# Word-Embeddings Distinguish Denominal and Root-Derived Verbs in Semitic

arXiv: https://arxiv.org/abs/2208.05721 · HTML: https://mlsanity.com/paper/2208.05721/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 9 of its 39 references that are in the map.
- First posted on arXiv 11 Aug 2022.

## Builds on (most cited first, 9 of 9)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [Efficient Estimation of Word Representations in Vector Space](https://mlsanity.com/paper/1301.3781/) (2013)
- [Enriching Word Vectors with Subword Information](https://mlsanity.com/paper/1607.04606/) (2016)
- [Learning Word Vectors for 157 Languages](https://mlsanity.com/paper/1802.06893/) (2018)
- [Analogies Explained: Towards Understanding Word Embeddings](https://mlsanity.com/paper/1901.09813/) (2019)
- [A Survey on Contextual Embeddings](https://mlsanity.com/paper/2003.07278/) (2020)
- [AlephBERT:A Hebrew Large Pre-Trained Language Model to Start-off your Hebrew NLP Application With](https://mlsanity.com/paper/2104.04052/) (2021)
- [Contextualized Word Embeddings Encode Aspects of Human-Like Word Sense Knowledge](https://mlsanity.com/paper/2010.13057/) (2020)

## Abstract (arXiv)

Proponents of the Distributed Morphology framework have posited the existence of two levels of morphological word formation: a lower one, leading to loose input-output semantic relationships; and an upper one, leading to tight input-output semantic relationships. In this work, we propose to test the validity of this assumption in the context of Hebrew word embeddings. If the two-level hypothesis is borne out, we expect state-of-the-art Hebrew word embeddings to encode (1) a noun, (2) a denominal derived from it (via an upper-level operation), and (3) a verb related to the noun (via a lower-level operation on the noun's root), in such a way that the denominal (2) should be closer in the embedding space to the noun (1) than the related verb (3) is to the same noun (1). We report that this hypothesis is verified by four embedding models of Hebrew: fastText, GloVe, Word2Vec and AlephBERT. This suggests that word embedding models are able to capture complex and fine-grained semantic properties that are morphologically motivated.