Demo: Developing Multi-lingual Conversational AI Applications

Watch Again

1. What restrictions do child flows have? Can the child's Flow's structure be edited or must it follow the Master?

The child flows must follow the structure of the master but are localized into the child solution’s language (training data provided in the target language, answers translated into the target language, etc). They might use different integrations (for instance, each region may have their own back-end systems such as ordering, CRM, databases, etc) and these variations can be easily parameterized to minimize the work needed for the children.

 

It is possible for a child to break its connection to the master if there is a requirement for a different process flow, but that decision is irreversible and the child will no longer be notified of or inherit changes in the master flow (for that flow only, not for other elements, where inheritance continues as before).

2. How does the robot accesses customer data?

Teneo access data through integrations with the appropriate backend systems generally using Java or Groovy.

 

Within Teneo Studio is a dedicated Integration Manager, separating the code components from the conversational components and allowing your programmers to create the appropriate integrations for use by non-technical conversational AI designers building the solution. You can create libraries of these integrations for reuse across multiple solutions and Artificial Solutions provide pre-built integrations for many common systems.

3. Do you have a library in pt br that can be downloaded to use on Teneo by default?

Yes, Artificial Solutions provide pre-built libraries for all the languages that we support including Brazilian Portuguese. Further note that these libraries can be extended by you to incorporate additional vocabulary you might have such as company-specific verbiage, product names, etc.

4. How well can Teneo do mixed language?

Supporting mixed language in the context of a chatbot is quite difficult due to the subtleties of language as well as the fact that one word may have different meanings in different languages. For example, in English a gift is a present and something desirable where as in German, Gift is poison. For that reason, chatbots are typically deployed to support a single language each. Teneo can detect when a user is using a language different than that of the solution and react accordingly.

 

As shown in the demo, this might include seamless transfer to a different chatbot which speaks that language, providing an alternative contact channel such as call center or live chat, or notifying the customer that their language isn’t currently supported and providing a list of those which are. Also of interest, this is all logged and can be reviewed to make future business decisions so that, for example, if 15% of users begin speaking an unsupported language, that might be targeted as the next chatbot solution language.

5. Isn't there a lot of work to create all those conversations/chains?

No, the goal of Teneo is to remove the complexity and effort from building conversational solutions. In a typical project, the first deployment of the bot can occur within 8 – 12 weeks of project start, allowing for rapid time-to-market. As an example, the “my bill is too high” flow seen in the demo too, only a couple hours to create (though there was refinement after that initial build). Further, as Teneo was created with the enterprise in mind, you do not need to rely on Artificial Solutions for development but can use your own non-technical subject matter experts to take complete control of development if you wish (though we do have a dedicated professional services team who can assist you ranging from providing occasional advice to completely building the solution for you). Artificial Solutions employ an iterative, agile methodology for solution creation, allowing you to quickly create and deploy usable bits of a solution without the need to fully deploy your entire project scope.

 

The particular focus of this webinar was deploying the same solution in multiple languages and the amount of effort saved with the master local functionality is tremendous, allowing you to build out your flow structures a single time and automatically recreate them (and update, optimize and add over time) in child solutions.

6. Do you support Arabic and their dialects?

Teneo does not currently support Arabic but it is a language on our roadmap and we are actively looking at providing support.

7. What languages does the solution support? Specifically Spanish, Portuguese, Mandarin, Cantonese?

Teneo currently supports 36 languages: Basque, Belarusian, Bosnian, Bulgarian, Catalan, Chinese (Mandarin), Croatian, Czech, Danish, Dutch, English, Esperanto, Finnish, French, Galician, German, Greek, Hungarian, Icelandic, Italian, Japanese, Latvian, Lithuanian, Macedonian, Norwegian, Polish, Portuguese, Romanian, Russian, Serbian, Slovak, Slovene, Spanish, Swedish, Turkish, Ukrainian. Learn More.

8. If we have one mega agent which is in English language and suppose we create multiple sub agents in different languages, does it automatically convert all the data in particular agent language for those sub agents?

As noted in the presentation, while it is possible to build a solution in one language (which needn’t be English) and integrate with machine translation to support additional languages, this will provide a sub-optimal, potentially even unusable, experience for users not using the language of the solution. We strongly recommend that the translation for child solutions be carried out by human translators who are able to understand the nuances of the language and accurately present the voice of your company in each of the target languages. That said, for more simple FAQ type solutions, machine translation might be a viable option.

9. Does multilingual solution imply any possible language or just the more developed and popular ones?

Teneo currently supports 36 languages with our unique hybrid model mixing a machine-learned classifier with our own linguistic conditions based on Teneo Language Resources (TLR), allowing for high accuracy intent recognition. Depending on the language, it would still be possible to use the classifier alone with currently unsupported languages and develop the TLR later. We can rapidly create TLRs to provide support for new languages in between 2 – 6 months, depending on the similarity to an already supported language.

Conversational AI for Contact Center Agility

Watch Again

1. Will these systems be integrated with homepod systems like Google Pods or Alexa?

Yes – Teneo can be integrated with any system that exposes a suitable API. Teneo Developers lists a variety of off the shelf connectors https://www.teneo.ai/engine/channels. Google Assistant and Amazon Alexa are covered.

2. How about accents - say a Spanish accent - when talking in English?

The quality of transcription for various accents is primarily dependent on the ASR provider you plan to use. Some providers in the space provide features that cater specifically for certain accents and some also provide the ability to train the transcription model on a set of possible and probable utterances. Artificial Solutions doesn’t offer ASR or TTS capabilities as part of the platform. We have however worked with most providers in the space over the years.

3. I'm from Europe and was wondering how the tool works with other languages (German, French, Dutch). Does each language needs it's own "learning"?

Teneo is capable of understanding over 35 languages. A conversational solution is configured with a language specific set of resources. These resources contain rules and language building blocks for the respective language. You are still required to provide language specific training utterances for each of the intents you plan on handling.

4. Did you work with Teneo ML only or do you also integrate 3rd party AI components?

Teneo provides a good coverage of all the ML components you would want in a Conversational AI platform. That said, Teneo is extensible and you can reach out and utilize any external APIs or ML models. This can either be done through custom integrations in Studio or you can expand the logic in one of the Teneo lifecycle hooks.

5. What languages is the speech recognition available in?

The demonstration in the Webinar utilizes the freely available Web Speech API available in supported browsers. The list of languages that the Web Speech API is extensive. The languages we typically demonstrate through the chat client are English, French, German, Italian, Japanese, Danish, Dutch, Norwegian, Spanish, Swedish and Russian. Coverage of speech recognition is really just a capability of the ASR provider you choose to use for your specific use case. Artificial Solutions has work with many over the years.

6. Can the interaction get transferred from Virtual Assistant on web to IVR channel or live voice agent?

Teneo can assist with this type of transfer where the capability exists in your communications provider’s stack.

7. Is your tone of voice (Text to Speech) personalized as well?

Text to Speech is configurable based on your preferences. In the demonstration we saw a couple different TTS providers playing a role. Through the IVR channel Amazon Polly was used and a specific voice was configured. It could just as well been a different provider and different voice. The voice used in the web chat client are also configurable.

8. Is your tone of voice (Text to Speech) personalized as well?

Text to Speech is configurable based on your preferences. In the demonstration we saw a couple different TTS providers playing a role. Through the IVR channel Amazon Polly was used and a specific voice was configured. It could just as well been a different provider and different voice. The voice used in the web chat client are also configurable.

9. Which SMS platform did you use that can present buttons like 'yes', 'no' etc.?

The button options you saw in the SMS demonstration are generated automatically from the actual SMS app on my Samsung Galaxy Note phone. We do send signals along with Teneo responses to tell channels that support it to show buttons and other rich media. You’re correct though that SMS just doesn’t provide that capability by default – what you saw is a bit of Samsung/Android magic.

10. Do you also take email text data?

Yes, email data can be useful in understanding the regularity of certain topics or associations in the data. This insight can be highlighted via the use of Teneo Discovery and used to scope projects and perform a detailed understanding of the sub topics and types of conversation you will likely need to cater for in a Conversational AI solution.

11. Is there a specific difference between a telephone access system and one that uses chat? Are you able to use the same Conversational AI for both?

The same conversational solution is used across channels. This ability to be aware of the channel initiating the request allows Teneo personalize responses per the capabilities of the channel. Teneo Developers provides a listing a ready to use channel connectors, all of which could be connected to the same Conversational AI solution: https://www.teneo.ai/engine/channels

12. How does your solution support compliant treatment of personal/medical information when used in the Healthcare vertical?

The conversational data always belongs to our clients and can be hosted within your own data centers or private cloud. Teneo is GDPR compliant and as such provides tooling in the platform to delete, redact or pseudo anonymize certain conversational data.

13. Do you use recorded voice or Text to Speech for your CAI solution? What percentage of one vs the other?

95% of time, it’s mainly the use of Text to Speech that is used.