We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.CL

Change to browse by:

cs

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computation and Language

Title: Interpreting Arabic Transformer Models

Abstract: Arabic is a Semitic language which is widely spoken with many dialects. Given the success of pre-trained language models, many transformer models trained on Arabic and its dialects have surfaced. While these models have been compared with respect to downstream NLP tasks, no evaluation has been carried out to directly compare the internal representations. We probe how linguistic information is encoded in Arabic pretrained models, trained on different varieties of Arabic language. We perform a layer and neuron analysis on the models using three intrinsic tasks: two morphological tagging tasks based on MSA (modern standard Arabic) and dialectal POS-tagging and a dialectal identification task. Our analysis enlightens interesting findings such as: i) word morphology is learned at the lower and middle layers ii) dialectal identification necessitate more knowledge and hence preserved even in the final layers, iii) despite a large overlap in their vocabulary, the MSA-based models fail to capture the nuances of Arabic dialects, iv) we found that neurons in embedding layers are polysemous in nature, while the neurons in middle layers are exclusive to specific properties.
Comments: 11 pages, 6 figures, 4 tables
Subjects: Computation and Language (cs.CL)
Cite as: arXiv:2201.07434 [cs.CL]
  (or arXiv:2201.07434v1 [cs.CL] for this version)

Submission history

From: Ahmed Abdelali [view email]
[v1] Wed, 19 Jan 2022 06:32:25 GMT (8920kb,D)

Link back to: arXiv, form interface, contact.