Large Pre-Trained Models with Extra-Large Vocabularies: A Contrastive Analysis of Hebrew BERT Models and a New One to Outperform Them All
This work addresses performance issues in Hebrew natural language processing by proposing a model with a larger vocabulary, which is incremental as it builds on existing pre-trained language models.
The authors tackled the problem of improving Hebrew language models by introducing AlephBERTGimmel, a model with a much larger vocabulary (128K items) that reduces token splits, and it achieved new state-of-the-art results across multiple Hebrew benchmarks, including Morphological Segmentation and NER.
We present a new pre-trained language model (PLM) for modern Hebrew, termed AlephBERTGimmel, which employs a much larger vocabulary (128K items) than standard Hebrew PLMs before. We perform a contrastive analysis of this model against all previous Hebrew PLMs (mBERT, heBERT, AlephBERT) and assess the effects of larger vocabularies on task performance. Our experiments show that larger vocabularies lead to fewer splits, and that reducing splits is better for model performance, across different tasks. All in all this new model achieves new SOTA on all available Hebrew benchmarks, including Morphological Segmentation, POS Tagging, Full Morphological Analysis, NER, and Sentiment Analysis. Subsequently we advocate for PLMs that are larger not only in terms of number of layers or training data, but also in terms of their vocabulary. We release the new model publicly for unrestricted use.