en: "nothing but N" vs. cs: "nic než N"

[nesor]

I came across the Czech equivalent of English nothing but ... as in He inherited nothing but debts. The single example in Czech PDT has než as deprel=mark & upos=SCONJ, while the following NOUN is deprel=advcl. This is how UDPipe parses similar Czech sentences.

UD 2.3 - Czech PDT: https://lindat.mff.cuni.cz/services/kontext/view?ctxattrs=word%2Cdeprel&attr_vmode=mixed&pagesize=40&refs=%3Ddata.id&q=~shO32i5rtQLq&viewmode=kwic&attrs=word%2Cdeprel&corpname=ud_23_cs_pdt_a&attr_allpos=all

Is this intended? I'm not happy mainly about the NOUN ending up as deprel=advcl. There's nothing clausal about it, is there? There are more English examples with nothing but in one of the treebanks, and they are consistently annotated in a more intuitive way: but is deprel=cc & upos=ADP while the following NOUN is deprel=conj:

UD 2.3 - English EWT: https://lindat.mff.cuni.cz/services/kontext/view?ctxattrs=word&attr_vmode=mouseover&pagesize=40&refs=%3Ddata.id&q=~HPQO0Pl1HrNb&viewmode=kwic&attrs=word&corpname=ud_23_en_ewt_a&attr_allpos=all

I'd be happy just to learn that the Czech structure should be annotated more like its English equivalent.

Thanks!

[martinpopel]

I am not sure what should be the correct annotation in Czech nor English. Just a few comments:

• The particular Czech sentence is a bit unusual because it contains a colon - "Nemyslíš na nic než: světlo." It would be more usual (and fully grammatical) without the colon. The colon may suggest there is an elided clause (at least, I don't see any other reason for the colon).
• There is no other occurrence of "nic než" in the Czech UD data. According to SYN2020, the frequency of "nic než" is about 8 instances per million (tokens). So this bigram is slightly underrepresented in Czech UD data (with over 1.5M tokens), but still it is perhaps too rare for any special treatment being mentioned in the guidelines.
• Czech "než" means "than" and most (>96%) occurrences have upos=SCONJ and deprel=mark. There is only a single occurrence of "než" with deprel=case in UD_Czech-PDT v2.8 despite this is the recommended annotation for comparatives.
• I can imagine "nic než" being treated as deprel=fixed, but the meaning is quite compositional and predictable (nothing other than), so if deprel=fixed is rejected I don't see much reasons for annotating this kind of "než" in a different way than other "než" in typical comparative constructions.

[nesor]

Hi Martin,

Thanks for your feedback! I know it's not one of the most common structures, but the annotation of the noun following než as advcl struck me as inappropriate. It was a colleague of mine, interested in comparing finite and non-finite structures across several languages and using UDPipe to parse her texts, who noticed that the NOUN following než is an advcl. It's actually not easy at all to specify rules distinguishing finite and non-finite clauses (or their heads) in UD, and some of the rules must be language specific. Cases like this one make it even harder.

Best,

Saša (Rosen)

[dan-zeman]

The conversion from PDT to UD seems to be buggy in this respect: if the phrase introduced by než is a simple nominal, it shouldn't be treated as a clause (i.e., než should be attached as case, and the nominal should be obl rather than advcl). If it is not a nominal, it should be treated as an adverbial clause, even if reduced. But in the converted data, it seems to be treated as advcl (or as dep) even for nominals.

The single occurrence of nic než “nothing but” could be regarded as an ellipsis of the more frequent nic jiného než “nothing else than”, which is one of the documented comparative structures. So I do not think we need a coordination analysis here.

[dan-zeman]

• 3 occurrences in CAC: http://hdl.handle.net/11346/PMLTQ-1TSO
• 1 occurrence in PDT: http://hdl.handle.net/11346/PMLTQ-QA5U
• no occurrences in the other Czech treebanks