How to consistently get a parse of "quoted phrases" that uses `fw_seq` with `quoted` predicates?

Based on this page, I was hoping that any phrase with quoting (e.g. “say ‘how are you’” or “she said ‘wow’”) would generate at least one parse using a combination of fw_seq with quoted predicates for the phrase in quotes. Any tips for how to do this? Is it possible that I’m not using the right quote marks or something?

Here’s what I’ve found with testing:

It appears that one word quotes mostly end up using fw_seq[xe] with no quoted predicate like this parse of “say ‘stop’”:

[ proper_q LBL: h5 ARG0: x4 [ x PERS: 3 NUM: sg ] RSTR: h6 BODY: h7 ]
[ fw_seq LBL: h8 ARG0: x4 [ x PERS: 3 NUM: sg ] ARG1: e9 ]
[ _stop_v_1 LBL: h8 ARG0: e9 [ e SF: prop TENSE: untensed MOOD: indicative PROG: - PERF: - ] ARG1: i10 ARG2: i11 ]
[ _say_v_1 LBL: h1 ARG0: e2 [ e SF: prop-or-ques TENSE: tensed MOOD: indicative PROG: - PERF: - ] ARG1: i3 ARG2: x4 ]

And anything with more than one word mostly ends up using fw_seq and quoted like this parse of “say ‘stop thief’”:

[ proper_q LBL: h5 ARG0: x4 [ x PERS: 3 NUM: sg ] RSTR: h6 BODY: h7 ]
[ fw_seq LBL: h8 ARG0: x9534 [ x ] ARG1: i9 ARG2: i10 ]
[ quoted LBL: h8 CARG: “stop” ARG0: i9 [ i ] ]
[ quoted LBL: h8 CARG: “thief” ARG0: i10 [ i ] ]
[ unknown LBL: h1 ARG0: e2 [ e SF: prop-or-ques ] ARG: x4 ]

But there are exceptions to both. Some examples I’ve tried generate combinations of the two approaches: "say ‘yell stop’"puts stop as quoted but yell as a real predicate. Makes it tough to handle quotes easily…

Note: I’ve tried to trick the system into thinking the quoted phrase was “foreign” by putting garbage in the front. E.g. changing “say ‘stop’” to “say ‘adaf stop’” but that just gets parses with nn_u_unknown__x in them

Belatedly following up on this thread: There were indeed inconsistencies in how quoted words and phrases were treated in the 2018 release of the ERG. I have made some improvements in the soon-to-be-declared 2020 release of the grammar, so I would encourage you to have a look once I get it announced.

Awesome! Looking forward to it. Thanks @Dan!

For others that might want to get the quoted phrases in 2018 to consistently generate fw_seq parses, I found a workaround that seems to work: It appears that phrases in quotes with less than three words don’t reliably get a FW parse. I also found that replacing spaces with dashes seemed to force the FW parse. So I did this as a workaround:

Preprocess the string before sending to Ace like this:

  1. Find all quoted strings in the utterance
  2. If there are less than 3 words add enough “noop” words to make it 3
  3. Replace all spaces with “-”
  4. Send the string to Ace
  5. In the MRS: Ignore the trailing “-” on any word in a FW predicate, and ignore any “noop” words as well

So the string “Stop Thief” would get converted to “Stop-Thief-noop” before sending through Ace.

FWIW