ufal / lindat-kontext

An alternative web front-end for the Manatee corpus search engine
GNU General Public License v2.0
5 stars 1 forks source link

Multiple trees in tree-view #196

Open vidiecan opened 6 years ago

vidiecan commented 6 years ago

Related to #53

PDT - Query: Königově (1 hit), click on tree, shows only one

Ansa211 commented 6 years ago

I do not understand what is meant by this issue. Should there be more trees for that tree?

What I can imagine is that we might like to show the t-layer tree. But in that case, we would have to provide some way of encoding it in the vertical, including generated nodes, and it is not clear how to do that without cluttering the vertical with artificial nodes that would break surface-layer search.

kosarko commented 6 years ago

@Ansa211 I guess this is what's mentioned in the linked issue #53 in the first version of the tree view we were displaying both the a-tree and t-tree (as shown in a picture by you on the sentence V Königově domě není pánů a hostů .) And I'd guess further that this issue is whether we want it and whether it's possible.

Ansa211 commented 6 years ago

I have an idea about treating the t-trees in KonText with only a slight extension to the current functionalities.

We could compile PDT data as a parallel corpus. It would be turned into two "corpora", one for the a-layer and another one for the t-layer. The a-layer corpus would contain the surface structure of the sentence, a-layer attributes and a small subset of t-layer attributes (e.g. functors of lexical words). The t-layer corpus would contain as many "words" for each sentence as there are t-layer nodes (including the generated ones), most of the t-layer annotation and possibly a small subset of corresponding a-layer attributes (e.g. concatenation of word forms corresponding to the nodes that are not generated; afun).

This issue then would be generalized to the question whether tree-view can be extended so that showing trees for aligned parallel corpora simultaneously would be possible. That would be a nice functionality also for aligned corpora treated with UDpipe.