As helpful as the GBI Syntax Trees are, I have disagreements with them. Randall and Andi are receptive to feedback but there are very different types of disagreement that can arise in syntactic analysis so I thought I’d start to note down what they are.

Somethings aren’t disagreements, just corrections. Some are differences of interpretation of the Greek. Some are differences in overall approach.

Here’s a first attempt at a more refined categorization of types. I’ll call the the person/group who did the initial (published) analysis A1 and the person/group who has the change/disagreement A2.

  • I. correction—A1 actually agrees with A2 but simply made a mistake and can uncontroversially update their analysis accordingly
  • II. ambiguity—both A1 and A2’s analysis is possible in the eyes of the other, but based on other factors, A1 and A2 disagree which analysis to go with. Perhaps this could further be refined into:
    • IIA. cases where A1 and A2 each think their own analysis is the more likely; versus
    • IIB. cases where A1 and A2 each their their own analysis is the only likely.
  • III. terminology/framework—A1 and A2 agree on structure and relationship up to a certain isomorphism but not in the specifics. This could be further split into:
    • IIIA. cases where A1 and A2’s analyses are structurally identical but just different in labels
    • IIIB. cases where A1 and A2’s analyses different in structure even though they are derivable from one another
  • IV. irreconcilable—A1 and A2 disagree on the way the language actually works and the analyses can’t easily be mapped to one another.

I think many of my disagreements with the GBI Trees so far are of type IIIB which means it is likely possible for me to programmatically generate an alternative analysis with my preferred structure. Indeed, converting to a dependency analysis is a simple example of this but even different choices of head within the constituent structure (which is a major source of systemic disagreement) are easy to make.

The great thing about type III in general is that even if you disagree with A1, you can still use the analysis to explore the syntactic phenomenon you want (you just have to map your queries to their labels and their conventions).

I should also note that an important aspect to dealing with this is proper documentation of conventions followed.

With these thoughts down, I’m now interested in other work that has already been done in this area.