Don't Blame the Annotator: Bias Already Starts in the Annotation Instructions

In recent years, progress in NLU has been driven by benchmarks. These benchmarks are typically collected by crowdsourcing, where annotators write examples based on annotation instructions crafted by dataset creators. In this work, we hypothesize that annotators pick up on patterns in the crowdsourcing instructions, which bias them to write similar examples that are then over-represented in the collected data. We study this form of bias, termed instruction bias, in 14 recent NLU benchmarks, showing that instruction examples often exhibit concrete patterns, which are propagated by crowdworkers to the collected data. This extends previous work (Geva et al., 2019) and raises a new concern of whether we are modeling the dataset creator's instructions, rather than the task. Through a series of experiments, we show that, indeed, instruction bias can lead to overestimation of model performance, and that models struggle to generalize beyond biases originating in the crowdsourcing instructions. We further analyze the influence of instruction bias in terms of pattern frequency and model size, and derive concrete recommendations for creating future NLU benchmarks.

Benchmarking Generalization via In-Context Instructions on 1,600+ Language Tasks

How can we measure the generalization of models to a variety of unseen tasks when provided with their language instructions? To facilitate progress in this goal, we introduce Natural-Instructions v2, a collection of 1,600+ diverse language tasks and their expert written instructions. More importantly, the benchmark covers 70+ distinct task types, such as tagging, in-filling, and rewriting. This benchmark is collected with contributions of NLP practitioners in the community and through an iterative peer review process to ensure their quality. This benchmark enables large-scale evaluation of cross-task generalization of the models -- training on a subset of tasks and evaluating on the remaining unseen ones. For instance, we are able to rigorously quantify generalization as a function of various scaling parameters, such as the number of observed tasks, the number of instances, and model sizes. As a by-product of these experiments. we introduce Tk-Instruct, an encoder-decoder Transformer that is trained to follow a variety of in-context instructions (plain language task definitions or k-shot examples) which outperforms existing larger models on our benchmark. We hope this benchmark facilitates future progress toward more general-purpose language understanding models.

In-BoXBART: Get Instructions into Biomedical Multi-Task Learning

Single-task models have proven pivotal in solving specific tasks; however, they have limitations in real-world applications where multi-tasking is necessary and domain shifts are exhibited. Recently, instructional prompts have shown significant improvement towards multi-task generalization; however, the effect of instructional prompts and Multi-Task Learning (MTL) has not been systematically studied in the biomedical domain. Motivated by this, this paper explores the impact of instructional prompts for biomedical MTL. We introduce the BoX, a collection of 32 instruction tasks for Biomedical NLP across (X) various categories. Using this meta-dataset, we propose a unified model termed In-BoXBART, that can jointly learn all tasks of the BoX without any task-specific modules. To the best of our knowledge, this is the first attempt to propose a unified model in the biomedical domain and use instructions to achieve generalization across several biomedical tasks. Experimental results indicate that the proposed model: 1) outperforms the single-task baseline by ~3% and multi-task (without instruction) baseline by ~18% on an average, and 2) shows ~23% improvement compared to the single-task baseline in few-shot learning (i.e., 32 instances per task) on an average. Our analysis indicates that there is significant room for improvement across tasks in the BoX, implying the scope for future research direction.

NumGLUE: A Suite of Fundamental yet Challenging Mathematical Reasoning Tasks

Given the ubiquitous nature of numbers in text, reasoning with numbers to perform simple calculations is an important skill of AI systems. While many datasets and models have been developed to this end, state-of-the-art AI systems are brittle; failing to perform the underlying mathematical reasoning when they appear in a slightly different scenario. Drawing inspiration from GLUE that was proposed in the context of natural language understanding, we propose NumGLUE, a multi-task benchmark that evaluates the performance of AI systems on eight different tasks, that at their core require simple arithmetic understanding. We show that this benchmark is far from being solved with neural models including state-of-the-art large-scale language models performing significantly worse than humans (lower by 46.4%). Further, NumGLUE promotes sharing knowledge across tasks, especially those with limited training data as evidenced by the superior performance (average gain of 3.4% on each task) when a model is jointly trained on all the tasks as opposed to task-specific modeling. Finally, we hope that NumGLUE will encourage systems that perform robust and general arithmetic reasoning within language, a first step towards being able to perform more complex mathematical reasoning.

To Find Waldo You Need Contextual Cues: Debiasing Who's Waldo

We present a debiased dataset for the Person-centric Visual Grounding (PCVG) task first proposed by Cui et al. (2021) in the Who's Waldo dataset. Given an image and a caption, PCVG requires pairing up a person's name mentioned in a caption with a bounding box that points to the person in the image. We find that the original Who's Waldo dataset compiled for this task contains a large number of biased samples that are solvable simply by heuristic methods; for instance, in many cases the first name in the sentence corresponds to the largest bounding box, or the sequence of names in the sentence corresponds to an exact left-to-right order in the image. Naturally, models trained on these biased data lead to over-estimation of performance on the benchmark. To enforce models being correct for the correct reasons, we design automated tools to filter and debias the original dataset by ruling out all examples of insufficient context, such as those with no verb or with a long chain of conjunct names in their captions. Our experiments show that our new sub-sampled dataset contains less bias with much lowered heuristic performances and widened gaps between heuristic and supervised methods. We also demonstrate the same benchmark model trained on our debiased training set outperforms that trained on the original biased (and larger) training set on our debiased test set. We argue our debiased dataset offers the PCVG task a more practical baseline for reliable benchmarking and future improvements.

How Many Data Samples is an Additional Instruction Worth?

Recently introduced instruction-paradigm empowers non-expert users to leverage NLP resources by defining a new task in natural language. Instruction-tuned models have significantly outperformed multitask learning models (without instruction); however they are far from state of the art task specific models. Conventional approaches to improve model performance via creating large datasets with lots of task instances or architectural/training changes in model may not be feasible for non-expert users. However, they can write alternate instructions to represent an instruction task. Is Instruction-augumentation helpful? We augment a subset of tasks in the expanded version of NATURAL INSTRUCTIONS with additional instructions and find that these significantly improve model performance (up to 35%), especially in the low-data regime. Our results indicate that an additional instruction can be equivalent to ~200 data samples on average across tasks.

Less is More: Summary of Long Instructions is Better for Program Synthesis

Despite the success of large pre-trained language models (LMs) such as Codex, they show below-par performance on the larger and more complicated programming related questions. We show that LMs benefit from the summarized version of complicated questions. Our findings show that superfluous information often present in problem description such as human characters, background stories, names (which are included to help humans in understanding a task) does not help models in understanding a task. To this extent, we create a meta-dataset from the frequently used APPS dataset for the program synthesis task. Our meta-dataset consists of human and synthesized summary of the long and complicated programming questions. Experimental results on Codex show that our proposed approach outperforms baseline by 8.13% on an average in terms of strict accuracy. Our analysis shows that summary significantly improve performance for introductory (9.86%) and interview (11.48%) related programming questions. However, it shows improvement by a small margin (~2%) for competitive programming questions, implying the scope for future research direction.

Generalized but not Robust? Comparing the Effects of Data Modification Methods on Out-of-Domain Generalization and Adversarial Robustness

Data modification, either via additional training datasets, data augmentation, debiasing, and dataset filtering, has been proposed as an effective solution for generalizing to out-of-domain (OOD) inputs, in both natural language processing and computer vision literature. However, the effect of data modification on adversarial robustness remains unclear. In this work, we conduct a comprehensive study of common data modification strategies and evaluate not only their in-domain and OOD performance, but also their adversarial robustness (AR). We also present results on a two-dimensional synthetic dataset to visualize the effect of each method on the training distribution. This work serves as an empirical study towards understanding the relationship between generalizing to unseen domains and defending against adversarial perturbations. Our findings suggest that more data (either via additional datasets or data augmentation) benefits both OOD accuracy and AR. However, data filtering (previously shown to improve OOD accuracy on natural language inference) hurts OOD accuracy on other tasks such as question answering and image classification. We provide insights from our experiments to inform future work in this direction.

Choose Your QA Model Wisely: A Systematic Study of Generative and Extractive Readers for Question Answering

While both extractive and generative readers have been successfully applied to the Question Answering (QA) task, little attention has been paid toward the systematic comparison of them. Characterizing the strengths and weaknesses of the two readers is crucial not only for making a more informed reader selection in practice but also for developing a deeper understanding to foster further research on improving readers in a principled manner. Motivated by this goal, we make the first attempt to systematically study the comparison of extractive and generative readers for question answering. To be aligned with the state-of-the-art, we explore nine transformer-based large pre-trained language models (PrLMs) as backbone architectures. Furthermore, we organize our findings under two main categories: (1) keeping the architecture invariant, and (2) varying the underlying PrLMs. Among several interesting findings, it is important to highlight that (1) the generative readers perform better in long context QA, (2) the extractive readers perform better in short context while also showing better out-of-domain generalization, and (3) the encoder of encoder-decoder PrLMs (e.g., T5) turns out to be a strong extractive reader and outperforms the standard choice of encoder-only PrLMs (e.g., RoBERTa). We also study the effect of multi-task learning on the two types of readers varying the underlying PrLMs and perform qualitative and quantitative diagnosis to provide further insights into future directions in modeling better readers.

ILDAE: Instance-Level Difficulty Analysis of Evaluation Data

Knowledge of questions' difficulty level helps a teacher in several ways, such as estimating students' potential quickly by asking carefully selected questions and improving quality of examination by modifying trivial and hard questions. Can we extract such benefits of instance difficulty in NLP? To this end, we conduct Instance-Level Difficulty Analysis of Evaluation data (ILDAE) in a large-scale setup of 23 datasets and demonstrate its five novel applications: 1) conducting efficient-yet-accurate evaluations with fewer instances saving computational cost and time, 2) improving quality of existing evaluation datasets by repairing erroneous and trivial instances, 3) selecting the best model based on application requirements, 4) analyzing dataset characteristics for guiding future data creation, 5) estimating Out-of-Domain performance reliably. Comprehensive experiments for these applications result in several interesting findings, such as evaluation using just 5% instances (selected via ILDAE) achieves as high as 0.93 Kendall correlation with evaluation using complete dataset and computing weighted accuracy using difficulty scores leads to 5.2% higher correlation with Out-of-Domain performance. We release the difficulty scores and hope our analyses and findings will bring more attention to this important yet understudied field of leveraging instance difficulty in evaluations.