BLM-CausH (Blackbird Language Matrices Causative and Passive Alternation in Hebrew)

Description

BLM-CausH is a dataset in Modern Hebrew for learning the causative alternation, developed in the Blackbird Language Matrices (BLM) framework. In this task, an instance consists of sequences of sentences with specific attributes. To predict the correct answer as the next element of the sequence, a model must correctly detect the underlying generative rules used to produce the dataset.

The instantiated data are extracted from natural data extracted from two treebanks of Universal Dependencies of Hebrew containing respectively news (HBT v.2.15, Tsarfaty 2013; McDonald et al. 2013; 114,648 tokens, 6,143 trees) and encyclopaedic entries (IAHLTWiki v. 2.15, henceforth IW; Zeldes et al. 2022; 103,395 tokens; 5,039 trees). We collected sentences where the main verb is annotated with relevant the morphosyntactic property HEBBINYAN.

The data comes grouped by target voice, in two groups SENT (full sentences) and VERB (verb only) and each subset is split into train/test. The statistics of the current iteration of the dataset are (train:test split information):

Reference

If you use this dataset, please cite the following publication:

Giuseppe Samo, Paola Merlo, Modelling the Morphology of Verbal Paradigms: A Case Study in the Tokenization of Turkish and Hebrew, paper accepted at the SigTurk – SIGTURK 2026 Workshop