ParisGrammarCatalogue

===Grammar Catalogue===

currently we have summary pages: eg. ErgTop

hands raised for ppl wanting more Top pages (or similar), similar number people willing to create this data

emily discusses possible fields (lexicon size, # types, canonical citation, etc)

antske: anonymity regarding grammars is good for submitting papers, so we should be able to have demos separate from identifying data

glenn: likes

laurie: if using database, easy to query for phenomena

• glenn: but can't "glance" at table to see overview of fields emily: we'll want a database, but interface can be

antonio: how much metadata? too much? perhaps we should do "stages" of adding info to the database.

francis: how many can we automatically extract? should we have standard names for components (e.g. unknown word handling)

ann: we should definitely ground numbers and things on test sets (or something objectively measurable)

• francis: so we need an expanded MRS test set? dan: for langs without a feature, typically the same sentence will appear twice for that feature. dan: related to matrix's item file creation with phenomenon labels

francis: canonical.bib for automating citation extraction

stephan: this database (and scripts) should be in a repository so we have versions

emily volunteers francis for making script to extract features

laurie: MRS test suite is too small to cover phenomena featured in some languages

stephan: a phenomena wiki?

laurie: is there any benefits for grammar developers to do this?

ann: the infrastructure for coordinating is lacking. E.g. we added some sentences to the MRS test suite that weren't adopted.

francis: we need a "not in my language" option.

dan: we need a translation, not just monolingual sentences with phenomena listed

??: but with english how to illustrate? a paragraph of explanation?

stephan: running out of time, but main point is we need a repository of phenomena, or find one that exists

francis: how many phenomena might exist?

• laurie: i originally asked for this to see languages with semantic perfect (fine-grained distinction) emily: probably 300 at least, if people do it maybe 1000

bart: we should create a framework for this catalogue so it's there, and fill in values later