Differing from typical memory networks, we extract three kinds of features to enrich the word representation of each context word.
The results suggest that the architecture underlying spoken word representation and processing is constant throughout development, even if some differences between children and adults emerged.
There are two main objectives: how well do popular deep learning architectures, namely LSTM and CNNs, perform on gender classification task and investigate how the choice of word representation effects the performance.
Pennington, J., Socher, R. & Manning, C. Glove: global vectors for word representation. in Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) 1532 1543 (ACL, 2014).
Three main steps are considered: word segmentation, word representation, and word weight calculation.
The proposed method includes extraction of the baseline and the word representation feature sets.
To achieve this goal, word representation through word vector and weight calculation method are introduced.
WR features: word representation features generated by Brown clustering [24], random indexing [25] and skip-gram [26].We followed the same method as in [31] to generate unsupervised word representation features using these three methods.
The generally applicable word representation features were reported to boost system performance significantly for both chemical and biomedical NER.
We observed that the whitespace tokenizer performs better than the BANNER simple tokenizer for extraction of word representation features.
For word representation features, we train Brown clustering models [28] and Word Vector (WV) models [17] on a large PubMed and PMC document collection.
