"'desk' is countable" generates proper_q

If I use the ACE parser with erg-2018-osx-0.9.31 to parse the phrase:

'desk' is countable

Both parses use proper_q as the quantifer. At best I expected undef_q. Is this a bug or a feature? And if a feature, what does it mean?

[ TOP: h0
INDEX: e2
RELS: < 
[ proper_q__xhh LBL: h4 ARG0: x3 [ x PERS: 3 NUM: sg ] RSTR: h5 BODY: h6 ]
[ _desk_n_1__x LBL: h7 ARG0: x8 [ x PERS: 3 NUM: sg IND: + ] ]
[ fw_seq__xx LBL: h7 ARG0: x3 [ x PERS: 3 NUM: sg ] ARG1: x8 ]
[ _countable_a_1__ex LBL: h1 ARG0: e2 [ e SF: prop TENSE: pres MOOD: indicative PROG: - PERF: - ] ARG1: x3 ]
>
HCONS: < h0 qeq h1 h5 qeq h7 > ]

[ TOP: h0
INDEX: e2
RELS: < [ 
proper_q__xhh LBL: h7 ARG0: x6 [ x PERS: 3 NUM: sg ] RSTR: h8 BODY: h9 ]
[ _desk_n_1__x LBL: h10 ARG0: x11 [ x PERS: 3 NUM: sg IND: + ] ]
[ fw_seq__xx LBL: h10 ARG0: x6 [ x PERS: 3 NUM: sg ] ARG1: x11 ]
[ implicit_conj__eee LBL: h1 ARG0: e2 [ e SF: prop-or-ques TENSE: tensed MOOD: indicative ] ARG1: e4 ARG2: e5 ]
[ unknown__ex LBL: h1 ARG0: e4 [ e SF: prop-or-ques TENSE: tensed MOOD: indicative ] ARG: x6 ]
[ _countable_a_1__ei LBL: h1 ARG0: e5 [ e SF: prop-or-ques TENSE: pres MOOD: indicative PROG: - PERF: - ] ARG1: i12 ]
>
HCONS: < h0 qeq h1 h8 qeq h10 > ]

I reckon it’s a feature, since the grammar is working as intended. The ERG deals with the use vs. mention issue by taking the paired quotes as indication that the quoted expression does not have its ordinary meaning, so it treats the expression (in your example a single token) as the name of a linguistic entity, with the quoted word’s predicate an argument of the general relation “fw_seq” (short for “foreign word sequence”, even though in this case the single token happens to be from the same language). Since it’s viewed as a named entity, the use of the same quantifier proper_q as that used for ordinary names should not be surprising, even if somewhat arbitrary.

1 Like

Fair enough. I was definitely thinking of proper_q as only for proper nouns.

Follow-on question then: If I run “‘water’ is countable” through ACE, the first two parses are below (there are many more). Why doesn’t Ace give “‘desk’ is countable” a parse like the first udef_q parse below? It seems like it would be a valid interpretation.

[ TOP: h0
INDEX: e2
RELS: < 
[udef_q__xhh LBL: h4 ARG0: x3 [ x PERS: 3 NUM: sg ] RSTR: h5 BODY: h6 ]
[ _water_n_1__x LBL: h7 ARG0: x3 [ x PERS: 3 NUM: sg ] ]
[ _countable_a_1__ex LBL: h1 ARG0: e2 [ e SF: prop TENSE: pres MOOD: indicative PROG: - PERF: - ] ARG1: x3 ]
>
HCONS: < h0 qeq h1 h5 qeq h7 > ]

[ TOP: h0
INDEX: e2
RELS: < [ 
proper_q__xhh LBL: h4 ARG0: x3 [ x PERS: 3 NUM: sg ] RSTR: h5 BODY: h6 ]
[ _water_n_1__x LBL: h7 ARG0: x8 [ x PERS: 3 NUM: sg ] ]
[ fw_seq__xx LBL: h7 ARG0: x3 [ x PERS: 3 NUM: sg ] ARG1: x8 ]
[ _countable_a_1__ex LBL: h1 ARG0: e2 [ e SF: prop TENSE: pres MOOD: indicative PROG: - PERF: - ] ARG1: x3 ]
>
HCONS: < h0 qeq h1 h5 qeq h7 > ]

More importantly for my purposes: I’ve never been able to get the foreign quoting feature to work like I’d expect. I assumed that anything in quotes that wasn’t understood would be turned into a series of fw* and quoted__ci predicates. But it is really inconsistent. For example:

'maison' is countable gives one parse that looks very different from the ones above. I was hoping that it would package all the foreign words nicely into something like the fw_seq__xxi/fw_seq__xii and quoted__ci predicates along with the proper_q__xhh and _countable_a_1__ex predicates as above. But instead it doesn’t even appear to use that rule and instead gives:

[ TOP: h0
INDEX: e2
RELS: < 
[ implicit_conj__eee LBL: h1 ARG0: e2 [ e SF: prop-or-ques TENSE: tensed MOOD: indicative ] ARG1: e4 ARG2: e5 ]
[ unknown__eu LBL: h1 ARG0: e4 [ e SF: prop-or-ques TENSE: tensed MOOD: indicative ] ARG: u6 ]
[ _maison/jj_u_unknown__eu LBL: h1 ARG0: e7 [ e SF: prop ] ARG1: u6 ]
[ _countable_a_1__ei LBL: h1 ARG0: e5 [ e SF: prop-or-ques TENSE: pres MOOD: indicative PROG: - PERF: - ] ARG1: i8 ]
>
HCONS: < h0 qeq h1 > ]

What is the underlying model for when the foreign word sequence rules kick in?

You won’t get the udef_q interpretation with desk' that you get with water’ because that interpretation is with a parse that essentially ignores the single quotes, so the analysis is equivalent to the sentence
water is countable
This is a good sentence (though with a silly reading) because “water” is a mass noun that does not require an article such as “the”. In contrast, “desk” is a count noun, requiring an article if it is singular, so the following is not grammatical:
desk is countable.
Of course, we humans can interpret this sentence the same as the original one with quotes, but the ERG really needs the quotes to signal that “desk” is being used as the name of the word. It would be too expensive in parsing to consider treating every word in every sentence both with its normal meaning and with the name-of-the-word meaning.

1 Like

Ahh, I get it. Great. Any thoughts on my other foreign word sequence question above? That one has been driving me nuts for a long time…