Morphological Recursive Neural Network (MorphoRNN) Embedding System
Jump to navigation
Jump to search
A Morphological Recursive Neural Network (MorphoRNN) Embedding System is a Morpheme Embedding System that is based on the training of a Morphological Recursive Neural Network.
- Example(s):
- Counter-Example(s):
- See: Word Embedding System, Character Embedding System, OOV Embedding System, Subword Embedding System, Morpheme Vector, Morpheme Embedding Model, Natural Language Processing System, Word Detection System, Language Model, POS Tagging System.
References
2013
- (Luong et al., 2013) ⇒ Thang Luong, Richard Socher, and Christopher Manning. (2013). “Better Word Representations with Recursive Neural Networks for Morphology.” In: Proceedings of the Seventeenth Conference on Computational Natural Language Learning (CoNLL-2013).
- QUOTE: Our morphological Recursive Neural Network (morphoRNN) is similar to (Socher et al., 2011b), but operates at the morpheme level instead of at the word level. Specifically, morphemes, the minimum meaning-bearing unit in languages, are modeled as real-valued vectors of parameters, and are used to build up more complex words. We assume access to a dictionary of morphemic analyses of words, which will be detailed in Section 4.
Following (Collobert and Weston, 2008), distinct morphemes are encoded by column vectors in a morphemic embedding matrix $\mathbf{W}_e \in \R^{d\times |M|}, where $d$ is the vector dimension and $M$ is an ordered set of all morphemes in a language.
As illustrated in Figure 1, vectors of morphologically complex words are gradually built up from their morphemic representations (...).
- QUOTE: Our morphological Recursive Neural Network (morphoRNN) is similar to (Socher et al., 2011b), but operates at the morpheme level instead of at the word level. Specifically, morphemes, the minimum meaning-bearing unit in languages, are modeled as real-valued vectors of parameters, and are used to build up more complex words. We assume access to a dictionary of morphemic analyses of words, which will be detailed in Section 4.
|