Ting-Rui Chiang

Jessica Huynh, Ting-Rui Chiang, Jeffrey Bigham et al.

Dialog system developers need high-quality data to train, fine-tune and assess their systems. They often use crowdsourcing for this since it provides large quantities of data from many workers. However, the data may not be of sufficiently good quality. This can be due to the way that the requester presents a task and how they interact with the workers. This paper introduces DialCrowd 2.0 to help requesters obtain higher quality data by, for example, presenting tasks more clearly and facilitating effective communication with workers. DialCrowd 2.0 guides developers in creating improved Human Intelligence Tasks (HITs) and is directly applicable to the workflows used currently by developers and researchers.

Ting-Rui Chiang, Xinyan Velocity Yu, Joshua Robinson et al. · amazon-science, uw

Augmenting a language model (LM) with $k$-nearest neighbors ($k$NN) retrieval on its training data alone can decrease its perplexity, though the underlying reasons for this remain elusive. In this work, we rule out one previously posited possibility -- the "softmax bottleneck." We then create a new dataset to evaluate LM generalization ability in the setting where training data contains additional information that is not causally relevant. This task is challenging even for GPT-3.5 Turbo. We show that, for both GPT-2 and Mistral 7B, $k$NN retrieval augmentation consistently improves performance in this setting. Finally, to make $k$NN retrieval more accessible, we propose using a multi-layer perceptron model that maps datastore keys to values as a drop-in replacement for traditional retrieval. This reduces storage costs by over 25x.

Ting-Rui Chiang, Dani Yogatama

We analyze the masked language modeling pretraining objective function from the perspective of the distributional hypothesis. We investigate whether better sample efficiency and the better generalization capability of models pretrained with masked language modeling can be attributed to the semantic similarity encoded in the pretraining data's distributional property. Via a synthetic dataset, our analysis suggests that distributional property indeed leads to the better sample efficiency of pretrained masked language models, but does not fully explain the generalization capability. We also conduct analyses over two real-world datasets and demonstrate that the distributional property does not explain the generalization ability of pretrained natural language models either. Our results illustrate our limited understanding of model pretraining and provide future research directions.

Ting-Rui Chiang, Yi-Ting Yeh

Dialogue state tracking models play an important role in a task-oriented dialogue system. However, most of them model the slot types conditionally independently given the input. We discover that it may cause the model to be confused by slot types that share the same data type. To mitigate this issue, we propose TripPy-MRF and TripPy-LSTM that models the slots jointly. Our results show that they are able to alleviate the confusion mentioned above, and they push the state-of-the-art on dataset MultiWoZ 2.1 from 58.7 to 61.3. Our implementation is available at https://github.com/CTinRay/Trippy-Joint.

Ting-Rui Chiang, Hao-Tong Ye, Yun-Nung Chen

With a lot of work about context-free question answering systems, there is an emerging trend of conversational question answering models in the natural language processing field. Thanks to the recently collected datasets, including QuAC and CoQA, there has been more work on conversational question answering, and recent work has achieved competitive performance on both datasets. However, to best of our knowledge, two important questions for conversational comprehension research have not been well studied: 1) How well can the benchmark dataset reflect models' content understanding? 2) Do the models well utilize the conversation content when answering questions? To investigate these questions, we design different training settings, testing settings, as well as an attack to verify the models' capability of content understanding on QuAC and CoQA. The experimental results indicate some potential hazards in the benchmark datasets, QuAC and CoQA, for conversational comprehension research. Our analysis also sheds light on both what models may learn and how datasets may bias the models. With deep investigation of the task, it is believed that this work can benefit the future progress of conversation comprehension. The source code is available at https://github.com/MiuLab/CQA-Study.

Ting-Rui Chiang, Dani Yogatama

Many existing theoretical analyses of in-context learning for natural language processing are based on latent variable models that leaves gaps between theory and practice. We aim to close these gaps by proposing a theoretical framework, the Pelican Soup Framework. In this framework, we introduce (1) the notion of a common sense knowledge base, (2) a general formalism for natural language classification tasks, and the notion of (3) meaning association. Under this framework, we can establish a $\mathcal{O}(1/T)$ loss bound for in-context learning, where $T$ is the number of example-label pairs in the demonstration. Compared with previous works, our bound reflects the effect of the choice of verbalizers and the effect of instruction tuning. An additional notion of \textit{atom concepts} makes our framework possible to explain the generalization to tasks unseen in the language model training data. Finally, we propose a toy setup, Calcutec, and a digit addition task that mimics types of distribution shifts a model needs to overcome to perform in-context learning. We also experiment with GPT2-Large on real-world NLP tasks. Our empirical results demonstrate the efficacy of our framework to explain in-context learning.

Ting-Rui Chiang, Dani Yogatama

The Rotary Position Embedding (RoPE) is widely used in the attention heads of many large language models (LLM). It rotates dimensions in the query and the key vectors by different angles according to their positions in the input sequence. For long context modeling, the range of positions may vary a lot, and thus RoPE rotates some dimensions by a great range of angles. We hypothesize that the wide range of rotation angles may prevent LLMs from utilizing those dimensions. To validate this hypothesis, we present a controlled experiment showing that applying RoPE causes low utility of certain dimensions. Our analyses on three LLMs also indicate that these dimensions do not help LLMs do long-context question answering.

Ting-Rui Chiang, Joshua Robinson, Xinyan Velocity Yu et al.

The ability to locate an object in an image according to natural language instructions is crucial for many real-world applications. In this work we propose LocateBench, a high-quality benchmark dedicated to evaluating this ability. We experiment with multiple prompting approaches, and measure the accuracy of several large vision language models. We find that even the accuracy of the strongest model, GPT-4o, lags behind human accuracy by more than 10%.

Ting-Rui Chiang, Yi-Pei Chen, Yi-Ting Yeh et al.

While multilingual training is now an essential ingredient in machine translation (MT) systems, recent work has demonstrated that it has different effects in different multilingual settings, such as many-to-one, one-to-many, and many-to-many learning. These training settings expose the encoder and the decoder in a machine translation model with different data distributions. In this paper, we examine how different varieties of multilingual training contribute to learning these two components of the MT model. Specifically, we compare bilingual models with encoders and/or decoders initialized by multilingual training. We show that multilingual training is beneficial to encoders in general, while it only benefits decoders for low-resource languages (LRLs). We further find the important attention heads for each language pair and compare their correlations during inference. Our analysis sheds light on how multilingual translation models work and enables us to propose methods to improve performance by training with highly related languages. Our many-to-one models for high-resource languages and one-to-many models for LRL outperform the best results reported by Aharoni et al. (2019)

Ting-Rui Chiang, Yi-Ting Yeh, Ta-Chung Chi et al.

ALFRED is a recently proposed benchmark that requires a model to complete tasks in simulated house environments specified by instructions in natural language. We hypothesize that key to success is accurately aligning the text modality with visual inputs. Motivated by this, we inspect how well existing models can align these modalities using our proposed intrinsic metric, boundary adherence score (BAS). The results show the previous models are indeed failing to perform proper alignment. To address this issue, we introduce approaches aimed at improving model alignment and demonstrate how improved alignment, improves end task performance.

Ting-Rui Chiang

Despite the success of pretrained masked language models (MLM), why MLM pretraining is useful is still a qeustion not fully answered. In this work we theoretically and empirically show that MLM pretraining makes models robust to lexicon-level spurious features, partly answer the question. We theoretically show that, when we can model the distribution of a spurious feature $Π$ conditioned on the context, then (1) $Π$ is at least as informative as the spurious feature, and (2) learning from $Π$ is at least as simple as learning from the spurious feature. Therefore, MLM pretraining rescues the model from the simplicity bias caused by the spurious feature. We also explore the efficacy of MLM pretraing in causal settings. Finally we close the gap between our theories and the real world practices by conducting experiments on the hate speech detection and the name entity recognition tasks.

Ting-Rui Chiang, Yun-Nung Chen

This work focuses on relating two mysteries in neural-based text generation: exposure bias, and text degeneration. Despite the long time since exposure bias was mentioned and the numerous studies for its remedy, to our knowledge, its impact on text generation has not yet been verified. Text degeneration is a problem that the widely-used pre-trained language model GPT-2 was recently found to suffer from (Holtzman et al., 2020). Motivated by the unknown causation of the text degeneration, in this paper we attempt to relate these two mysteries. Specifically, we first qualitatively quantitatively identify mistakes made before text degeneration occurs. Then we investigate the significance of the mistakes by inspecting the hidden states in GPT-2. Our results show that text degeneration is likely to be partly caused by exposure bias. We also study the self-reinforcing mechanism of text degeneration, explaining why the mistakes amplify. In sum, our study provides a more concrete foundation for further investigation on exposure bias and text degeneration problems.

Ting-Rui Chiang, Yun-Nung Chen

The huge size of the widely used BERT family models has led to recent efforts about model distillation. The main goal of distillation is to create a task-agnostic pre-trained model that can be fine-tuned on downstream tasks without fine-tuning its full-sized version. Despite the progress of distillation, to what degree and for what reason a task-agnostic model can be created from distillation has not been well studied. Also, the mechanisms behind transfer learning of those BERT models are not well investigated either. Therefore, this work focuses on analyzing the acceptable deduction when distillation for guiding the future distillation procedure. Specifically, we first inspect the prunability of the Transformer heads in RoBERTa and ALBERT using their head importance estimation proposed by Michel et al. (2019), and then check the coherence of the important heads between the pre-trained task and downstream tasks. Hence, the acceptable deduction of performance on the pre-trained task when distilling a model can be derived from the results, and we further compare the behavior of the pruned model before and after fine-tuning. Our studies provide guidance for future directions about BERT family model distillation.

Ting-Rui Chiang, Chao-Wei Huang, Shang-Yu Su et al.

With the increasing research interest in dialogue response generation, there is an emerging branch formulating this task as selecting next sentences, where given the partial dialogue contexts, the goal is to determine the most probable next sentence. Following the recent success of the Transformer model, this paper proposes (1) a new variant of attention mechanism based on multi-head attention, called highway attention, and (2) a recurrent model based on transformer and the proposed highway attention, so-called Highway Recurrent Transformer. Experiments on the response selection task in the seventh Dialog System Technology Challenge (DSTC7) show the capability of the proposed model of modeling both utterance-level and dialogue-level information; the effectiveness of each module is further analyzed as well.

Chao-Wei Huang, Ting-Rui Chiang, Shang-Yu Su et al.

The response selection has been an emerging research topic due to the growing interest in dialogue modeling, where the goal of the task is to select an appropriate response for continuing dialogues. To further push the end-to-end dialogue model toward real-world scenarios, the seventh Dialog System Technology Challenge (DSTC7) proposed a challenging track based on real chatlog datasets. The competition focuses on dialogue modeling with several advanced characteristics: (1) natural language diversity, (2) capability of precisely selecting a proper response from a large set of candidates or the scenario without any correct answer, and (3) knowledge grounding. This paper introduces recurrent attention pooling networks (RAP-Net), a novel framework for response selection, which can well estimate the relevance between the dialogue contexts and the candidates. The proposed RAP-Net is shown to be effective and can be generalized across different datasets and settings in the DSTC7 experiments.

Ting-Rui Chiang, Yun-Nung Chen

Solving math word problems is a challenging task that requires accurate natural language understanding to bridge natural language texts and math expressions. Motivated by the intuition about how human generates the equations given the problem texts, this paper presents a neural approach to automatically solve math word problems by operating symbols according to their semantic meanings in texts. This paper views the process of generating equation as a bridge between the semantic world and the symbolic world, where the proposed neural math solver is based on an encoder-decoder framework. In the proposed model, the encoder is designed to understand the semantics of problems, and the decoder focuses on tracking semantic meanings of the generated symbols and then deciding which symbol to generate next. The preliminary experiments are conducted in a dataset Math23K, and our model significantly outperforms both the state-of-the-art single model and the best non-retrieval-based model over about 10% accuracy, demonstrating the effectiveness of bridging the symbolic and semantic worlds from math word problems.