DanielGallagherIRE/georgian-case-alignment

Georgian Case-Alignment Dataset

This dataset contains examples for testing Georgian case-alignment patterns across different verb series and argument structures. It can be used for Targeted Syntactic Evaluation of language models on Georgian case alignment and was put together using Grew-TSE applied to the Georgian Language Corpus UD Treebank. Additional word forms were added by a native speaker where necessary to ensure each test contained the nominative, dative, and ergative forms. A special thank you to Dr. Tamara Tartarashvili for her validation of the data.

Dataset Structure

This dataset contains multiple configurations, each representing a specific syntactic structure. All configurations include a training split with the corresponding data file.

Config Name	Description	Data File
intransitive-nom-subj	Sentences with intransitive verbs (no objects), where the nominative subject is tested.	intransitive-nom-subj/train.csv
transitive-nom-dat-subj	Transitive sentences with a nominative subject and dative object, where the subject is tested.	transitive-nom-dat-subj/train.csv
transitive-nom-dat-obj	Transitive sentences with a nominative subject and dative object, where the object is tested.	transitive-nom-dat-obj/train.csv
transitive-erg-nom-subj	Transitive sentences with an ergative subject and nominative object, where the subject is tested.	transitive-erg-nom-subj/train.csv
transitive-erg-nom-obj	Transitive sentences with an ergative subject and nominative object, where the object is tested.	transitive-erg-nom-obj/train.csv
transitive-dat-nom-subj	Transitive sentences with a dative subject and nominative object, where the subject is tested.	transitive-dat-nom-subj/train.csv
transitive-dat-nom-obj	Transitive sentences with an ergative subject and nominative object, where the object is tested.	transitive-dat-nom-obj/train.csv

Features

Each configuration contains the following fields:

• sentence_id (string): Sentence ID from the Georgian Language Corpus UD Treebank.
• match_id (int32): Target word is the ith word in the sentence, where punctuation is counted individually as words.
• masked_text (string): Sentence with masked argument.
• form_grammatical (string): Grammatically correct case form.
• form_ungrammatical_* (string): Ungrammatical case alternatives, one of nominative, dative, or ergative.

Note: Which form is grammatical and which ungrammatical depends on the specific verbal paradigm (Series 1,2, or 3).

Usage

from datasets import load_dataset

# Load a specific configuration
dataset = load_dataset("DanielGallagherIRE/georgian-case-alignment", "intransitive")

# Load all configurations
dataset = load_dataset("DanielGallagherIRE/georgian-case-alignment")