In the end of last year we (Speaktoit) have launched our new speech interaction platform allowing to design voice and text-based agents for multiple different platforms. We support speech recognition, NLU, context handling, dialog management.
api.ai also powers our first product - Assistant, used by more than 17 million users.

Feedback of Chatbots community is really appreciated!

Nicely done, and great to see Cordova support!!

So I don’t see anything in the dev guides (quick glance) around NLU. How do you handle semantic reasoning? CCG, FOL, DRT?

Best
Rob Ellis

@Ilya
I have seen and browsed your API, seems to be pretty complete, but… I still do not grasp howdy you get this done in so many languages, you only offer English? ¿do you use automated translation?
¿Does the API offer how to handle the context of a conversation, may be complex to hold all this in a real dialog system!?

BTW: I have developed a open dialog language to build multilingual Dialog Avatars, and still cannot let the system program itself, nor do I lack to need a parser for every language, and its particular grammar, also the tokenizers and morphological analyzers are complicated, also the Named Entity Recognizers has to be heavily trained for every language and context, etc.

I finished only a simple parto of all this for Spanish and English after years of work! (do not use open source datasets and software like Freeling because the word and training coverage is too bad)

Also cannot use EuroWordnet because of licensing issues!

Have some questions

¿what dictionary/morphoogic-anayzers do you use?

¿Are there real scientific papers (publications) backing up all the bells and whistles?

¿What is your % of success/failure, or some F-score intead on specific tasks? cause there is nothing like perfect NLU, - or give me a break!

¿Does this system correct ortography? (in case of written text)

more later….

Also I am Intrigued en the subjacent parsers characteristics: ¿Do you parse grammatical and ungramamtical sentences?

> What are the best dialog system APIs for the Internet of Things?

api.ai is included on Quora, above.  There is also a dedicated Quora topic at quora.com/api-ai.

On Quora, see also:

> Is there an API that returns Siri like responses in text when questions are asked in text

> What companies are doing cool work in natural language processing

> What is the difference between WIT.AI, API.AI, and AskZiggy

> Is there a free alternative to Cleverbot/Cleverscript

Ilya Gelfenbeyn - May 5, 2015:

we (Speaktoit) have launched our new speech interaction platform

Hey Ilya, what about an update? I’d love to hear more!

Hi Ilya, are there answers for my enquiry? we would be delighted to hear more…

Rob Ellis - May 5, 2015:

Nicely done, and great to see Cordova support!!

So I don’t see anything in the dev guides (quick glance) around NLU. How do you handle semantic reasoning? CCG, FOL, DRT?

Best
Rob Ellis

Hi Rob, there is no semantic reasoning capabilities, as it is not a question answering system, but rather a conversational voice interfaces building platform. Reasoning can be used on the client side after getting recognized intent.

Andres Hohendahl - May 6, 2015:

Hi Ilya, are there answers for my enquiry? we would be delighted to hear more…

Sorry for the delay. The answers are below.

- We support 11 languages. We have 1-3 people in team working on each language. Not using automated translation.
- Yes, handling contexts is one of important parts of the system.
- Dictionaries/morphology analyzers - depends on language. Most of them proprietary, developed in the last 5 years.
- We are not publishing papers, as we are focused heavily on business
- No, not doing spell checking now (most of inputs are by voice)
- F-score and other KPI depend on how narrow the domain is and what does the audience look like. We are constantly monitoring those for Assistant app, but not disclosing
- Both grammatical and not grammatical, yes.

Hi Ilya, thanks for the responses!

My thoughts is that if I buy a product, I wanna know how fast it is (for a car) and the gas consumption, in case of a lingüistic producto, the least I need to know is some measures, because the word-space is huge, I need to know %of coverage, the precision of grammatical analysis, and so on.

I sincerely hope you sell lots of REST requests (app) using this, but I guess that serious users need to know the quality of the engine driving the guts of the system.

Another question, does the morphologic analyzer do spell correction? - in Spanish as long as in most flexion-full languages, its hardly needed (Italian, Portuguese, etc.)

My platform does a regular spell correction, but is just 30 times faster than MS-Word and 5 times more precise! (for the first guess) which is the one needed for a NLU & comprehension system.

I just defended my These in Engineering using this corrector, incorporated into the frontend!

BTW do you have a dialog engine service available?
- just in case you might be interested in mine! and I do disclose specs!

best!

Hi Andres,

We are not selling Assistant as a linguistic product, that’s why stats are not disclosed. As for the platform itself, precision, recall, F-scores, etc vastly depend on inputs and type of agents you work on, so it is hard to give some “average” numbers.

We do have a dialog engine, but would love to learn about yours and understand whether we could work together.

Hi, you are now an ultrafast-response-agent!
Send me a way to contact you, via skype o other, and I send you some papers, brochure, and techincal information, to explain the product we have developed.

Actually, the development is mature, and recently we built a web-editor with a powerful “syntax-highliter” focused onto our DDL = Dialog Definition Language, which helps a lot for understanding the engine itself. it’s fast, written mostly in JavaScript powered by AJAX (JQuery)

We have also a web-interface, to allow a developer team to build a dialog engine, and test+debug it online, If we come to an arrangement, I can let you test the engine and the inteface, its not pretty nor light and delightfull for the masses but enough robust to build and test/debug a dialog script with an easy to understand approach!

BTW: Our approach is so similar to how we think, that it’s like effortless to design a dialog system, and may be this approach is good enough for people who are not skilled programmers, to assemble effortles and efficiently a simple but robust dialog system.

May be with your entities-detection, the system may shine!

Actually we have only developed a full shallow parser-chunker, capable of getting the ‘best’ chunk of many different types, Noun=Phrase, Time-Phrase, Verbal-Phrase, including very complex and nested-ones, the system allows you to commit many typing errors (not too much) and despite them, it delivers a cool analysis too!

give me your mail, or write me a e-letter or just google me, I am available elsewhere, no hiding-games!
cheers!

> Adding Insight To Communications, api.ai Integrates With Twilio To Create Siri-Like Automation (June 2015)

An example in practice: instead of filling out reservations in an online form, consumers can simply text the details of what they want to a Twilio-integrated app. api.ai’s platform will then decipher and clarify user intent, reply via text, and pass the request back to the company. An example of how the integration is working, in this case with reservation management service SeatMe

> Twilio - APIs for Text Messaging, VoIP & Voice in the Cloud

Move over phonegap.com & superscriptjs.com, make room for twilio.com & api.ai....