Unification of verb, noun, and adjective framefiles in PropBank

1 Purpose:

Unify derived words that share the same base form under a single framefile. For example, verb and noun forms of fear and adjective afraid will be united under a unified framefile "fear". Consequently, we are doing away with the -n, -v, and -j suffixes in the the framefile names.

2 Procedure:

There are two levels of unification involved in this procedure: (1) the unification of the current verb, noun, and adjective framefiles into a single framefile, and (2) the unification of the framesets within the framefiles that share the same argument structure.

2.1 Unification of framefiles Framefiles of the framesets sharing the same etymological base form will be joined under a same framefile regardless of the semantic differences in the different derivative forms.

Let's say we currently have:

 lemmaX-v has 3 senses 01,02, and 03
 lemmaX-j has 2 senses 01 and 02
 lemmaX-n has 1 sense 01

Each of the framesets within the framefiles will be entered as a distinct sense under the unified framefile (assuming no frameset unification; see section 2.2). Hence, in the above example the unified framefile lemmaX will have 6 framesets.

Lemma naming scheme hierarchy: verb > noun > adjective. If a verb framefile is currently available, the unified framefile will take the verb form as its lemma. If the verb framefile is not, then the unified framefile will take the noun form as its lemma.

 e.g. 
 fear-v, fear-n, and afraid-j will be united under fear
 promise-n and promise-v would be united under promise
 destruction-n and' 'destroy-v would be united under destroy
 good-j will be united (with itself) under good

 (to consider: take-v and takeover-n)

Handling of framesets: Barring any specification for frameset unification (see section 2.2), the framesets under the framefiles will be enumerated under their respective predicates.

 e.g.

Handling of predicates: Barring any specification for frameset unification (see section 2.2), VPCs or idiomatic expressions that have a separate predicate within the framefile will retain their predicate status in the unified framefiles. If there are

 e.g. Currently write-n and write-v have the predicate "write_up". 
 The framesets from these two files will be enumerated under the predicate "write_up" in the write framefile 
 (assuming no frameset unification specified).

2.2 Unification of framesets

During the framefile unification process, any framesets that are related in argument structure will be manually identified and evaluated then unitifed under a single frameset within the unified framefile. For example:

Let's say we currently have:

 lemmaX-v has 3 senses 01,02, and 03
 lemmaX-j has 2 senses 01 and 02
 lemmaX-n has 1 sense 01

Let's also say that lemmaX-j.01 has been identified to have the same semantics and semantic argument structure as as lemmaX-v.01 and that lemmaX-n.01 can be easily unfied also with lemmaX-v.01. This will result in the unified framefile lemmaX with 4 senses.

 lemmaX.01: unified frameset of previous lemmaX-v,-j,-n 01s
 lemmaX.02: previous lemmaX-v.02 
 lemmaX.03: previous lemmaX-v.03  
 lemmaX.04: previous lemmaX-j.02

Notekeeping in progress: Currently the annotators who are evaluating the relationships between the framesets are keeping notes on which framesets should be unified and what the unified frameset should look like.

3 Aliasing:

text

---

Considerations:

Automatic: - Framenet mapping: framenet_name.pos space delimited - Join in all of the VN cls specification -