Open Issues Meta Discussion

This page logs some of the discussion we had about managing open issues and how to best channel effort (e.g., student research) to projects which are of high priority but not otherwise currently being addressed. The main page for open issues is OpenissuesTop, and instructions for adding or modifying open issues are given there.

Many topics come up repeatedly which are too complicated to solve in one meeting or one research project for which they are not central. Need better connection to DELPH-IN external and internal folks who are looking for things to work on (PhD theses, MA theses)

People download our stuff, play with it, but don’t keep in contact. Sometimes useful products we don’t know about, sometimes things that could have been more useful/better if they were created in communication with us. Somehow make it easier to know who to talk to within DELPH-IN and that they are willing to be approached!

Especially needed for linguists who are primarily linguists but a bit computationally inclined. DELPH-IN provides big infrastructure that can’t be created by any individuals, esp. not by linguists. But, only really works with human connection to someone inside DELPH-IN.

At the same time, there are ways in which these open issues can interest the broader computational community, and DELPH-IN can benefit from that interest: state of the art statistical technology on top of our grammars, extract features from our representations in a computationally feasible way, run existing toolkits in combination. DELPH-IN resources can articulate problems, provide large, open-source multilingual resources — present and present our needs to ML people.

Making the resources accessible to ML people requires some repackaging. Extracting treebanks from [incr tsdb()] so that e.g., supertagging is a shared task. TimBaldwin volunteers to maintain repository of repackaged treebanks and/or include export scripts in DELPH-IN repository. If we can present a DELPH-IN shared task, the large ML community will probably come to call. There are some issues with managing it so that folks don’t obsess over less-relevant small tasks. What are the interesting open issues that are amenable to this approach? Minimize the number of assumptions (over-simplifying) in the task set up.

Last update: 2006-06-29 by EmilyBender [edit]