Rachel Carrington (Internal Seminar) ----> 11th of February at 2pm 

Title: Semi-supervised word embeddings 

Abstract: Word embeddings are a popular way of modelling language, in which words are represented as low-dimensional vectors. The aim is that distances between vectors correspond to relationships between words: words with similar meanings should be closer together in the embedding space. Recently this has been a growing area of interest, with applications including sentiment analysis, machine translation, and artificial intelligence. However, a disadvantage is that large amounts of data are generally needed to train word embedding models. In this talk I will first give an overview of how word embeddings are generated, and then I will outline a novel method of generating semi-supervised word embeddings, based on multidimensional scaling, which has the potential to reduce the amount of data needed to generate accurate word embeddings.