Lin Pan

Shuaichen Chang, Jun Wang, Mingwen Dong et al. · amazon-science, ibm-research

Neural text-to-SQL models have achieved remarkable performance in translating natural language questions into SQL queries. However, recent studies reveal that text-to-SQL models are vulnerable to task-specific perturbations. Previous curated robustness test sets usually focus on individual phenomena. In this paper, we propose a comprehensive robustness benchmark based on Spider, a cross-domain text-to-SQL benchmark, to diagnose the model robustness. We design 17 perturbations on databases, natural language questions, and SQL queries to measure the robustness from different angles. In order to collect more diversified natural question perturbations, we utilize large pretrained language models (PLMs) to simulate human behaviors in creating natural questions. We conduct a diagnostic study of the state-of-the-art models on the robustness set. Experimental results reveal that even the most robust model suffers from a 14.0% performance drop overall and a 50.7% performance drop on the most challenging perturbation. We also present a breakdown analysis regarding text-to-SQL model designs and provide insights for improving model robustness.

Yiyun Zhao, Jiarong Jiang, Yiqun Hu et al. · amazon-science, ibm-research

Recently, there has been increasing interest in synthesizing data to improve downstream text-to-SQL tasks. In this paper, we first examined the existing synthesized datasets and discovered that state-of-the-art text-to-SQL algorithms did not further improve on popular benchmarks when trained with augmented synthetic data. We observed two shortcomings: illogical synthetic SQL queries from independent column sampling and arbitrary table joins. To address these issues, we propose a novel synthesis framework that incorporates key relationships from schema, imposes strong typing, and conducts schema-distance-weighted column sampling. We also adopt an intermediate representation (IR) for the SQL-to-text task to further improve the quality of the generated natural language questions. When existing powerful semantic parsers are pre-finetuned on our high-quality synthesized data, our experiments show that these models have significant accuracy boosts on popular benchmarks, including new state-of-the-art performance on Spider.

Yuanyuan Peng, Lin Pan, Pengpeng Luan et al.

Automatic segmentation of curvilinear objects in medical images plays an important role in the diagnosis and evaluation of human diseases, yet it is a challenging uncertainty in the complex segmentation tasks due to different issues such as various image appearances, low contrast between curvilinear objects and their surrounding backgrounds, thin and uneven curvilinear structures, and improper background illumination conditions. To overcome these challenges, we present a unique curvilinear structure segmentation framework based on an oriented derivative of stick (ODoS) filter and a deep learning network for curvilinear object segmentation in medical images. Currently, a large number of deep learning models emphasize developing deep architectures and ignore capturing the structural features of curvilinear objects, which may lead to unsatisfactory results. Consequently, a new approach that incorporates an ODoS filter as part of a deep learning network is presented to improve the spatial attention of curvilinear objects. Specifically, the input image is transfered into four-channel image constructed by the ODoS filter. In which, the original image is considered the principal part to describe various image appearance and complex background illumination conditions, a multi-step strategy is used to enhance the contrast between curvilinear objects and their surrounding backgrounds, and a vector field is applied to discriminate thin and uneven curvilinear structures. Subsequently, a deep learning framework is employed to extract various structural features for curvilinear object segmentation in medical images. The performance of the computational model is validated in experiments conducted on the publicly available DRIVE, STARE and CHASEDB1 datasets. The experimental results indicate that the presented model yields surprising results compared with those of some state-of-the-art methods.

Mingwen Dong, Nischal Ashok Kumar, Yiqun Hu et al. · amazon-science

Previous text-to-SQL datasets and systems have primarily focused on user questions with clear intentions that can be answered. However, real user questions can often be ambiguous with multiple interpretations or unanswerable due to a lack of relevant data. In this work, we construct a practical conversational text-to-SQL dataset called PRACTIQ, consisting of ambiguous and unanswerable questions inspired by real-world user questions. We first identified four categories of ambiguous questions and four categories of unanswerable questions by studying existing text-to-SQL datasets. Then, we generate conversations with four turns: the initial user question, an assistant response seeking clarification, the user's clarification, and the assistant's clarified SQL response with the natural language explanation of the execution results. For some ambiguous queries, we also directly generate helpful SQL responses, that consider multiple aspects of ambiguity, instead of requesting user clarification. To benchmark the performance on ambiguous, unanswerable, and answerable questions, we implemented large language model (LLM)-based baselines using various LLMs. Our approach involves two steps: question category classification and clarification SQL prediction. Our experiments reveal that state-of-the-art systems struggle to handle ambiguous and unanswerable questions effectively. We will release our code for data generation and experiments on GitHub.

Wuwei Lan, Zhiguo Wang, Anuj Chauhan et al.

A practical text-to-SQL system should generalize well on a wide variety of natural language questions, unseen database schemas, and novel SQL query structures. To comprehensively evaluate text-to-SQL systems, we introduce a UNIfied benchmark for Text-to-SQL Evaluation (UNITE). It is composed of publicly available text-to-SQL datasets, containing natural language questions from more than 12 domains, SQL queries from more than 3.9K patterns, and 29K databases. Compared to the widely used Spider benchmark, we introduce $\sim$120K additional examples and a threefold increase in SQL patterns, such as comparative and boolean questions. We conduct a systematic study of six state-of-the-art (SOTA) text-to-SQL parsers on our new benchmark and show that: 1) Codex performs surprisingly well on out-of-domain datasets; 2) specially designed decoding methods (e.g. constrained beam search) can improve performance for both in-domain and out-of-domain settings; 3) explicitly modeling the relationship between questions and schemas further improves the Seq2Seq models. More importantly, our benchmark presents key challenges towards compositional generalization and robustness issues -- which these SOTA models cannot address well. Our code and data processing script are available at https://github.com/awslabs/unified-text2sql-benchmark

Jiaqing Yuan, Lin Pan, Chung-Wei Hang et al. · amazon-science

Large language models (LLMs) have shown remarkable performance on a variety of NLP tasks, and are being rapidly adopted in a wide range of use cases. It is therefore of vital importance to holistically evaluate the factuality of their generated outputs, as hallucinations remain a challenging issue. In this work, we focus on assessing LLMs' ability to recall factual knowledge learned from pretraining, and the factors that affect this ability. To that end, we construct FACT-BENCH, a representative benchmark covering 20 domains, 134 property types, 3 answer types, and different knowledge popularity levels. We benchmark 31 models from 10 model families and provide a holistic assessment of their strengths and weaknesses. We observe that instruction-tuning hurts knowledge recall, as pretraining-only models consistently outperform their instruction-tuned counterparts, and positive effects of model scaling, as larger models outperform smaller ones for all model families. However, the best performance from GPT-4 still represents a large gap with the upper-bound. We additionally study the role of in-context exemplars using counterfactual demonstrations, which lead to significant degradation of factual knowledge recall for large models. By further decoupling model known and unknown knowledge, we find the degradation is attributed to exemplars that contradict a model's known knowledge, as well as the number of such exemplars. Lastly, we fine-tune LLaMA-7B in different settings of known and unknown knowledge. In particular, fine-tuning on a model's known knowledge is beneficial, and consistently outperforms fine-tuning on unknown and mixed knowledge. We will make our benchmark publicly available.

Lin Pan, Chung-Wei Hang, Avirup Sil et al.

We propose a simple and general method to regularize the fine-tuning of Transformer-based encoders for text classification tasks. Specifically, during fine-tuning we generate adversarial examples by perturbing the word embeddings of the model and perform contrastive learning on clean and adversarial examples in order to teach the model to learn noise-invariant representations. By training on both clean and adversarial examples along with the additional contrastive objective, we observe consistent improvement over standard fine-tuning on clean examples. On several GLUE benchmark tasks, our fine-tuned BERT Large model outperforms BERT Large baseline by 1.7% on average, and our fine-tuned RoBERTa Large improves over RoBERTa Large baseline by 1.3%. We additionally validate our method in different domains using three intent classification datasets, where our fine-tuned RoBERTa Large outperforms RoBERTa Large baseline by 1-2% on average.

Lin Pan, Yaoyong Zheng, Liqin Huang et al.

With the development of medical computer-aided diagnostic systems, pulmonary artery-vein(A/V) separation plays a crucial role in assisting doctors in preoperative planning for lung cancer surgery. However, distinguishing arterial from venous irrigation in chest CT images remains a challenge due to the similarity and complex structure of the arteries and veins. We propose a novel method for automatic separation of pulmonary arteries and veins from chest CT images. The method consists of three parts. First, global connection information and local feature information are used to construct a complete topological tree and ensure the continuity of vessel reconstruction. Second, the Twin-Pipe network proposed can automatically learn the differences between arteries and veins at different levels to reduce classification errors caused by changes in terminal vessel characteristics. Finally, the topology optimizer considers interbranch and intrabranch topological relationships to maintain spatial consistency to avoid the misclassification of A/V irrigations. We validate the performance of the method on chest CT images. Compared with manual classification, the proposed method achieves an average accuracy of 96.2% on noncontrast chest CT. In addition, the method has been proven to have good generalization, that is, the accuracies of 93.8% and 94.8% are obtained for CT scans from other devices and other modes, respectively. The result of pulmonary artery-vein obtained by the proposed method can provide better assistance for preoperative planning of lung cancer surgery.

Jinquan Guo, Rongda Fu, Lin Pan et al.

Automatic airway segmentation from chest computed tomography (CT) scans plays an important role in pulmonary disease diagnosis and computer-assisted therapy. However, low contrast at peripheral branches and complex tree-like structures remain as two mainly challenges for airway segmentation. Recent research has illustrated that deep learning methods perform well in segmentation tasks. Motivated by these works, a coarse-to-fine segmentation framework is proposed to obtain a complete airway tree. Our framework segments the overall airway and small branches via the multi-information fusion convolution neural network (Mif-CNN) and the CNN-based region growing, respectively. In Mif-CNN, atrous spatial pyramid pooling (ASPP) is integrated into a u-shaped network, and it can expend the receptive field and capture multi-scale information. Meanwhile, boundary and location information are incorporated into semantic information. These information are fused to help Mif-CNN utilize additional context knowledge and useful features. To improve the performance of the segmentation result, the CNN-based region growing method is designed to focus on obtaining small branches. A voxel classification network (VCN), which can entirely capture the rich information around each voxel, is applied to classify the voxels into airway and non-airway. In addition, a shape reconstruction method is used to refine the airway tree.

Shaohua Zheng, Zhiqiang Shen, Chenhao Peia et al.

Background and Objective:Computer-aided diagnosis (CAD) systems promote diagnosis effectiveness and alleviate pressure of radiologists. A CAD system for lung cancer diagnosis includes nodule candidate detection and nodule malignancy evaluation. Recently, deep learning-based pulmonary nodule detection has reached satisfactory performance ready for clinical application. However, deep learning-based nodule malignancy evaluation depends on heuristic inference from low-dose computed tomography volume to malignant probability, which lacks clinical cognition. Methods:In this paper, we propose a joint radiology analysis and malignancy evaluation network (R2MNet) to evaluate the pulmonary nodule malignancy via radiology characteristics analysis. Radiological features are extracted as channel descriptor to highlight specific regions of the input volume that are critical for nodule malignancy evaluation. In addition, for model explanations, we propose channel-dependent activation mapping to visualize the features and shed light on the decision process of deep neural network. Results:Experimental results on the LIDC-IDRI dataset demonstrate that the proposed method achieved area under curve of 96.27% on nodule radiology analysis and AUC of 97.52% on nodule malignancy evaluation. In addition, explanations of CDAM features proved that the shape and density of nodule regions were two critical factors that influence a nodule to be inferred as malignant, which conforms with the diagnosis cognition of experienced radiologists. Conclusion:Incorporating radiology analysis with nodule malignant evaluation, the network inference process conforms to the diagnostic procedure of radiologists and increases the confidence of evaluation results. Besides, model interpretation with CDAM features shed light on the regions which DNNs focus on when they estimate nodule malignancy probabilities.

Mihaela Bornea, Lin Pan, Sara Rosenthal et al.

Prior work on multilingual question answering has mostly focused on using large multilingual pre-trained language models (LM) to perform zero-shot language-wise learning: train a QA model on English and test on other languages. In this work, we explore strategies that improve cross-lingual transfer by bringing the multilingual embeddings closer in the semantic space. Our first strategy augments the original English training data with machine translation-generated data. This results in a corpus of multilingual silver-labeled QA pairs that is 14 times larger than the original training set. In addition, we propose two novel strategies, language adversarial training and language arbitration framework, which significantly improve the (zero-resource) cross-lingual transfer performance and result in LM embeddings that are less language-variant. Empirically, we show that the proposed models outperform the previous zero-shot baseline on the recently introduced multilingual MLQA and TyDiQA datasets.

Haode Qi, Lin Pan, Atin Sood et al.

Intent detection is a key component of modern goal-oriented dialog systems that accomplish a user task by predicting the intent of users' text input. There are three primary challenges in designing robust and accurate intent detection models. First, typical intent detection models require a large amount of labeled data to achieve high accuracy. Unfortunately, in practical scenarios it is more common to find small, unbalanced, and noisy datasets. Secondly, even with large training data, the intent detection models can see a different distribution of test data when being deployed in the real world, leading to poor accuracy. Finally, a practical intent detection model must be computationally efficient in both training and single query inference so that it can be used continuously and re-trained frequently. We benchmark intent detection methods on a variety of datasets. Our results show that Watson Assistant's intent detection model outperforms other commercial solutions and is comparable to large pretrained language models while requiring only a fraction of computational resources and training data. Watson Assistant demonstrates a higher degree of robustness when the training and test distributions differ.

Lin Pan, Chung-Wei Hang, Haode Qi et al.

We propose a simple method to align multilingual contextual embeddings as a post-pretraining step for improved zero-shot cross-lingual transferability of the pretrained models. Using parallel data, our method aligns embeddings on the word level through the recently proposed Translation Language Modeling objective as well as on the sentence level via contrastive learning and random input shuffling. We also perform sentence-level code-switching with English when finetuning on downstream tasks. On XNLI, our best model (initialized from mBERT) improves over mBERT by 4.7% in the zero-shot setting and achieves comparable result to XLM for translate-train while using less than 18% of the same parallel data and 31% less model parameters. On MLQA, our model outperforms XLM-R_Base that has 57% more parameters than ours.

Vittorio Castelli, Rishav Chakravarti, Saswati Dana et al.

We introduce TechQA, a domain-adaptation question answering dataset for the technical support domain. The TechQA corpus highlights two real-world issues from the automated customer support domain. First, it contains actual questions posed by users on a technical forum, rather than questions generated specifically for a competition or a task. Second, it has a real-world size -- 600 training, 310 dev, and 490 evaluation question/answer pairs -- thus reflecting the cost of creating large labeled datasets with actual data. Consequently, TechQA is meant to stimulate research in domain adaptation rather than being a resource to build QA systems from scratch. The dataset was obtained by crawling the IBM Developer and IBM DeveloperWorks forums for questions with accepted answers that appear in a published IBM Technote---a technical document that addresses a specific technical issue. We also release a collection of the 801,998 publicly available Technotes as of April 4, 2019 as a companion resource that might be used for pretraining, to learn representations of the IT domain language.

Anthony Ferritto, Lin Pan, Rishav Chakravarti et al.

Many of the top question answering systems today utilize ensembling to improve their performance on tasks such as the Stanford Question Answering Dataset (SQuAD) and Natural Questions (NQ) challenges. Unfortunately most of these systems do not publish their ensembling strategies used in their leaderboard submissions. In this work, we investigate a number of ensembling techniques and demonstrate a strategy which improves our F1 score for short answers on the dev set for NQ by 2.3 F1 points over our single model (which outperforms the previous SOTA by 1.9 F1 points).

Lin Pan, Rishav Chakravarti, Anthony Ferritto et al.

Existing literature on Question Answering (QA) mostly focuses on algorithmic novelty, data augmentation, or increasingly large pre-trained language models like XLNet and RoBERTa. Additionally, a lot of systems on the QA leaderboards do not have associated research documentation in order to successfully replicate their experiments. In this paper, we outline these algorithmic components such as Attention-over-Attention, coupled with data augmentation and ensembling strategies that have shown to yield state-of-the-art results on benchmark datasets like SQuAD, even achieving super-human performance. Contrary to these prior results, when we evaluate on the recently proposed Natural Questions benchmark dataset, we find that an incredibly simple approach of transfer learning from BERT outperforms the previous state-of-the-art system trained on 4 million more examples than ours by 1.9 F1 points. Adding ensembling strategies further improves that number by 2.3 F1 points.

Michael Glass, Alfio Gliozzo, Rishav Chakravarti et al.

BERT (Bidirectional Encoder Representations from Transformers) and related pre-trained Transformers have provided large gains across many language understanding tasks, achieving a new state-of-the-art (SOTA). BERT is pre-trained on two auxiliary tasks: Masked Language Model and Next Sentence Prediction. In this paper we introduce a new pre-training task inspired by reading comprehension to better align the pre-training from memorization to understanding. Span Selection Pre-Training (SSPT) poses cloze-like training instances, but rather than draw the answer from the model's parameters, it is selected from a relevant passage. We find significant and consistent improvements over both BERT-BASE and BERT-LARGE on multiple reading comprehension (MRC) datasets. Specifically, our proposed model has strong empirical evidence as it obtains SOTA results on Natural Questions, a new benchmark MRC dataset, outperforming BERT-LARGE by 3 F1 points on short answer prediction. We also show significant impact in HotpotQA, improving answer prediction F1 by 4 points and supporting fact prediction F1 by 1 point and outperforming the previous best system. Moreover, we show that our pre-training approach is particularly effective when training data is limited, improving the learning curve by a large amount.

Rishav Chakravarti, Cezar Pendus, Andrzej Sakrajda et al.

This paper introduces a novel orchestration framework, called CFO (COMPUTATION FLOW ORCHESTRATOR), for building, experimenting with, and deploying interactive NLP (Natural Language Processing) and IR (Information Retrieval) systems to production environments. We then demonstrate a question answering system built using this framework which incorporates state-of-the-art BERT based MRC (Machine Reading Comprehension) with IR components to enable end-to-end answer retrieval. Results from the demo system are shown to be high quality in both academic and industry domain specific settings. Finally, we discuss best practices when (pre-)training BERT based MRC models for production systems.

Zhiguo Wang, Yue Zhang, Mo Yu et al.

Self-explaining text categorization requires a classifier to make a prediction along with supporting evidence. A popular type of evidence is sub-sequences extracted from the input text which are sufficient for the classifier to make the prediction. In this work, we define multi-granular ngrams as basic units for explanation, and organize all ngrams into a hierarchical structure, so that shorter ngrams can be reused while computing longer ngrams. We leverage a tree-structured LSTM to learn a context-independent representation for each unit via parameter sharing. Experiments on medical disease classification show that our model is more accurate, efficient and compact than BiLSTM and CNN baselines. More importantly, our model can extract intuitive multi-granular evidence to support its predictions.