[developers] PyDelphin's future

[Michael Wayne Goodman]

Hi everyone,

The quiz has been open for about 2.5 weeks and I now have 13 responses.
Thanks to those who participated! I'll leave the quiz open for a bit in
case this email reminds someone who wanted to respond. A summary of the
responses is below:

The most common usage of PyDelphin is as a Python 3 library (85%), then
indirectly (e.g., through a demo; 54%), then as a Python 2 library (31%),
and as a command (23%).

The most used features are, as I expected, *MRS inspection (77%), the ACE
interface (62%), and SimpleMRS (de)serialization (54%), and [incr tsdb()]
support (39%). I did not expect that SimpleDMRS serialization, DMRS-PENMAN
serialization, and TDL parsing and inspection were also commonly used, all
at 39%. The remaining serialization formats are MRX (23%), MRS-JSON (31%),
DMRX (31%), DMRS-JSON (23%), DMRS-tikz (for LaTeX; 15%), EDS (native; 15%),
EDS-JSON (0%), EDS-PENMAN (0%). I was happy to see 2 respondents (15%)
using the relatively new SEM-I module, and one (8%) using the VPM module.
The YY Tokens and Derivation modules both had 3 users (23% each), and the
web interface had 4 users (31%).

Responses were low (just 4) for the command usage, so I won't summarize
here.

There were 9 responses for desired features, and I was not surprised to see
"more documentation" as a popular choice (56%). PyDelphin has fairly
thorough API documentation and some literate unit tests, but not so much of
beginner-level tutorials and small examples. I was, however, surprised to
see a demand for more TDL features (compiling at 44%, serialization at 33%,
and interactive unification at 22%). For *MRS, people wanted scope
resolution (44%) and modification (33%), but none of the respondents wanted
RMRS or DM support (although I note that DM is often used by researchers
outside of DELPH-IN; e.g., https://arxiv.org/pdf/1804.05990.pdf). There
were 3 people who wanted REPP tokenization in some form. Features with one
vote include: increasing performance, integration with NLTK, DMRX
serialization without the root <dmrs-list> node, and predicate string
manipulation.

I'll be considering these responses as I prepare the next releases of
PyDelphin. My current plan looks like this (listing the major changes only):

* v0.7.0 (nearly complete) improves [incr tsdb()] support and adds
MRS-Prolog exporting and maybe a REPP tokenizer

* v0.8.0 improve TDL support

* v0.9.0 *MRS improvements: manipulation and scope resolution

* v1.0.0 restructure the PyDelphin module hierarchy and remove deprecated
functionality, add more testing and documentation

I don't know if I'll be able to continue working on PyDelphin once I find
employment, so for v1.0.0 (if I get there) I want to make the code clean,
correct, and easy to maintain so that someone else can easily take over as
the maintainer, or as a co-maintainer, of PyDelphin (email me if you're
interested!). If nobody volunteers I'll arrow Francis for the job, but I'm
hoping an interested student will step up.

Thank you

On Sat, Apr 7, 2018 at 9:20 PM, Michael Wayne Goodman <goodmami at uw.edu>
wrote:

> Hello developers,
>
> I recently finished my degree and while I look for jobs I'm going to work
> on PyDelphin a bit. If you use, have used, or want to use PyDelphin, please
> consider taking this short survey (~2-3 minutes) to help me understand
> which current features are useful and which potential features are desired:
>
>     https://goo.gl/forms/BjTQuSkOgR9HBGZn1
>
> GitHub tells me that PyDelphin is downloaded by ~2 people per week and the
> project is viewed by ~3 people per day, so I know people besides me use it,
> but I don't know how they use it. Thanks for any help!
>
> --
> Michael Wayne Goodman
>



-- 
Michael Wayne Goodman
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.delph-in.net/archives/developers/attachments/20180425/6864ca9b/attachment.html>