Create dataset loader for UP2.0

[SamuelCahyawijaya]

Dataloader name: up2/up2.py
DataCatalogue: http://seacrowd.github.io/seacrowd-catalogue/card.html?up2

Dataset	up2
Description	Southeast Asian language subsets from Universal Propositions (UP) 2.0 dataset. Semantic role labeling (SRL) is a shallow semantic parsing task that identifies “who did what to whom when, where etc” for each predicate in a sentence. It provides an intermediate (shallow) level of a semantic representation that helps the map from syntactic parse structures to more fully-specified representations of meaning.
Subsets	ind, vie
Languages	ind, vie
Tasks	Semantic Role Labeling
License	Community Data License Agreement – Permissive, Version 1.0 (cdla-permissive-1.0)
Homepage	https://github.com/UniversalPropositions/UP_Indonesian-GSD, https://github.com/UniversalPropositions/UP_Vietnamese-VTB
HF URL	-
Paper URL	https://aclanthology.org/2022.lrec-1.181.pdf

[fhudi]

#self-assign

[fhudi]

@SamuelCahyawijaya @holylovenia Hi, I have some doubts here:

1. There is no Tasks.SEMANTIC_ROLE_LABELING, shall I make it as a sequential labeling schema (token-to-label)?
   Does anyone know where to get the whole set of roles?

2. It is necessary to merge with UD to get the token, but it seems the original tools for merging requires additional library. Shall I use try-catch or is it better to re-implement the merging with UD, i.e. mapping ID to get only the sequence of token?

[fhudi]

For the time being, I implemented this as source-only.
Please advice when possible.
@SamuelCahyawijaya @holylovenia

[holylovenia]

@SamuelCahyawijaya @holylovenia Hi, I have some doubts here:

1. There is no Tasks.SEMANTIC_ROLE_LABELING, shall I make it as a sequential labeling schema (token-to-label)?
   Does anyone know where to get the whole set of roles?
2. It is necessary to merge with UD to get the token, but it seems the original tools for merging requires additional library. Shall I use try-catch or is it better to re-implement the merging with UD, i.e. mapping ID to get only the sequence of token?

Hi @fhudi, sorry for the late response.

1. We can do a source-only schema for this dataloader. What do you mean by "the whole set of roles"?
2. Have you taken a look at common_parser.py and if it can help or not? If it's insufficient, we can use try-except to import the additional library.

[fhudi]

Thanks @holylovenia for the reply.

1. We can do a source-only schema for this dataloader. What do you mean by "the whole set of roles"?
2. Have you taken a look at common_parser.py and if it can help or not? If it's insufficient, we can use try-except to import the additional library.

1. The semantic roles (column 11 of UP).

2. Yeah, I wrote that parser. I do believe the common_parser.py suffices.
   The reason I highlighted this is due to not following the steps from original UP dataset reconstruction.

Tag: @ijindal
Could you please clarify these 2 points?

[ijindal]

@fhudi
The whole set of roles in UP means all EN propbank rolesets. Be careful though all the roleset are consistent with V3.0 of EN propbank. The current labeling schema may not work with the current version of EN propbank which is v3.4.

• Yes, But the steps mentioned are specific to UP2.0 datasets not for semantic role labeling in general.