Midterm report TODO

[yhamoudi]

Some propositions (updated):

• correct/remove the remaining parts in red in the report
• add a link in intro/overview to gitbuh project. add links/footnotes in each chapter to the relevant repositories https://github.com/ProjetPP/Papers/pull/16
• recall workpackages, goals, organization (= short version of the proposal) in introduction
• add a subsection "future work" to all sections that need one -> wikidata
• merge Machine learning 1 & 2 into a single section, and divide it into 2 subsections
• have a more beautiful first page (ex: http://origram.org/files/rapport-final-origram.pdf) :)
• dont' use sandalone
• replace ML1 and ML2 by something more explicit (rename .tex files too)
• in datamodel, explain how trees are represented in a single-line structure (as displayed in the ppp actually): what the holes/brackets/... mean.The NLP section doesn't have to explain it.
• rename .tex files with the new sections names https://github.com/ProjetPP/Papers/commit/d23fde2c3d161b8ecdf9aef5db7dad46950b419a (still ML1/ML2 to rename...)
• Add a section (after intro) about overall progress of the project. Recall the Gantt diagramm and explain wether deadlines have been respected or not (and eventually more precisions in each section). Introduce the planed evolution of the project until december
• add clearly visible (!= footnote) links to http://ppp.pony.ovh/ (and https://projetpp.github.io/)

[progval]

Approved.

[yhamoudi]

More:

• I've seen the word "sandalone" quite often. Is it really what you want to say?
• Classical Natural Language Processing cannot be a title of section ("classical" means nothing and all the project is about NLP). I propose a section Question simplification (if you have a better idea...) with 3 subsections: Grammatical approach, ML 1 and ML 2 (ML1 and ML2 should be replaced by something more explicit, such as the name of the algorithm used).
• Add a section (after intro) about overall progress of the project. Recall the Gantt diagramm and explain wether deadlines have been respected or not (and eventually more precisions in each sections). Introduce the planed evolution of the project until december

[progval]

On 31/10/2014 11:41, Yassine wrote:

More:

• I've seen the word "sandalone" quite often. Is it really what you want to say? The typo is even in the name of the repository, actually
• Add a section (after intro) about overall progress of the project. Recall the Gantt diagramm and explain wether deadlines have been respected or not (and then, more precisions in each sections). Introduce the planed evolution of the project until december I already detailed the progress in the introduction and the conclusion, in addition to my parts on the core and the logging.

[Ezibenroc]

I've seen the word "sandalone" quite often. Is it really what you want to say?

Yes this is a typo which exists since the creation of the repository. This is a runing gag now :) (but it would be more serious to correct it).

Classical Natural Language Processing cannot be a title of section ("classical" means nothing and all the project is about NLP). I propose a section Question simplification (if you have a better idea...) with 3 subsections: Grammatical approach, ML 1 and ML 2 (ML1 and ML2 should be replaced by something more explicit, such as the name of the algorithm used).

Our module strongly depends on the Stanford library, which is a "grammarian" library (creating it must have required a very strong knowledge of the English language). The ML modules is more "Computer Science" flavoured, with machine learning. I believe that this is in this sens that our module is "classical".

[yhamoudi]

Yes but classical may be interpreted as "old, ancient, naive" vs "modern, powerful" methods. Moreover "classical" implies that there is a "canonical existing way" to product triples from graph dependency, whereas the literature review has shown that nobody gives an explicit way to perform that.

[Ezibenroc]

What name then? Grammatical approach? And if we have a section for "question simplification" and a section for "core" (for instance), the first one will be much bigger than the second one...

[yhamoudi]

Grammatical approach if it seems you good. There are many things common to the 3 question simplification algorithms. I think we must write some global info at the beginning of the section before starting subsections (what is our purpose,our goals/difficulties (depth of the tree, relevance of the triples...)...). If we divide this into 3 sections, where do we put common info? I don't think it's a problem to have a section bigger than the other one. Question simplification is a big part, more than half of us have worded on it, it's normal to have a a big section.

[Ezibenroc]

Ok : https://github.com/ProjetPP/Papers/commit/5efc3bdd8171ebb1ee74f842078b6c7b90910b85

[progval]

Why the appendix?

[Ezibenroc]

"Grammatical approach" has three pages of images (we could maybe reduce to two, but not less). We can not put this in the middle of the report.

[yhamoudi]

More:

• in datamodel, explain how trees are represented in a single-line structure (as displayed in the ppp actually): what the holes/brackets/... mean.The NLP section doesn't have to explain it.
• get out this horrible "neon green" in references, please :'( https://github.com/ProjetPP/Papers/commit/58aa2cec21c96efad2b3dd82b6900cca8073d5c2
• modify the numerotation of figures. Actually a figure can be numbered 5.3 and references to it: 5.1.3 https://github.com/ProjetPP/Papers/commit/fd6f4eb5253af34e01cf65ee3e595a0086c2ea00
• rename .tex files with the new sections names https://github.com/ProjetPP/Papers/commit/d23fde2c3d161b8ecdf9aef5db7dad46950b419a (still ML1/ML2 to rename...)

[yhamoudi]

More:

• Rename NLP by question parsing on fig 3.1

Some sentences I don't understand (perhaps they're good, just verify they mean what you want):

• until the former gets a complete response.
• made its way from a question to an useful answer
• contains few more information such that the language used.
• The current state is the Core is successfully able with all modules that have been written
• The new work package "Computer Algebra System" has to implement such a system in the PPP.

[Tpt]

More:

• links to the source code

[progval]

One link to https://github.com/ProjetPP/ , or specific links for each section?

[yhamoudi]

or both ? (one big link in intro or overview, and then each section can put a footnote to its specific repository if it wants)

[Tpt]

I agree with Yassine

[Ezibenroc]

or both ? (one big link in intro or overview, and then each section can put a footnote to its specific repository if it wants)

+1

[yhamoudi]

We should rename the repositories now, otherwise the links in the report will be obsolete later. I propose:

• PPP-NLP-classical -> PPP-QParsing-Grammatical
• PPP-NLP-ML -> PPP-QParsing-ML-Reformulation
• PPP-NLP-ML-standalone -> ??? Propose something else if you have better ideas :)

[progval]

Not needed; when we rename a repository, GitHub redirects the old link to the new one.

[yhamoudi]

ok, good

Quentin, can you rename your NLP part in the report? (Machine Learning: Triples from scratch?)

[robocop]

Ok!

Le 4 nov. 2014 à 20:29, Yassine notifications@github.com a écrit :

ok, good

Quentin, can you rename your NLP part in the report? (Machine Learning: Triples from scratch?)

— Reply to this email directly or view it on GitHub.

[Ezibenroc]

Not needed; when we rename a repository, GitHub redirects the old link to the new one.

This would still be ugly to have old links in the report...

[yhamoudi]

it will be fixed for final report. ppp-nlp-classical still appears in other parts of the project (router, ....). It we can avoid making all corrections in emergency it's better

[progval]

+1 Yassine

[Ezibenroc]

Ok

[yhamoudi]

no more things :)

reviews are still welcomed of course