Detecting vocabulary skill level and correcting misalignment in remote interactions

A communication is received comprising a customer communication device utilized by a customer and an automated resource holding the communication. A customer language proficiency rating of the customer is accessed. A pool of language proficiency ratings associated with each agent in a pool of agents is also accessed. An agent is selected from the pool of agents that has an associated language proficiency rating that best matches the customer language proficiency rating, and the communication is connected to comprise an agent communication device associated with the selected agent. Accessing the customer language proficiency may be achieved by scoring prior communication with the customer against a scoring model. A further aspect of the invention determines whether the agent language proficiency is dissimilar from the customer language proficiency and if so then the agent communication device presents an indication of this dissimilarity. Another aspect of the invention provides training of a neural network for language proficiency utilising a first set of words, a set of words that are synonyms of the first set and a set of words that have a different meaning to the first set.