Stephen Carlson wrote: ↑December 20th, 2017, 8:18 pm
Jonathan Robie wrote: ↑December 19th, 2017, 12:13 pm
Sure. And examples like ἥψατο μού τις make me think that the patient might be quite affected for some of these.
I'm not sure this is a good example of a patient. "Touch" doesn't have a patient, because the "object" is usually not affected and affectedness is not part of its meaning. The "object" is more of a location where the contact takes place. The motivation for why the object/complement is a partitive genitive is because touching something only contacts part of it.
Jonathan Robie wrote: ↑December 19th, 2017, 12:13 pm
I also wonder if semantic roles could be helpful here -
That's what I was referring to with the Paul Danove case frame stuff. In general, it doesn't solve the analytical problems but pushes them down to another level detail.
To really know whether I agree with you on ἥψατο μού τις, I would have to spend significant time studying similar constructions and how others have modeled agent / patient relationships. So the problem turtles. Building one model depends on another model, and each model needs to be examined in light of the data.
I think the basic takeaway is that we need to look at all the different kinds of objects that occur, at the same time, to make really good decisions here, and we can benefit by looking at the decisions others have made in their systems. There are
various treebanks that have had to look at this question, and they have come up with different approaches to modeling objects.
I'm currently focused on creating an environment where we can compare the decisions made by these existing systems and see what kinds of queries are easily supported and what kinds of queries are not. And it's really helpful for me to make a list of the questions.
I saw Paul Danove's case frame presentation at SBL 2017. I would love to have this as a dataset ... I agree that adding this level of detail doesn't answer all the questions, but it certainly gives more precise data for making good modeling decisions.
Stephen Carlson wrote: ↑December 20th, 2017, 8:18 pmI refer to this as the
fractal nature of grammar: no matter how you define things at any give level of resolution, you'll find things that don't fit. Edward Sapir's way of saying this is:
All grammars leak.
Exactly. And different models leak in different ways. When you have to annotate an entire corpus, you can't avoid this.
That's one of the advantages of having multiple analyses. But every model is just a model. None of them completely captures the language.
ἐξίσταντο δὲ πάντες καὶ διηποροῦντο, ἄλλος πρὸς ἄλλον λέγοντες, τί θέλει τοῦτο εἶναι;
http://jonathanrobie.biblicalhumanities.org/