A transformer is a neural network – a set of (often complex) algorithms. Most of these algorithms aren’t directly related to linguistics – but probability.
For example, the TF-IDF algorithm (term frequency, inverse document frequency) finds the value of the TF (the frequency of a specific term in a given text) divided by the inverse of the DF (the number of total corpuses containing that term). The higher this value, the more relevant a term is likely to be.
A transformer combines the results of many of these algorithms in order to “comprehend” a given text, and to attempt to produce a relevant response.
Latest Answers