Recent research shows that pre-trained language models (PLMs) suffer from "prompt bias" in factual knowledge extraction, i.e., prompts tend to introduce biases toward specific labels. Prompt bias presents a significant challenge in assessing the factual knowledge within PLMs. Therefore, this paper aims to improve the reliability of existing benchmarks by thoroughly investigating and mitigating prompt bias. We show that: 1) all prompts in the experiments exhibit non-negligible bias, with gradient-based prompts like AutoPrompt and OptiPrompt displaying significantly higher levels of bias; 2) prompt bias can amplify benchmark accuracy unreasonably by overfitting the test datasets, especially on imbalanced datasets like LAMA. Based on these findings, we propose a representation-based approach to mitigate the prompt bias during inference time. Specifically, we first estimate the biased representation using prompt-only querying, and then remove it from the model's internal representations to generate the debiased representations, which are used to produce the final debiased outputs. Experiments across various prompts, PLMs, and benchmarks show that our approach can not only correct the overfitted performance caused by prompt bias, but also significantly improve the prompt retrieval capability (up to 10% absolute performance gain). These results indicate that our approach effectively alleviates prompt bias in knowledge evaluation, thereby enhancing the reliability of benchmark assessments. Hopefully, our plug-and-play approach can be a golden standard to strengthen PLMs toward reliable knowledge bases.

This paper introduces a set of English translations for a 123-hour subset of the CallHome Mandarin Chinese data and the HKUST Mandarin Telephone Speech data for the task of speech translation. Paired source-language speech and target-language text is essential for training end-to-end speech translation systems and can provide substantial performance improvements for cascaded systems as well, relative to training on more widely available text data sets. We demonstrate that fine-tuning a general-purpose translation model to our Mandarin-English conversational telephone speech training set improves target-domain BLEU by more than 8 points, highlighting the importance of matched training data.

Phrase-level dense retrieval has shown many appealing characteristics in downstream NLP tasks by leveraging the fine-grained information that phrases offer. In our work, we propose a new task formulation of dense retrieval, cross-lingual contextualized phrase retrieval, which aims to augment cross-lingual applications by addressing polysemy using context information. However, the lack of specific training data and models are the primary challenges to achieve our goal. As a result, we extract pairs of cross-lingual phrases using word alignment information automatically induced from parallel sentences. Subsequently, we train our Cross-lingual Contextualized Phrase Retriever (CCPR) using contrastive learning, which encourages the hidden representations of phrases with similar contexts and semantics to align closely. Comprehensive experiments on both the cross-lingual phrase retrieval task and a downstream task, i.e, machine translation, demonstrate the effectiveness of CCPR. On the phrase retrieval task, CCPR surpasses baselines by a significant margin, achieving a top-1 accuracy that is at least 13 points higher. When utilizing CCPR to augment the large-language-model-based translator, it achieves average gains of 0.7 and 1.5 in BERTScore for translations from X=>En and vice versa, respectively, on WMT16 dataset. Our code and data are available at \url{https://github.com/ghrua/ccpr_release}.

Multilingual pretraining and fine-tuning have remarkably succeeded in various natural language processing tasks. Transferring representations from one language to another is especially crucial for cross-lingual learning. One can expect machine translation objectives to be well suited to fostering such capabilities, as they involve the explicit alignment of semantically equivalent sentences from different languages. This paper investigates the potential benefits of employing machine translation as a continued training objective to enhance language representation learning, bridging multilingual pretraining and cross-lingual applications. We study this question through two lenses: a quantitative evaluation of the performance of existing models and an analysis of their latent representations. Our results show that, contrary to expectations, machine translation as the continued training fails to enhance cross-lingual representation learning in multiple cross-lingual natural language understanding tasks. We conclude that explicit sentence-level alignment in the cross-lingual scenario is detrimental to cross-lingual transfer pretraining, which has important implications for future cross-lingual transfer studies. We furthermore provide evidence through similarity measures and investigation of parameters that this lack of positive influence is due to output separability -- which we argue is of use for machine translation but detrimental elsewhere.

The advent of scalable deep models and large datasets has improved the performance of Neural Machine Translation. Knowledge Distillation (KD) enhances efficiency by transferring knowledge from a teacher model to a more compact student model. However, KD approaches to Transformer architecture often rely on heuristics, particularly when deciding which teacher layers to distill from. In this paper, we introduce the 'Align-to-Distill' (A2D) strategy, designed to address the feature mapping problem by adaptively aligning student attention heads with their teacher counterparts during training. The Attention Alignment Module in A2D performs a dense head-by-head comparison between student and teacher attention heads across layers, turning the combinatorial mapping heuristics into a learning problem. Our experiments show the efficacy of A2D, demonstrating gains of up to +3.61 and +0.63 BLEU points for WMT-2022 De->Dsb and WMT-2014 En->De, respectively, compared to Transformer baselines.

We present a detailed replication study of the BASS framework, an abstractive summarization system based on the notion of Unified Semantic Graphs. Our investigation includes challenges in replicating key components and an ablation study to systematically isolate error sources rooted in replicating novel components. Our findings reveal discrepancies in performance compared to the original work. We highlight the significance of paying careful attention even to reasonably omitted details for replicating advanced frameworks like BASS, and emphasize key practices for writing replicable papers.

Users are dissatisfied with services. Since the service is not tailor-made for a user, it is natural for dissatisfaction to arise. The problem is, that even if users are dissatisfied, they often do not have the means to resolve their dissatisfaction. The user cannot alter the source code of the service, nor can they force the service provider to change. The user has no choice but to remain dissatisfied or quit the service. User-side realization offers proactive solutions to this problem by providing general algorithms to deal with common problems on the user's side. These algorithms run on the user's side and solve the problems without having the service provider change the service itself.

This thesis investigates how natural language understanding and generation with transformer models can benefit from grounding the models with knowledge representations. Currently, the most prevailing paradigm for training language models is through pre-training on abundant raw text data and fine-tuning on downstream tasks. Although language models continue to advance, especially the recent trend of Large Language Models (LLMs) such as ChatGPT, there seem to be limits to what can be achieved with text data alone and it is desirable to study the impact of applying and integrating rich forms of knowledge representation to improve model performance. The most widely used form of knowledge for language modelling is structured knowledge in the form of triples consisting of entities and their relationships, often in English. This thesis explores beyond this conventional approach and aims to address several key questions: Can knowledge of entities extend its benefits beyond entity-centric tasks such as entity linking? How can we faithfully and effectively extract such structured knowledge from raw text, especially noisy web text? How do other types of knowledge, beyond structured knowledge, contribute to improving NLP tasks?

Recent studies on adversarial examples expose vulnerabilities of natural language processing (NLP) models. Existing techniques for generating adversarial examples are typically driven by deterministic hierarchical rules that are agnostic to the optimal adversarial examples, a strategy that often results in adversarial samples with a suboptimal balance between magnitudes of changes and attack successes. To this end, in this research we propose two algorithms, Reversible Jump Attack (RJA) and Metropolis-Hasting Modification Reduction (MMR), to generate highly effective adversarial examples and to improve the imperceptibility of the examples, respectively. RJA utilizes a novel randomization mechanism to enlarge the search space and efficiently adapts to a number of perturbed words for adversarial examples. With these generated adversarial examples, MMR applies the Metropolis-Hasting sampler to enhance the imperceptibility of adversarial examples. Extensive experiments demonstrate that RJA-MMR outperforms current state-of-the-art methods in attack performance, imperceptibility, fluency and grammar correctness.

Automatic measures of similarity between utterances are invaluable for training speech synthesizers, evaluating machine translation, and assessing learner productions. While there exist measures for semantic similarity and prosodic similarity, there are as yet none for pragmatic similarity. To enable the training of such measures, we developed the first collection of human judgments of pragmatic similarity between utterance pairs. Each pair consisting of an utterance extracted from a recorded dialog and a re-enactment of that utterance. Re-enactments were done under various conditions designed to create a variety of degrees of similarity. Each pair was rated on a continuous scale by 6 to 9 judges. The average inter-judge correlation was as high as 0.72 for English and 0.66 for Spanish.