMA on two simple language-based reasoning tasks (ProofWriter & AI2 Reasoning Challenge). We show that MechanisticProbe is able to detect the information of the reasoning tree from the model’s attentions for most examples, suggesting that the LM indeed is going through a process of multi-step reasoning within its architecture in many cases.

Improving the quality of academic writing is a meaningful but challenging task. Conventional methods of language refinement focus on narrow, specific linguistic features within isolated sentences, such as grammatical errors and improper word use. We propose a more general task, Academic Writing Formalization (AWF), to improve the overall quality of formal academic writing at the paragraph level. We formulate this language refinement task as a formal text style transfer task which transfers informal-academic text to formal-academic and contribute a large-scale non-parallel dataset, Doolittle, for this purpose. Concurrently, we apply a method named metric-oriented reinforcement learning (MORL) to two large language models (LLM) where we incorporate different levels of automatic feedback into the training process. Our experiments reveal that existing text transfer models and grammatical error correction models address certain aspects of AWF but still have a significant performance gap compared to human performance. Meanwhile, language models fine-tuned with our MORL method exhibit considerably improved performance, rivaling the latest chatbot ChatGPT, but still have a non-negligible gap compared to the ground truth formal-academic texts in Doolittle.

In this paper, we introduce Cross-View Language Modeling, a simple and effective pre-training framework that unifies cross-lingual and cross-modal pre-training with shared architectures and objectives. Our approach is motivated by a key observation that cross-lingual and cross-modal pre-training share the same goal of aligning two different views of the same object into a common semantic space. To this end, the cross-view language modeling framework considers both multi-modal data (i.e., image-caption pairs) and multi-lingual data (i.e., parallel sentence pairs) as two different views of the same object, and trains the model to align the two views by maximizing the mutual information between them with conditional masked language modeling and contrastive learning. We pre-train CCLM, a Cross-lingual Cross-modal Language Model, with the cross-view language modeling framework. Empirical results on IGLUE, a multi-lingual multi-modal benchmark, and two multi-lingual image-text retrieval datasets show that while conceptually simpler, CCLM significantly outperforms the prior state-of-the-art with an average absolute improvement of over 10%. Moreover, CCLM is the first multi-lingual multi-modal pre-trained model that surpasses the translate-test performance of representative English vision-language models by zero-shot cross-lingual transfer.

Machine translation quality estimation (QE) predicts human judgements of a translation hypothesis without seeing the reference. State-of-the-art QE systems based on pretrained language models have been achieving remarkable correlations with human judgements yet they are computationally heavy and require human annotations, which are slow and expensive to create. To address these limitations, we define the problem of metric estimation (ME) where one predicts the automated metric scores also without the reference. We show that even without access to the reference, our model can estimate automated metrics (ρ = 60% for BLEU, ρ = 51% for other metrics) at the sentence-level. Because automated metrics correlate with human judgements, we can leverage the ME task for pre-training a QE model. For the QE task, we find that pre-training on TER is better (ρ = 23%) than training for scratch (ρ = 20%).

The WMT 2023 Terminology Shared Task investigates progress in machine translation of texts with specialized vocabulary. The participants were given the source text and segment-level terminology dictionaries for three language pairs: Chinese→English, English→Czech, and German→English. We evaluate 21 submissions from 7 teams on two main criteria: general translation quality and the effectiveness of translating specialized terminology. Systems took varied approaches — incorporating terminology at inference time or weakly supervised training that uses terminology access. While incorporating terminology dictionaries leads to improvement in the translation quality, incorporating an equal amount of information from the reference leads to similar results. This challenges the position of terminologies being the crux of meaning in translation, it can also be explained by inadequate metrics which are not terminology-centric.

Currently, masked language modeling (e.g., BERT) is the prime choice to learn contextualized representations. Due to the pervasiveness, it naturally raises an interesting question: how do masked language models (MLMs) learn contextual representations? In this work, we analyze the learning dynamics of MLMs and find that it adopts sampled embeddings as anchors to estimate and inject contextual semantics to representations, which limits the efficiency and effectiveness of MLMs. To address these problems, we propose TACO, a simple yet effective representation learning approach to directly model global semantics. To be specific, TACO extracts and aligns contextual semantics hidden in contextualized representations to encourage models to attend global semantics when generating contextualized representations. Experiments on the GLUE benchmark show that TACO achieves up to 5x speedup and up to 1.2 points average improvement over MLM.

We present Knowledge Distillation with Meta Learning (MetaDistil), a simple yet effective alternative to traditional knowledge distillation (KD) methods where the teacher model is fixed during training. We show the teacher network can learn to better transfer knowledge to the student network (i.e., learning to teach) with the feedback from the performance of the distilled student network in a meta learning framework. Moreover, we introduce a pilot update mechanism to improve the alignment between the inner-learner and meta-learner in meta learning algorithms that focus on an improved inner-learner. Experiments on various benchmarks show that MetaDistil can yield significant improvements compared with traditional KD algorithms and is less sensitive to the choice of different student capacity and hyperparameters, facilitating the use of KD on different tasks and models.

Intermediate-task transfer can benefit a wide range of NLP tasks with properly selected source datasets. However, it is computationally infeasible to experiment with all intermediate transfer combinations, making choosing a useful source task a challenging problem. In this paper, we anticipate that task-specific parameters updated in parameter-efficient tuning methods are likely to encode task-specific information. Therefore, such parameters can be predictive for inter-task transferability. Thus, we propose to exploit these efficiently tuned parameters as off-the-shelf task embeddings for the efficient selection of source datasets for intermediate-task transfer. We experiment with 11 text classification tasks and 11 question answering tasks. Experimental results show that our approach consistently outperforms existing inter-task transferability prediction methods while being conceptually simple and computationally efficient. Our analysis also reveals that the ability of efficiently tuned parameters on transferability prediction is disentangled with their in-task performance. This allows us to use parameters from early checkpoints as task embeddings to further improve efficiency.

Cant is important for understanding advertising, comedies and dog-whistle politics. However, computational research on cant is hindered by a lack of available datasets. In this paper, we propose a large and diverse Chinese dataset for creating and understanding cant from a computational linguistics perspective. We formulate a task for cant understanding and provide both quantitative and qualitative analysis for tested word embedding similarity and pretrained language models. Experiments suggest that such a task requires deep language understanding, common sense, and world knowledge and thus can be a good testbed for pretrained language models and help models perform better on other tasks.

In this paper, we propose Inverse Adversarial Training (IAT) algorithm for training neural dialogue systems to avoid generic responses and model dialogue history better. In contrast to standard adversarial training algorithms, IAT encourages the model to be sensitive to the perturbation in the dialogue history and therefore learning from perturbations. By giving higher rewards for responses whose output probability reduces more significantly when dialogue history is perturbed, the model is encouraged to generate more diverse and consistent responses. By penalizing the model when generating the same response given perturbed dialogue history, the model is forced to better capture dialogue history and generate more informative responses. Experimental results on two benchmark datasets show that our approach can better model dialogue history and generate more diverse and consistent responses. In addition, we point out a problem of the widely used maximum mutual information (MMI) based methods for improving the diversity of dialogue response generation models and demonstrate it empirically.

In this paper, we propose Sequence Span Rewriting (SSR), a self-supervised task for sequence-to-sequence (Seq2Seq) pre-training. SSR learns to refine the machine-generated imperfect text spans into ground truth text. SSR provides more fine-grained and informative supervision in addition to the original text-infilling objective. Compared to the prevalent text infilling objectives for Seq2Seq pre-training, SSR is naturally more consistent with many downstream generation tasks that require sentence rewriting (e.g., text summarization, question generation, grammatical error correction, and paraphrase generation). We conduct extensive experiments by using SSR to improve the typical Seq2Seq pre-trained model T5 in a continual pre-training setting and show substantial improvements over T5 on various natural language generation tasks.

Recent studies on compression of pretrained language models (e.g., BERT) usually use preserved accuracy as the metric for evaluation. In this paper, we propose two new metrics, label loyalty and probability loyalty that measure how closely a compressed model (i.e., student) mimics the original model (i.e., teacher). We also explore the effect of compression with regard to robustness under adversarial attacks. We benchmark quantization, pruning, knowledge distillation and progressive module replacing with loyalty and robustness. By combining multiple compression techniques, we provide a practical strategy to achieve better accuracy, loyalty and robustness.

We propose a novel data synthesis method to generate diverse error-corrected sentence pairs for improving grammatical error correction, which is based on a pair of machine translation models (e.g., Chinese to English) of different qualities (i.e., poor and good). The poor translation model can resemble the ESL (English as a second language) learner and tends to generate translations of low quality in terms of fluency and grammaticality, while the good translation model generally generates fluent and grammatically correct translations. With the pair of translation models, we can generate unlimited numbers of poor to good English sentence pairs from text in the source language (e.g., Chinese) of the translators. Our approach can generate various error-corrected patterns and nicely complement the other data synthesis approaches for GEC. Experimental results demonstrate the data generated by our approach can effectively help a GEC model to improve the performance and achieve the state-of-the-art single-model performance in BEA-19 and CoNLL-14 benchmark datasets.

Local sequence transduction (LST) tasks are sequence transduction tasks where there exists massive overlapping between the source and target sequences, such as grammatical error correction and spell or OCR correction. Motivated by this characteristic of LST tasks, we propose Pseudo-Bidirectional Decoding (PBD), a simple but versatile approach for LST tasks. PBD copies the representation of source tokens to the decoder as pseudo future context that enables the decoder self-attention to attends to its bi-directional context. In addition, the bidirectional decoding scheme and the characteristic of LST tasks motivate us to share the encoder and the decoder of LST models. Our approach provides right-side context information for the decoder, reduces the number of parameters by half, and provides good regularization effects. Experimental results on several benchmark datasets show that our approach consistently improves the performance of standard seq2seq models on LST tasks.

Recently, large-scale pre-trained language models have demonstrated impressive performance on several commonsense-reasoning benchmark datasets. However, building machines with commonsense to compose realistically plausible sentences remains challenging. In this paper, we present a constrained text generation task, CommonGen associated with a benchmark dataset, to explicitly test machines for the ability of generative commonsense reasoning. Given a set of common concepts (e.g., dog, frisbee, catch, throw); the task is to generate a coherent sentence describing an everyday scenario using these concepts (e.g., “a man throws a frisbee and his dog catches it”). The CommonGen task is challenging because it inherently requires 1) relational reasoning with background commonsense knowledge and 2) compositional generalization ability to work on unseen concept combinations. Our dataset, constructed through a combination of crowdsourced and existing caption corpora, consists of 77k commonsense descriptions over 35k unique concept-sets. Experiments show that there is a large gap between state-of-the-art text generation models (e.g., T5) and human performance (31.6% v.s. 63.5% in SPICE metric). Furthermore, we demonstrate that the learned generative commonsense reasoning capability can be transferred to improve downstream tasks such as CommonsenseQA (76.9% to 78.4 in dev accuracy) by generating additional context.

We introduce DropHead, a structured dropout method specifically designed for regularizing the multi-head attention mechanism which is a key component of transformer. In contrast to the conventional dropout mechanism which randomly drops units or connections, DropHead drops entire attention heads during training to prevent the multi-head attention model from being dominated by a small portion of attention heads. It can help reduce the risk of overfitting and allow the models to better benefit from the multi-head attention. Given the interaction between multi-headedness and training dynamics, we further propose a novel dropout rate scheduler to adjust the dropout rate of DropHead throughout training, which results in a better regularization effect. Experimental results demonstrate that our proposed approach can improve transformer models by 0.9 BLEU score on WMT14 En-De translation task and around 1.0 accuracy for various text classification tasks.

Fact verification models have enjoyed a fast advancement in the last two years with the development of pre-trained language models like BERT and the release of large scale datasets such as FEVER. However, the challenging problem of fake news detection has not benefited from the improvement of fact verification models, which is closely related to fake news detection. In this paper, we propose a simple yet effective approach to connect the dots between fact verification and fake news detection. Our approach first employs a text summarization model pre-trained on news corpora to summarize the long news article into a short claim. Then we use a fact verification model pre-trained on the FEVER dataset to detect whether the input news article is real or fake. Our approach makes use of the recent success of fact verification models and enables zero-shot fake news detection, alleviating the need of large scale training data to train fake news detection models. Experimental results on FakenewsNet, a benchmark dataset for fake news detection, demonstrate the effectiveness of our proposed approach.

In this paper, we propose a novel model compression approach to effectively compress BERT by progressive module replacing. Our approach first divides the original BERT into several modules and builds their compact substitutes. Then, we randomly replace the original modules with their substitutes to train the compact modules to mimic the behavior of the original modules. We progressively increase the probability of replacement through the training. In this way, our approach brings a deeper level of interaction between the original and compact models. Compared to the previous knowledge distillation approaches for BERT compression, our approach does not introduce any additional loss function. Our approach outperforms existing knowledge distillation approaches on GLUE benchmark, showing a new perspective of model compression.

Previous studies on lexical substitution tend to obtain substitute candidates by finding the target word’s synonyms from lexical resources (e.g., WordNet) and then rank the candidates based on its contexts. These approaches have two limitations: (1) They are likely to overlook good substitute candidates that are not the synonyms of the target words in the lexical resources; (2) They fail to take into account the substitution’s influence on the global context of the sentence. To address these issues, we propose an end-to-end BERT-based lexical substitution approach which can propose and validate substitute candidates without using any annotated data or manually curated resources. Our approach first applies dropout to the target word’s embedding for partially masking the word, allowing BERT to take balanced consideration of the target word’s semantics and contexts for proposing substitute candidates, and then validates the candidates based on their substitution’s influence on the global contextualized representation of the sentence. Experiments show our approach performs well in both proposing and ranking substitute candidates, achieving the state-of-the-art results in both LS07 and LS14 benchmarks.