Jump to content

Discrepancy in result hits


Λύχνις Δαν

Recommended Posts

Hi ya,

 

I have just run through http://www.accordancebible.com/forums/index.php?showtopic=6943 (Marco Fabri's first example Greek Syntax search). I am seeing very different numbers of hits. The screen shots he provided only show the initial hits and I am getting those correctly. I am unsure though why the hit counts differ. Has the counting of hits in these searches (or perhaps in searches in general) changed since Acc 9 in which his examples were done ? Or is there something wrong ? I don't have Acc 9 to perform a comparison with.

 

Just taking the four examples in the post above I get the following hit counts versus Marco's :

 

For adjunct I get 296 hits versus Fabri’s 459,

For complement I get 526 vs 683,

For subject I get 31 vs 50,

For appositive I get 44 vs 112.

 

I'm pretty sure I've followed the example properly but if anyone can reproduce the original results listed in the post, I'd love to compare notes and find the difference.

 

Note that this is on 10.1.3 (on Mac OSX 10.8.2) as I know that update includes some syntax fixes.

 

Thx

D

Link to comment
Share on other sites

There could be two things at work here. Marco is responsible for producing the Syntax Database for the GNT-T; thus, it could be that Marco is running his search against a larger database than what is currently released to the public.

 

Second, I am aware of several significant changes in the search functionality of the Syntax Construct Searches since Marco's original post. I can check later, but one would need to see if the above searches provide linguistic viable results to what the query is, not whether numbers match (although I can understand why you are doing this). Perhaps, Marco or Joel can chime in and let you know either way.

Link to comment
Share on other sites

One thing all users of the syntax modules need to be aware of is that the search results complicate the "hits numbers". Accordance counts by words, but syntax is done by "constituents (which may be from a single particle, like a preposition, all the way to a very, very long phrase). The hits # do not count syntactic constituents, unless the requested constituent (e.g., a subject) coincidently overlaps with a single word.

 

Since I can hardly imagine what programming magic it would take to count constituents versus words and I have no desire to add more to the overloaded programmers' shoulders, I would not hold my breath for an accurate count of syntactic hits to appear.

 

Rather, I suggest taking the verse count of the hits as the baseline and then scrolling to get a sense of how many syntactic items searched for appear in the typical verse. Anyone who wants a more accurate count will, like any good researcher, do the counting manually anyway.

Link to comment
Share on other sites

Hi James, Robert,

 

Thanks for the thoughts on this one. Robert I was mainly looking at this test result with my software developer hat on - one I'm much more qualified to wear that my NT Greek linguist hat which is woefully ill-made so far. My main concern was faith in the outcomes I am seeing. Certainly I have no argument that one needs to analyze the cases individually to see what's what. I was not concerned with what was being counted so much as the stability of the count given that the data set and operations were supposed to be the same. Of course, as James points out, that might not be so. I work in databases and correctness of search results is a prime concern. And I understand that the precise classification of particular instances of constructions is not always clear cut and without divergent opinion. Had the difference been slight I'd have put it down to slight adjustments in the text, the tagging or such. Given the magnitude of the differences it could point to changes in counting methods perhaps not so far removed from the magical algorithm to count constituents as opposed to words. Equally it could be just as James suggests.

 

Robert I was intrigued by your last paragraph. Checking verses for a particular construction seems like an application of the search results - to determine a particular writer's predilection for a particular construction, or its prevalence in a specific type of writing. Is that what you were getting at ? And yes its a given that to validate the results directly one would have to read and classify and count for oneself. I hope to be so capable one day.

 

Thx
D

Link to comment
Share on other sites

Daniel,

 

I can't answer for Marco, but I am always working with a different database than the officially released one. Currently, mine has 4 large additional books, with 3 more to be added soon. I then proof, check, revise before sending a new module for Oaktree to release. Moreover, the syntax searching itself has been refined with each Accordance upgrade. So until 1) the modules are completely finished and 2) the syntax searching is perfected (and it's getting close), then the hits will change with each upgrade, of both kinds.

Link to comment
Share on other sites

Hi Robert,

 

Makes sense. As my understanding of syntax improves I'll be better placed to judge reliability of results directly. I'll likely have to abandon my simple approach for now.

 

Many thanx

D

Link to comment
Share on other sites

  • 3 months later...

Sorry for chiming in so late: Robert is right. My searches were performed on my database, that already contained the rest of the NT.

I am now re-checking it for release.

 

Even so, if I had limited the search range to Matthew-Acts, I should have gotten the same results.

I say I should, because searches are better refined now in Accordance than they were when I prepared those searches. They would now be close to those results.
Link to comment
Share on other sites

Hi Marco,

 

Thanx for confirming this. I very much look forward to the full NT syntax.

 

Thx

D

Link to comment
Share on other sites

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now
×
×
  • Create New...