Scientists, engineers, and educators commonly need to make graphs that quickly illustrate quantitative ideas yet are not based on specific data sets. We call these graphs quantitative concept diagrams (QCDs). Existing charting and illustration programs make it possible to produce such graphs, but they are so time-consuming that users tend to sketch the graphs by hand instead. To reduce the cost of creating QCDs, we developed Graph Sketcher, a quantitative graphing tool that deeply integrates the data plotting capabilities of charting programs with the direct manipulation techniques of illustration programs. We show that our integrated interface substantially reduces the time needed to create QCDs, and we further show that real Graph Sketcher users both enjoy and take advantage of the interface improvements to create QCDs in a wide range of fields.

Traditional interfaces for information access do not fully support queries that rely on semantic relationships between terms. To better support such queries, we introduce a system that automatically extracts subject-verb-object concepts from unstructured text documents and dynamically presents them to the user as navigable refinements. This approach, which we call "idea navigation," makes subject-verb-object querying as simple as selecting successive refinements. It also supports exploratory search by providing a view of the most common ideas in the current result set. First-time users of a prototype system successfully used idea navigation to solve realistic search tasks, demonstrating its effectiveness.

A grammatical method of combining two kinds of speech repair cues is presented. One cue, prosodic disjuncture, is detected by a decision tree-based ensemble classifier that uses acoustic cues to identify where normal prosody seems to be interrupted (Lickley, 1996). The other cue, syntactic parallelism, codifies the expectation that repairs continue a syntactic category that was left unfinished in the reparandum (Levelt, 1983). The two cues are combined in a Treebank PCFG whose states are split using a few simple tree transformations. Parsing performance on the Switchboard and Fisher corpora suggests that these two cues help to locate speech repairs in a synergistic way.

We present a reranking approach to sentence-like unit (SU) boundary detection, one of the EARS metadata extraction tasks. Techniques for generating relatively small n-best lists with high oracle accuracy are presented. For each candidate, features are derived from a range of information sources, including the output of a number of parsers. Our approach yields significant improvements over the best performing system from the NIST RT-04F community evaluation