An official website of the United States government
Compilers primarily give feedback about problems to developers through the use of error messages. Unfortunately, developers routinely find these messages to be confusing and unhelpful. In this paper, we postulate that because error messages present poor explanations, theories of explanation---such as Toulmin's model of argument---can be applied to improve their quality. To understand how compilers should present explanations to developers, we conducted a comparative evaluation with 68 professional software developers and an empirical study of compiler error messages found in Stack Overflow questions across seven different programming languages. Our findings suggest that, given a pair of error messages, developers significantly prefer the error message that employs proper argument structure over a deficient argument structure when neither offers a resolution---but will accept a deficient argument structure if it provides a resolution to the problem. Human-authored explanations on Stack Overflow converge to one of the three argument structures: those that provide a resolution to the error, simple arguments, and extended arguments that provide additional evidence for the problem. Finally, we contribute three practical design principles to inform the design and evaluation of compiler error messages.
more »
« less
Explaining spreadsheets with spreadsheets
Based on the concept of explanation sheets, we present an approach to make spreadsheets easier to understand and thus easier to use and maintain. We identify the notion of explanation soundness and show that explanation sheets which conform to simple rules of formula coverage provide sound explanations. We also present a practical evaluation of explanation sheets based on samples drawn from widely used spreadsheet corpora and based on a small user study. In addition to supporting spreadsheet understanding and maintenance, our work on explanation sheets has also uncovered several general principles of explanation languages that can help guide the design of explanations for other programming and domain-specific languages.
more »
« less
- Award ID(s):
- 1717300
- PAR ID:
- 10096836
- Date Published:
- Journal Name:
- 17th ACM SIGPLAN International Conference on Generative Programming: Concepts and Experiences
- Page Range / eLocation ID:
- 161 to 167
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Numerous methods have been developed to explain the inner mechanism of deep neural network (DNN) based classifiers. Existing explanation methods are often limited to explaining predictions of a pre-specified class, which answers the question “why is the input classified into this class?” However, such explanations with respect to a single class are inherently insufficient because they do not capture features with class-discriminative power. That is, features that are important for predicting one class may also be important for other classes. To capture features with true class-discriminative power, we should instead ask “why is the input classified into this class, but not others?” To answer this question, we propose a weighted contrastive framework for explaining DNNs. Our framework can easily convert any existing back-propagation explanation methods to build class-contrastive explanations. We theoretically validate our weighted contrast explanation in general back-propagation explanations, and show that our framework enables class-contrastive explanations with significant improvements in both qualitative and quantitative experiments. Based on the results, we point out an important blind spot in the current explainable artificial intelligence (XAI) study, where explanations towards the predicted logits and the probabilities are obfuscated. We suggest that these two aspects should be distinguished explicitly any time explanation methods are applied.more » « less
-
null (Ed.)Feature attributions and counterfactual explanations are popular approaches to explain a ML model. The former assigns an importance score to each input feature, while the latter provides input examples with minimal changes to alter the model's predictions. To unify these approaches, we provide an interpretation based on the actual causality framework and present two key results in terms of their use. First, we present a method to generate feature attribution explanations from a set of counterfactual examples. These feature attributions convey how important a feature is to changing the classification outcome of a model, especially on whether a subset of features is necessary and/or sufficient for that change, which attribution-based methods are unable to provide. Second, we show how counterfactual examples can be used to evaluate the goodness of an attribution-based explanation in terms of its necessity and sufficiency. As a result, we highlight the complimentary of these two approaches. Our evaluation on three benchmark datasets --- Adult-Income, LendingClub, and German-Credit --- confirms the complimentary. Feature attribution methods like LIME and SHAP and counterfactual explanation methods like Wachter et al. and DiCE often do not agree on feature importance rankings. In addition, by restricting the features that can be modified for generating counterfactual examples, we find that the top-k features from LIME or SHAP are often neither necessary nor sufficient explanations of a model's prediction. Finally, we present a case study of different explanation methods on a real-world hospital triage problem.more » « less
-
In human-aware planning systems, a planning agent might need to explain its plan to a human user when that plan appears to be non-feasible or sub-optimal. A popular approach, called model reconciliation, has been proposed as a way to bring the model of the human user closer to the agent’s model. To do so, the agent provides an explanation that can be used to update the model of human such that the agent’s plan is feasible or optimal to the human user. Existing approaches to solve this problem have been based on automated planning methods and have been limited to classical planning problems only. In this paper, we approach the model reconciliation problem from a different perspective, that of knowledge representation and reasoning, and demonstrate that our approach can be applied not only to classical planning problems but also hybrid systems planning problems with durative actions and events/processes. In particular, we propose a logic-based framework for explanation generation, where given a knowledge base KBa (of an agent) and a knowledge base KBh (of a human user), each encoding their knowledge of a planning problem, and that KBa entails a query q (e.g., that a proposed plan of the agent is valid), the goal is to identify an explanation ε ⊆ KBa such that when it is used to update KBh, then the updated KBh also entails q. More specifically, we make the following contributions in this paper: (1) We formally define the notion of logic-based explanations in the context of model reconciliation problems; (2) We introduce a number of cost functions that can be used to reflect preferences between explanations; (3) We present algorithms to compute explanations for both classical planning and hybrid systems planning problems; and (4) We empirically evaluate their performance on such problems. Our empirical results demonstrate that, on classical planning problems, our approach is faster than the state of the art when the explanations are long or when the size of the knowledge base is small (e.g., the plans to be explained are short). They also demonstrate that our approach is efficient for hybrid systems planning problems. Finally, we evaluate the real-world efficacy of explanations generated by our algorithms through a controlled human user study, where we develop a proof-of-concept visualization system and use it as a medium for explanation communication.more » « less
-
Abstract In this study, we used Epistemic Network Analysis (ENA) to represent data generated by Natural Language Processing (NLP) analytics during an activity based on the Knowledge Integration (KI) framework. The activity features a web-based adaptive dialog about energy transfer in photosynthesis and cellular respiration. Students write an initial explanation, respond to two adaptive prompts in the dialog, and write a revised explanation. The NLP models score the KI level of the initial and revised explanations. They also detect the ideas in the explanations and the dialog responses. The dialog uses the detected ideas to prompt students to elaborate and refine their explanations. Participants were 196 8th-grade students at a public school in the Western United States. We used ENA to represent the idea networks at each KI score level for the revised explanations. We also used ENA to analyze the idea trajectories for the initial explanation, the two dialog responses, and the final explanation. Higher KI levels were associated with more links and increased frequency of mechanistic ideas in ENA representations. Representation of the trajectories suggests that the NLP adaptive dialog helped students who started with descriptive and macroscopic ideas to add more microscopic ideas. The dialog also helped students who started with partially linked ideas to keep linking the microscopic ideas to mechanistic ideas. We discuss implications for STEM teachers and researchers who are interested in how students build on their ideas to integrate their ideas.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1145/3278122.3278136
