Actually, these days the knowledgebase is probably more valuable than the interpreter, or engine. Although, answer discovery does depend somewhat on the quality of the interpreter. If you’re serious, the ALICE A.I. Foundation Superbot (999 USD) is a good place to start; because, regardless of what interpreter or schema you use, most things most people will ask are already in the can. Otherwise, there are plenty of free AIML sets floating around.
In building any chatbot knowledgebase, think question generation and question-answer pairs. It’s generally not that difficult to move a good knowledgebase from one interpreter or schema to another. So, focus on your data and metadata, and don’t waste time re-inventing the wheel trying to develop new interpreters; move your data from one engine to another, until you find the one you like. Be aware though, that different interpreters, or engines, will have differing functionality.
Note with machine learning, what is most important are question-answer pairs. In chatbots this conforms most closely with chatlogs. In other words, today what is most valuable are neither the knowledge nor the engines of all old chatbots, but in fact their chatlogs, or record of conversations, in the form of question-answer pairs. What this means is that modern chatbots can be started with rule-based pattern-matching, but with enough usage can be transitioned to machine learning, based off the chatlogs.
See also my answer to: