ADV -> ADJ -> N -> V

[kevincrawfordknight]

AMR stems down to representative concepts -- so, instead of "behavior", "behave", "behavioral", and "behaviorally", we just use "behave-01" in AMR. Then of course, there are many ways to realize (in English) an AMR that contains "behave-01".

The general plan is "ADV -> ADJ -> N -> V".

Between ADJ and N, some possible exceptions crop up. For example, I have seen AMR annotators render the word "patient" (English) as "patience" (in AMR) -- do we want to do this? I believe Ulf has seen several cases where the N is a morphological variant of the ADJ.

As long as we use one canonical concept, it doesn't matter, and we don't want to get caught up in English morphology. But we need a rule for deciding what the canonical concept is. "ADV -> ADJ -> N -> V" is one such rule, and we might decide to stick with it.

[uhermjakob]

Very good point. There are indeed many adjectives that should not be mapped to a noun such as happy, important, stable (not to be mapped to happiness, importance, stability).

If the noun is a tangible object or an agent, we probably want to continue to map from the adjective to the noun: microbial -> microbe, naval -> navy, Chinese -> China, organizational -> organization.

But for attributes such as sad, strong, important, stable, happy, free etc. we probably want to stay with the adjective.

In the vast majority of cases, the proper AMR concept will be the shorter word (sad/sadness -> sad; naval/navy -> navy), but there are some exceptions (German -> Germany, fame -> famous).

I've already started to structure my pertainym list to reflect a preferred direction. But additional insights are certainly welcome.

[nschneid]

Where an adjective adds a suffix to a noun (naval) or vice versa (happiness, stability) the rule should be clear: we prefer the version with less morphology. The tricky case noted above is where a root takes adjectival and nominal endings in alternation: patient~patience (denominal adjective or a deadjectival noun?). I think a good rule would be to prefer the noun in such a case.

A second issue is where the derivational morpheme adds more information than simply changing the part of speech: e.g. bliss~blissful or friend~friendly. In these cases I think we should keep the full adjective as the concept.

[uhermjakob]

For many of the words in the patient/patience class of adjective/noun pairs, annotators prefer the adjective form, e.g.

• important (195 times)
• importance (29 times)

and this despite of a guidelines bias towards nouns.

I agree with the majority of annotators that it is more natural to use the concept "important" when annotating phrases such as

• an important project
• The project is important.
• The project is of great importance.
• The president stressed the importance of the project. (The president stressed that the project is important.)

When trying to determine the more "basic" word, the sole reliance on English morphology can result in language-specific, idiosyncratic decisions, as illustrated by this example (French/English):

• beauté beauty (noun)
• beau beautiful (adjective)

"beauty" looks more basic in English, but not in French and other languages.

To support annotators in building consistent AMRs, I have been building adjective/noun lists and would like to share the following list as a proposal, with suggestions welcome: http://www.isi.edu/~ulf/amr/tmp/new-pert.txt which contains

• 2503 adjectives to be mapped to nouns
• 618 nouns to be mapped to adjectives
• 44 adjective/noun pairs to be mapped to verbs
• 45 adjective/noun pairs without a recommendation (often with a similar but distinct meaning)

Note that the adjective -> noun mapping still dominates, but with a sizeable minority for the opposite direction.

Applied to our current full corpus of AMRs, this produces the following recommendations: http://www.isi.edu/~ulf/amr/tmp/new-pert-stats.txt

Overall, when there is an adjective/noun pair in the new-pert.txt list, annotators already agree 89% with that list. For the rest, annotators have used the adjective form instead of the noun form, or vice versa, and mostly somewhat inconsistently.

[kevincrawfordknight]

love it

On May 23, 2013, at 8:53 PM, Ulf Hermjakob notifications@github.com wrote:

For many of the words in the patient/patience class of adjective/noun pairs, annotators prefer the adjective form, e.g.

important (195 times) importance (29 times) and this despite of a guidelines bias towards nouns.

I agree with the majority of annotators that it is more natural to use the concept "important" when annotating phrases such as

an important project The project is important. The project is of great importance. The president stressed the importance of the project. (The president stressed that the project is important.) When trying to determine the more "basic" word, the sole reliance on English morphology can result in language-specific, idiosyncratic decisions, as illustrated by this example (French/English):

beauté beauty (noun) beau beautiful (adjective) "beauty" looks more basic in English, but not in French and other languages.

To support annotators in building consistent AMRs, I have been building adjective/noun lists and would like to share the following list as a proposal, with suggestions welcome: http://www.isi.edu/~ulf/amr/tmp/new-pert.txt which contains

2503 adjectives to be mapped to nouns 618 nouns to be mapped to adjectives 44 adjective/noun pairs to be mapped to verbs 45 adjective/noun pairs without a recommendation (often with a similar but distinct meaning) Note that the adjective -> noun mapping still dominates, but with a sizeable minority for the opposite direction.

Applied to our current full corpus of AMRs, this produces the following recommendations: http://www.isi.edu/~ulf/amr/tmp/new-pert-stats.txt

Overall, when there is an adjective/noun pair in the new-pert.txt list, annotators already agree 89% with that list. For the rest, annotators have used the adjective form instead of the noun form, or vice versa, and mostly somewhat inconsistently.

— Reply to this email directly or view it on GitHub.

[nschneid]

Looks good to me! I think a generalization is: keep the adjective if it is more basic than the noun morphologically, is a "-ful" adjective, or belongs to an alternation like "-ant/ance", "-ent"/"-ency", "-ous/ion", etc.