Google Translate works by tracking patterns of texts through a computer and translating the patterns. Accuracy of the results improves with each use of a language or new uploads of text because pattern recognition improves the more the program accesses the statistics of a language.
Google Translate software takes large amounts of text data and incorporates what Google calls "statistical machine translation." By using both monolingual text and aligned text, or human translations from one language to another, the software creates a translation model for each language. Google has uploaded multilingual documents and millions of pages of indexes to ensure the most accurate translations possible, although Google acknowledges that translations are not all 100 percent accurate.
One limitation to this approach is the fact that documents from some languages are not readily available, so Google Translate does not have complete or quality translations for those languages compared to more widely used languages. Another limitation is that data translations lack some of the finesse of human translations, especially for colloquial translations or languages used in smaller countries. However, the statistical translation offered by Google Translation has provided an option outside of rule-based translation that continuously improves over time, making accurate translations easier to access from anywhere.