This paper introduces a novel approach to improve word representations by leveraging recursive neural networks (RNNs) for morphology. The proposed model, called morphoRNN, builds representations for morphologically complex words from their morphemes, combining RNNs with neural language models (NLMs) to capture contextual information. The model outperforms existing word representations on word similarity tasks across multiple datasets, including a newly introduced dataset focused on rare words. The key idea is to treat each morpheme as a basic unit in the RNNs and construct representations for complex words on the fly from their morphemes. By training a neural language model and integrating RNN structures for complex words, the model utilizes contextual information to learn morphemic semantics and their compositional properties. This approach allows the model to build representations for any new unseen word composed of known morphemes, providing an infinite (though incomplete) vocabulary coverage. The paper also discusses related work, highlighting the limitations of existing word representations in capturing morphological relationships and the need for models that can handle rare and complex words. The proposed morphoRNN model addresses these limitations by explicitly modeling morphological structures, leading to better performance in capturing both syntactic and semantic information. Experiments show that the morphoRNN model outperforms existing embeddings on word similarity tasks, particularly on datasets with rare words. The model's ability to blend syntactic and semantic information is demonstrated through analysis of nearest neighbors for various complex words. The paper also introduces a new dataset focused on rare words to complement existing ones, highlighting the importance of considering rare words in word representation learning. The model is evaluated on multiple datasets, including WS353, MC, RG, SCWS*, and the newly introduced rare word dataset. Results show that the morphoRNN model, especially the context-sensitive version (csmRNN), significantly outperforms existing embeddings in capturing word similarity. The model's ability to handle rare and complex words is attributed to its explicit modeling of morphological structures and integration of contextual information. The paper concludes that the proposed approach provides a more principled way of handling unseen words and improves the representation of rare and complex words. The model's effectiveness is demonstrated through its performance on word similarity tasks and its ability to blend syntactic and semantic information in clustering related words. The authors also highlight the potential of the model for application in other morphologically complex languages and domains.This paper introduces a novel approach to improve word representations by leveraging recursive neural networks (RNNs) for morphology. The proposed model, called morphoRNN, builds representations for morphologically complex words from their morphemes, combining RNNs with neural language models (NLMs) to capture contextual information. The model outperforms existing word representations on word similarity tasks across multiple datasets, including a newly introduced dataset focused on rare words. The key idea is to treat each morpheme as a basic unit in the RNNs and construct representations for complex words on the fly from their morphemes. By training a neural language model and integrating RNN structures for complex words, the model utilizes contextual information to learn morphemic semantics and their compositional properties. This approach allows the model to build representations for any new unseen word composed of known morphemes, providing an infinite (though incomplete) vocabulary coverage. The paper also discusses related work, highlighting the limitations of existing word representations in capturing morphological relationships and the need for models that can handle rare and complex words. The proposed morphoRNN model addresses these limitations by explicitly modeling morphological structures, leading to better performance in capturing both syntactic and semantic information. Experiments show that the morphoRNN model outperforms existing embeddings on word similarity tasks, particularly on datasets with rare words. The model's ability to blend syntactic and semantic information is demonstrated through analysis of nearest neighbors for various complex words. The paper also introduces a new dataset focused on rare words to complement existing ones, highlighting the importance of considering rare words in word representation learning. The model is evaluated on multiple datasets, including WS353, MC, RG, SCWS*, and the newly introduced rare word dataset. Results show that the morphoRNN model, especially the context-sensitive version (csmRNN), significantly outperforms existing embeddings in capturing word similarity. The model's ability to handle rare and complex words is attributed to its explicit modeling of morphological structures and integration of contextual information. The paper concludes that the proposed approach provides a more principled way of handling unseen words and improves the representation of rare and complex words. The model's effectiveness is demonstrated through its performance on word similarity tasks and its ability to blend syntactic and semantic information in clustering related words. The authors also highlight the potential of the model for application in other morphologically complex languages and domains.