When input text is first entered into ME, a SymbolGraph of characters is built. The individual character and its properties is called a Symbol and includes the whitespace following the character. A series of SymbolRules are run to determine as much information about a Symbol as possible. For instance, a period character '.' could also possibly be a decimal point, or indicate an abbreviation. The decimal needs to be considered as part of a number, while a period is given a property of 'IsToken' and a property that it represents an end mark. Symbol Rules do not change the structure of the SymbolGraph, but only add properties to describe the individual Symbols that will be used when building the TokenGraph. This information is then used for processing further down the Rule pipeline.

A TokenGraph is built on top of the SymbolGraph and is focused on determining the type and organization of a word or group of words. TokenRules takes infomation from the SymbolGraph and uses it to structure Symbol groups as Tokens. Unlike the SymbolGraph, Tokens can be built up of other Tokens and can represent Words, Numbers, multi-word Complex Words, Punctuation, or raw (unidentified) Text. TokenRules begin identifying the primary meanings of the Token and adds properties that help describe the additional qualities.

MeaningRules provide the last step of understanding for the Meaning of a Token. Much of the structural meaning of the Tokens will have been established by this point. This where we further narrow the most appropriate Meaning from the possible meanings that a Token can have. For instance, is the WordToken "HI" a greeting, or is being used in the context of the abbreviation of a US state?

Once the TokenGraph is complete and the most likely Meanings have been calculated, we can generate a Pattern of the structure of the Tokens. Based on the best match with the list of Patterns in our data store, an appropriate AnswerRule is returned. This rule can run any arbitrary code and will return an appropriate string response.