One key point in the article hasn't been mentioned yet:
“Suddenly, I got the urge to do something similar for Android,” writes Dexetra on its blog. “Since we have been working on NLP [natural language processing] and Machine learning for over an year now, [emphasis mine] I had a crazy belief that I could pull this off."
So to re-phrase: It took them eight hours to put a Siri-like wrapper on a year+ long project. I'm not deriding their work on AI, NLP and machine learning - that stuff is devilishly hard - but it's disingenuous to claim they "cloned Siri" in eight hours. In eight hours they did the following:
- Created a pretty interface
- Utilized pre-existing voice-recognition and text-to-speech libraries for their input and output
- Hard-coded a few funny responses to impart some personality
- Linked to their year+ old NLP project
Similarly, here's what Apple did for Siri:
- Created a pretty interface
- Utilized (and likely refined) various voice-recognition and text-to-speech libraries.
- Likely influenced the personality of the product, including its various humorous responses
- Linked to (and likely refined) the NLP capabilities purchased from SRI, the development of which began back in 2005 it seems.
So Dexetra did in eight hours something that Apple probably could have done in eight hours as well after purchasing SIRI, but decided to sit on and work on for a while. The keys are - can they polish it enough to make a usable product, like Apple did? And, is their NLP back-end as good as SRI's?