A couple of questions about the construct “say ‘[something]’”:
With the 2020 grammar I got 24 readings for “say ‘what are you doing?’” but they were all SF: ques. Shouldn’t at least one be SF: comm since this is a command to say a question?
Also: The first parse doesn’t include a predicate for “say” at all. Is there an interpretation that somehow ignores it?
% ace -g erg-2020-osx-perplexity.dat -Tf
say "what are you doing?"
SENT: say "what are you doing?"
[ LTOP: h0
INDEX: e2 [ e SF: ques TENSE: pres MOOD: indicative PROG: + PERF: - ]
RELS: < [ thing<5:9> LBL: h4 ARG0: x5 ]
[ which_q<5:9> LBL: h6 ARG0: x5 RSTR: h7 BODY: h8 ]
[ pron<14:17> LBL: h9 ARG0: x3 [ x PERS: 2 IND: + PT: std ] ]
[ pronoun_q<14:17> LBL: h10 ARG0: x3 RSTR: h11 BODY: h12 ]
[ _do_v_1<18:23> LBL: h1 ARG0: e2 ARG1: x3 ARG2: x5 ] >
HCONS: < h0 qeq h1 h7 qeq h4 h11 qeq h9 >
...23 more readings...
There is a use of ‘say’ as a discourse particle as in “We need to pick up say ten bags of chips”, and this particle can also be used sentence-initially as a rough equivalent of “hey” to draw the interlocutor’s attention, as in “hey, what are you doing?”. The 2020 grammar is still rather inconsistent in how it treats those discourse particles semantically, as you can see with a similar particle “yo” as in “yo, what are you doing” where an EP appears in the MRS for “yo”, though none for “hey” or “say”. I’ll work on correcting the lexical type for those semantically invisible discourse particles so they each appear in the MRS.
As for the SF ques in the parses where “say” is a verb, you’re right to expect there to be a reading where the SF has the value ‘comm’ but this issue of sentence-final punctuation inside of quoted sentential complements has proved to be a vexing one. There are varying conventions about whether that question mark inside of the quotes applies to the embedded question or to the whole utterance (search Google for “question mark inside or outside quotes”, for example). And there is a technical issue I haven’t solved yet with getting the SF value right in the grammar even when we want to have the question mark only apply to the embedded question, when it is at the end of the whole utterance. So you’ve bumped into an untidy corner of the grammarian’s garden, alas.
I definitely did since i figured it was just me getting it wrong. Although it seemed like there was consensus that putting it inside the quotes meant that the thing in quotes was a question. I’m not sure how much my google searching would hold up in academic circles though.
Regardless, please record my vote to solve that technical issue and record SF: comm. The current behavior breaks this phrase in my system since it runs the whole thing as a query which is obviously very different.
I have now managed to improve the analysis of embedded questions in the ERG so examples like “say ‘what are you doing?’” now correctly provide a sentence-force for a command instead of a question. I’ll check in the improvements to the trunk soon, and they’ll be in the next stable release.
Lovely! Thanks so much @Dan!