Kenneth C. Arnold

An interface for targeted collection of common sense knowledge using a mixture model

We present a game-based interface for acquiring common sense knowledge. In addition to being interactive and entertaining, our interface guides the knowledge acquisition process to learn about the most salient characteristics of a particular concept. We use statistical classification methods to discover the most informative characteristics in the Open Mind Common Sense knowledge base, and use these characteristics to play a game of 20 Questions with the user. Our interface also allows users to enter knowledge more quickly than a more traditional knowledge-acquisition interface. An evaluation showed that users enjoyed the game and that it increased the speed of knowledge acquisition.

Managing ambiguity in programming by finding unambiguous examples

We propose a new way to raise the level of discourse in the programming process: permit ambiguity, but manage it by linking it to unambiguous examples. This allows programming environments to work with informal descriptions that lack precise semantics, such as natural language descriptions or conceptual diagrams, without requiring programmers to formulate their ideas in a formal language first. As an example of this idea, we present Zones, a code search and reuse interface that connects code with ambiguous natural language statements about its purpose. The backend, called ProcedureSpace, relates purpose statements, static code analysis features, and natural language background knowledge. ProcedureSpace can search for code given statements of purpose or vice versa, and can find code that was never annotated or commented. Since completed Zones searches become annotations, system coverage grows with user interaction. Users in a preliminary study found that reasoning jointly over natural language and programming language helped them reuse code.

Embracing ambiguity

Software helps people fulfill their goals, but development tools lack understanding of those goals. But if development tools did understand how software artifacts relate to higher-level intents and goals, they could help developers reuse code, solve problems, and develop systems that are more robust and easier to use. In this paper, we suggest that supporting software development at a stage before concrete formalization is an area of opportunity for software engineering research. We discuss three aspects that are both core challenges and opportunities for this research area: handling ambiguity, understanding human situations, and flexible reflection about failure, and identify research results suggesting that substantial progress can be made on these problems within a decade. We believe that this research will make it easier to develop software that is more broadly useful and robust, even in the face of everyday uncertainty and failure.