SingaporeOnlineDemos

Demophin demo

Mike doing a demo of demophin.py

• Instructions on how to install.
• Discussion on assumptions regarding setup.
  • Eg ace version must match that the grammar was compiled with.
  • Improvements to demophin.py (and other tools that use ACE) is passing on this kind of error message to the user for robustness.

Discussion

Woodley: how does one go about dropping drms drawing into another app such as the FFTB?

Mike: Very possible, just need to call the right python function on the server to generate a json object which there is javascript code to parse/render.

Woodley: What is the formal status of converting to DMRS from a partial MRS? Mike: It should work, or if not, can get it to work in short order.

Woodley: do we foresee this eventually replacing the existing web demos?

Mike: Possibly? That was not the idea, however open to this -- but obviously many features need to be added. Would need assistance.

Mike: TJ Trimble is also working on a fully featured browser-based IDE using LUI.

Guy: There is the problem of having all these different modules being implemented multiple times in different codebases.

Mike: Demophin has a separation of server from visualisation. Need to keep the visualisers separate from the server functionality.

Ned: Some people have been asking about the (semi) interactive trees from Typediff. These are generated using JSON produced by C code that calls ACE libraries to generate a both short and long labelled trees along with character spans of nodes. Would be nice on explicitly supporting this.

Woodley: What exactly is the concern with the current setup?

Ned: That by hooking into ACE libraries, later changes by Woodley might break this setup.

Woodley: Making new programs that link to ACE libraries is a common pattern that I make use of. Happy to declare the functionality of reading in a derivation tree and reconstructing the graph stable and unlikely to change.

Perhaps we should look at creating a repository for modules providing different kinds of visualisations of DELPH-IN data structures. These could be composed of server side generation (producing JSON or other exchange format) and client side (Javascript) rendering pairs.

Petter: What about if I want ACE to generate the full AVM?

Woodley: You can kind of get that using the lui interface (it's a bit of a hack) however this is the trimmed AVM without all the daughters. Some extra work will be required in order to get the full AVM (and also make it more accessible). Will work on this (??).

Woodley: Additionally, the --report-labels flag could be useful to people. Currently --report-labels only provides the short labels (and not the longer labels/token mapping information etc derivation), however no one seems to be depending on this so perhaps this could be changed to

Dan: popular request is parsing/representing longer chunks of text with possibly multiple sentences. How should these be represented? A sequence of top-ranked parse trees? Perhaps looking at linking discourse entities (a la discourse representation structure)? It would be nice to paint a bigger picture.

Woodley: DELPH-IN doesn't have much to say about larger discourse structures.

Dan: Yes but if we had some kind of representation with missing information/linkages, this could help motivate working towards these sorts of problems, and also entice people working more with discourse related frameworks into our ecosystem.

(Discussion of future features for demophin, missed this.)

Dan: From the perspective of making our infrastructure more appealing, it would be nice to produce a consistent front door to the various windows of different kind of analyses.

Dan: We have various people working on interfaces/visualisations at the moment, using modern technologies, perhaps they could join forces in some way?

Mike: I would be happy to moderate/curate this collaboration if other people are willing to submit features.

Ned: should this be located within the demophin repository?

Mike: It would be nice to keep the visualisers separate from the server. Also people outside of DELPH-IN might like to use the DMRS visualisation so keeping it separate would be advantageous.

Tuan Anh: There is also the DMRS viewer that Francis and I made.

Dan: is this simply a different skin on the kind of thing that demophin is doing?

Mike: looks a bit different. Also using a different javascript graphics library.

Dan: How long is it going to take to enact these plans?

Ned: Do we want this to be geared more towards a unified demo or also multiple server/viewer pairs which can be dropped into other apps.

??: Both... but don't worry too much about locking in a standard. If there are modules that people find useful then hopefully they will use them.

Woodley: We shouldn't expect there to be only eg one tree viewer. Multiple solutions should be encouraged in this package.

Also, the JSON (or whatever exchange format) shouldn't be standardised across different visualisers. The server side code and client side pairs should just target each other.

Mike: Suggest creating a new package in the DELPH-IN group for the various visualisers.

Woodley??: Would also be nice to have latex generators for various data structures.

Chris: I can put my hand up for being responsible for latex generation --- need this for my own research (and have funding).