-
Notifications
You must be signed in to change notification settings - Fork 4
MatrixMrsCatb
[#candb This] is an early essay on Open Source. It is about 800 sentences, which is small, but there are more essays if we want more data. There are several good translations (not all linked to the main page). [http://en.wikipedia.org/wiki/The_Cathedral_and_the_Bazaar Wikipedia] also has a number of links to different translations: C (see on the left) (AF). It is freely available, but I (FCB) checked with the author anyway as a matter of courtesy and he was enthusiastic about us using it. There will be some clean up work involved in getting the translations aligned (there are several versions of the essay).
It was proposed (by FrancisBond) and accepted (by everyone) at the Kyoto Summit (2008) that we use this as a multilingual shared test suite to enable us to compare parses across different grammars. This page describes the steps we are taking to prepare the translations of the essays as a corpus. As the data becomes available, we will also link it to this pages.
At NiCT we also have a 201 sentence aligned subset of en,ko,zh,de,pt,it,fr which we use for MT testing.
This is the timeline agreed on at the Kyoto Summit.
-
0 Prepare and release partially filled skeletons (2008-08: FCB) 1 Make profile (2008-09)
-
align and correct translations
-
translate remaining text (if the original translation is incomplete)
-
feedback translations to the translators
-
segment if necessary
-
link the profile in the table above
-
add the profile to your grammar (e.g., jacy/tsdb/skeletons/catb.ja)
-
NiCT will also link as n(n-1) parallel corpora
2 Treebank profile(2009-03)
-
translate
-
treebank
-
share the treebank to allow for comparisons
-
include it with your grammar (e.g., grammar/gold/catb.ja)
3 Compare treebanks at the next DELPH-IN summit (2009-??)
-
Treebanking this text leads to several interesting issues with text cleansing: italics, embedded quotations, list numbers and so forth. In this section we will discuss what we have done in non-straightforward cases.
Note that we are not treating this as a corpus for testing the robustness of our systems to raw text, but rather as a set of sentences for comparing the semantic representations across languages. Therefore, we will try to make the input text as easy to parse as possible. In our corpus all markup is removed and obvious infelicities (typos, mispellings, bad translations) should be corrected. If and when we want to look at robustness issues, we will choose a new text (possibly the next essay in this series).
We have removed all markup (hyperlinks, italics, paragraph boundaries, ...). These can be added in when we have more of a handle on how to deal with them.
Examples:
-
Perhaps this is not only the future of <emphasis>open-source</emphasis> software.
- → Perhaps this is not only the future of open-source software.
-
Other examples are legion, as a visit to <ulink url="http://freshmeat.net/">Freshmeat</ulink> on any given day will quickly prove.
- → Other examples are legion, as a visit to Freshmeat on any given day will quickly prove.
Mark headers as headers (with a preceding + in the text profile, as XP in the item file):
- +The Cathedral and the Bazaar
Add list item numbers to the first sentence in the list item.
- 6. Treating your users as co-developers is your least-hassle route to rapid code improvement and effective debugging.
- When I expressed this opinion in his presence once, he smiled and quietly repeated something he has often said: "I'm basically a very lazy person who likes to get credit for things other people actually do."
We should correct obvious typos in the profile, and also send them upstream to the maintainer of the essay/translation.
-
the costs of duplicated work tend to scale sub-qadratically with team size
- → the costs of duplicated work tend to scale sub-quadratically with team size
Anything that is not clearly in error should be left as is.
Home | Forum | Discussions | Events