The UMR guidelines discuss degree constructions in Part
3-3-6.
That part focus on degree adverbs like
very and somewhat, and on their morphological counterparts in other
languages. The keyword degree
is used in two different ways:
- If there is a lexical degree adverb, it is used as a child node of
the relation
:degree
(they refer to it as “attribute” here but it should probably be a “relation” when it gets child concepts).
velmi velký “very large”
(v/ velký-001 'large'
:degree (v2/ velmi 'very'))
The AMR guidelines provide more examples, as, e.g., I hardly know her.
(k / know-01
:ARG0 (i / i)
:ARG1 (s / she)
:degree (h / hardly))
- When the degree is expressed morphologically,
:degree
becomes an attribute with one of the valuesIntensifier
,Downtoner
,Equal
. The example is from Sanapaná:
ak-yav-ay'-a “very large”
(e/ enyavay'a-00 'large'
:degree Intensifier)
Compare example 3-3-6 (1) from the Guidelines -- the first part represents the annotation for the morphologically expressed “very” intensifier (in Sanapaná), the second one is for the lexical degree adverb very (in English).
ak-yav-ay'-a=ngkoye yamet
2/3F-be.large-TI-PFV.NMLZ=INTNS tree
'The tree is very large.'
(h/ have-mod-91
:ARG1 (y/ yamet 'tree')
:ARG2 (e/ enyavay'a-00 'large'
:degree Intensifier)
:aspect State
:modstr FullAff)
(h/ have-mod-91
:ARG1 (t/ tree)
:ARG2 (l/ large
:degree (v/ very))
:aspect State
:modstr FullAff)
Lexicon: Very - Intensifier
They suggest that in Czech, the lexicon will explain that velmi is an
Intensifier
. (They actually show it on the example of English very.)
While the UMR guidelines do not analyze comparison, the AMR guidelines do
provide some examples.
Since UMR does not negate the AMR approach, maybe it is done the same way
there. AMR employs the abstract event have-degree-91
.
The UMR guidelines mention this abstract event only in one example in Part
3-1-3-8. However, the abstract predicate have-degree-91
is listed in the UMR abstract roleset list, with tyhe following set of roles:
Have-degree-91
Arg1: domain, entity characterized by attribute (e.g. girl)
Arg2: attribute (e.g. tall)
Arg3: degree itself (e.g. more, most, less, least, times, equal, enough, too, so, to-the-point, at-least, times)
Arg4: compared-to (e.g. (than the) BOY)
Arg5: superlative: reference to superset
Arg6: reference, threshold of sufficiency (e.g. (tall enough, to ride the rollercoaster) )
the most expensive dress
(d/ dress
:ARG1-of (h/ have-degree-91
:ARG2 (e/ expensive)
:ARG3 (m/ most)))
The superlative in Czech is morphological but the :degree Intensifier
attribute would not be appropriate here because nejvíce is not the same
concept as velmi. Also, it is not clear how we would use the attribute
instead of the have-degree-91
event.
nejdražší šaty “the most expensive dress”
(š/ šaty
:ARG1-of (h/ have-degree-91
:ARG2 (d/ drahý-001)
:ARG3 (n/ nejvíce)))
AMR uses :ARG3 and :ARG4 of have-degree-91
to express the standard of
comparison.
According to the AMR Guidelines, "Annotators are encouraged to use have-degree-91 as the root concept (...) when a comparison seems to be the main focus of the sentence, which include cases of the copular construction (e.g., the girl is taller than the boy, she is the tallest girl on the team—see below, see below).
dívka je vyšší než chlapec “the girl is taller than the boy”
(h/ have-degree-91
:ARG1 (d/ dívka)
:ARG2 (v/ vysoký-001)
:ARG3 (v2/ více)
:ARG4 (ch/ chlapec))
AMR also uses :ARG6 to express purpose or result (it is small enough to fit in your pocket, he was so tired that he slept for 10 hours).
Problem: :degree
vs. have-degree-91
:
Where we should use the :degree
relation and where the have-degree-91
predicate?
Possible solution: Reserve have-degree-91
only for explicitly expressed comparisons and superlatives (or for cases annotated as comparison in PDT).
Feedback from Julia: I use these in the same way you mentioned above. For sentences like He was the tallest boy in the room, I would use have-degree-91 as the top node of the graph and treat it as an event, with :aspect/:modstr/:temporal dependency.
But for a sentence like she ate the biggest banana of the bunch, I would use have-degree-91 so that I could capture the superset entity as a numbered argument (the bunch), but I wouldn't treat it as an event.
The counterexamples in
AMR indicate that the
abstract event have-degree-91
should not be used when comparing quantities, as in I need more money or Most students don't like math.
On the other
hand, the way the AMR examples are annotated probably is not right in UMR,
where we have the :quant
attribute / relation. See
quantity.md.
However, AMR also offers the have-quant-91
abstract predicate which should be used in a parallel way as have-degree-91
for comparison of qualities of things (i.e., if some amount is compared to some reference amount), see Quantity.