Stephen Carlson wrote: Stephen Hughes wrote:
Stephen Carlson wrote:Looks like you would prefer dependency tree instead of constituency trees. Other treebanks do the former, e.g. PROIEL.
I don't have the training or background to understand the difference between them. I'll try to scratch something and show you what I'm thinking of.
The Wikipedia page on Dependency Grammar http://en.wikipedia.org/wiki/Dependency_grammar
has been updated by a Timothy Osborne, a linguist who's interested in the distinction between these two type of grammars and trees. That would be a good place to start scratching.
Any interest in scratching on this question here in B-Greek? There are basically three open data treebanks for the Greek New Testament: PROIEL (dependency), GBI (headed phrase structure grammar), or Lowfat (derived from GBI, uses a mongrel model where the verb is modeled more or less as in dependency grammars and phrases are modeled more or less like headed phrase structure grammars).
Dag Haug has done some work comparing them and converting from dependency to constituency in his paper, From Dependency Structure to LFG Representations
, and people whose opinion I respect claim that you can, in general, transform dependency into constituency and vice versa. I think the basic information content is at least very similar in general, and it's at least plausible to me that transformations can be made in both directions.
The PROIEL trees have added tags to indicate whether information is new or old, etc.
Code: Select all
<value tag="new" summary="new"/>
<value tag="kind" summary="kind"/>
<value tag="acc_gen" summary="acc-gen"/>
<value tag="acc_sit" summary="acc-sit"/>
<value tag="acc_inf" summary="acc-inf"/>
<value tag="old" summary="old"/>
<value tag="old_inact" summary="old-inact"/>
<value tag="no_info_status" summary="annotatable (undecided)"/>
<value tag="info_unannotatable" summary="unannotatable"/>
<value tag="quant" summary="quantifier restriction"/>
<value tag="non_spec" summary="non-specific"/>
<value tag="non_spec_inf" summary="inferred from non-specific"/>
<value tag="non_spec_old" summary="non-specific old"/>
We don't have that, it could be added, but we do not have it now.
If we wanted to do this, it might be interesting to take a set of passages, explore how they are represented in each of these models, compare the information content available in each, and think about the questions each could be used to answer and how you would go about answering those questions. Any of you linguists have time and energy to help guide me through this? I can do queries on PROIEL, Lowfat, and GBI, maybe someone else can work with Cascadia?