While MRS gives a nicely sorted list of MRS readings based on tree-banking, the same is not available for the well-formed trees that get generated for each MRS. So far, I’ve simply been trying them out in the order my tree builder generates them, which is effectively random. When one works, I use it. The problem is that it can come up with unexpected results (to me at least).
After working through some scenarios, I’m wondering if anyone knows of research on whether the forward reading tree is the most common initial interpretation, in general. Is this something that has been studied? It seems intuitively right, and could at least be a better heuristic than I have now (which is none).
Here’s a scenario that shows the problem:
In a file system that looks like this:
\desktop\file1.txt \desktop\file2.txt \system\file3.txt
The user might say: “which files are in a folder?”
It just so happens that my well-formed tree builder builds the reverse reading tree first:
┌────── _folder_n_of(x9,i14) _a_q(x9,RSTR,BODY) ┌────── _file_n_of(x3,i8) └─ _which_q(x3,RSTR,BODY) └─ _in_p_loc(e2,x3,x9)
file is plural, only
a folder that has >1
file in it will be found. In the world state above, this means the answer is “file1.txt” and “file2.txt”.
In the forward reading tree:
┌────── _file_n_of(x3,i8) _which_q(x3,RSTR,BODY) ┌────── _folder_n_of(x9,i14) └─ _a_q(x9,RSTR,BODY) └─ _in_p_loc(e2,x3,x9)
via the magic of distributive readings, each “1 folder” can be per file, and, as long as the total number of files adds up to >1, they are licensed. So, “files” can return even files that are the sole files in a folder. So the answer now is “file1.txt”, “file2.txt”, “file3.txt”.
I suspect that the latter (forward) reading is the one people would expect, although this is a somewhat odd question to ask (since every file is in a folder in a file system).