"yell 'I am free!'": invalid MRS? or alternative way to scope variables?

I was happy to see that there are parse variants for quoted text that don’t attempt to interpret the text semantically and just pass it through with a series of linked fw_seq and quoted predicates.

The parse of "yell 'I am free!'" that I want to use has this MRS (which I marked up using handle = [] to show one tree that I think was intended to be valid):

[ LTOP: h0 INDEX: e2 [ e SF: comm TENSE: pres MOOD: indicative PROG: - PERF: - ] 
RELS: < 
    h0 = [ pronoun_q<0:17> LBL: h4 ARG0: x3 [ x PERS: 2 PT: zero ] RSTR: h5 BODY: h6 ]  
        h5 = [ pron<0:17> LBL: h7 ARG0: x3 ]  
        h6 = [ proper_q<5:17> LBL: h9 ARG0: x8 RSTR: h10 BODY: h11 ]  
            h10 = [ fw_seq<5:10> LBL: h12 ARG0: x13 ARG1: i15 ARG2: i16 ]
                  [ fw_seq<5:17> LBL: h12 ARG0: x8 ARG1: x13 ARG2: i14 ]  
                  [ quoted<5:7> LBL: h12 CARG: "I" ARG0: i15 ]  
                  [ quoted<8:10> LBL: h12 CARG: "am" ARG0: i16 ]  
                  [ quoted<11:17> LBL: h12 CARG: "free" ARG0: i14 ] > 
            h11 = [ _yell_v_1<0:4> LBL: h1 ARG0: e2 ARG1: x3 ARG2: x8 [ x PERS: 3 NUM: sg ] ]
HCONS: < h0 qeq h1 h5 qeq h7 h10 qeq h12 > ICONS: < > ]

The problem is that there is no quantifier that binds x13. It is introduced by “plain old” predicate fw_seq<5:10> and then consumed by another predicate, fw_seq<5:17>, in the conjunction of predicates created by h12.

Are variables introduced by a predicate in a conjunction “bound/in scope” for others in that conjunction (in which case I should add a case to my resolver)? Or is this just an ERG bug?

Let me know if I should log a bug. For now I’m going to assume, based on this example, that this is a valid way to scope variables. So, any x variables not introduced by a quantifier are existentially quantified for the whole tree.

I’ll give two answers. Firstly, implicit existential quantification is already necessary for e variables. So formally speaking, this mechanism is already there.

But usually this isn’t done for x variables. In this case, I don’t think it’s really clear what the semantic representation should be, because the words are all quoted – should there just be one x for the whole sequence? One x for each token? So I’m not sure if this is a bug or a kind of placeholder analysis. The semantic documentation is very brief, but it does say “The ERG does not attempt to assign any linguistically motivated syntactic or semantic internal structure to foreign word sequences. Rather, they are represented with a subgraph which can then integrated into the larger MRS structure.”


So I would say, use the fw_seq and quoted predicates in whatever way is useful for you.

Thanks, that is the conclusion I came to as well and it works great for my scenarios.

But I still don’t understand how to make quoted expressions reliably use this fw_seq mechanism, especially with one word quotes like “yell ‘boo’”. Any suggestions there would be appreciated. I haven’t found any good workarounds

Maybe the interesting question is how ERG identifies a sequence of words as a foreign word sequences. I using ERG 2018, I got two complete different analysis for 1-2 vs. 3-4 (same for ")

  1. Abrams said ‘eu te amo’!
  2. Abrams said “eu te amo”!
  3. Abrams said ‘eu te amo!’
  4. Abrams said “eu te amo!”