Localizing Speech Applications - Part 2 Video



  • In the second part of our series on localizing speech applications, we look at how moving speech applications affects grammars and how using custom pronunciations can be added to grammars to help with this process. We also cover advanced features of languages, such as considering formality and politeness, and how dialects within a language can affect recognition.
  • RUNTIME 8:33


Video Transcription

Localizing Speech Applications - Part 2

Welcome to the second part of Localizing Speech Applications. We're going to pick up where we left off in Part 1 and talk about localizing grammars, some of the key differences between dialects and languages and what the distinction is, and we'll also talk about the importance of tuning a speech application.

Localizing Grammars

To reiterate, grammars are a list of words and phrases that determine what kinds of things a user can say at a given time. The key thing for grammars is: don't just use literal translations of existing grammars. You can't just take an English-to-Spanish dictionary and take my English grammar and look up each word for it in Spanish. People use words in completely different ways when you go from one language to another, so you want to make sure that you're working with a native speaker, and not just some guy who took a couple semesters of Spanish in high school. People are going to use words in different ways. They're going to come up with different combinations of words than what you'll expect, and word order changes between certain languages. In English, you have an adjective and then a noun, and in the Latin-based languages, you have a noun and then an adjective. Your grammar needs to reflect this kind of change.

You'll also want to do this after your prompts have changed. Remember, the prompts will always influence what the caller is going to say, so then the grammars need to reflect this new influence. And like always, consider how the users will respond. Keep that key thing in mind, "How will a user interact with our system?" That's what you'll always want to prepare for in creating your grammars.

Custom Pronunciations

One thing you can put in your grammars is custom pronunciations. There are times when, for whatever reason, our dictionary doesn't contain a word. It might be because the word isn't in the dictionary, or it might be a word from outside a given language, so what we allow you to do is tell us precisely how a word is pronounced. It's useful for when we don't have a local pronunciation for a word. For example, we have a dialect, and a dialect is just a different way of speaking within a language. There are different dialects for pretty much every language out there. Often times place names are very heavily influenced by a local dialect. A person who lives in a place may pronounce it differently than every other place. An example is, in America, most of the country says "New Or-leans", and people who live there say something closer to "New Aw-lins". You might want to add those kinds of custom pronunciations in. We see this all the time in City/State directory grammars. Also, for words not common in a language, like foreign names, it's a good idea to spell them out phonetically. Elsewhere on the site, we have a whole section on adding in custom pronunciations, foreign words, using phonetic spelling, so you can look up those if you want more information.

Formality and Politeness

There are two components to consider when moving between languages: formality and politeness. Formality is the more strict, linguistic-type idea. In English it's hard to demonstrate, but in, say, a Latin language, you conjugate verbs differently when you're being formal. In French you have "vous" or "tu" to mean "you". "Vous" is formal, and then your verbs take on different endings depending on being formal. So make sure that if you're moving from English, which has basically no concept of formality, that your new language accommodates that. If you have a Spanish grammar, you probably want to accommodate both the formal and the informal verb conjugations.

There's also the idea of politeness. This can often differ by a region within a language. Politeness is saying things like "please" and "thank you". We see this a lot in English. We had a customer who was making a banking application, and it was working fine, except they kept finding that in certain transactions they would get issues. Most of the time when people withdraw or transfer money, they do it in lump sums, like $100 or $150. But they found that they were getting these interactions where customers were recognized as saying "$100.23". It turns out what happened was their grammar would allow you to say "$100.23", but what users were actually saying was "$100, ma'am", because they were responding to a female prompt. It turns out that users in the southern part of America tend to be formal, particularly when speaking to a female voice. As soon as you add the possibility for users to end a sentence with "ma'am", the problem went away. So, always understand how users speak in a region and you'll be much better off.

The Importance of Tuning

All of this comes back to tuning. Tuning an application is when you make changes to it after it's already been deployed, after you collect data that shows how users are actually interacting with it. That's when you can determine if your callers are actually saying "please" after the sentence, to add that into your grammar. This is always an important part of speech application development, but it's especially important when localizing. There's a temptation to think we'll just translate it and we're done, but no, you have to tune it again because people are going to be using it differently. We have a Speech Tuner application, it's free to use. If you're using the LumenVox Speech Engine you can tune your applications like that. We have all sorts of videos and resources on how to tune. Please be sure that once you've gone through these that ultimately the users are in control. They will tell you how they'll use it by actually using it. Listen to them use it, find out where you've gone wrong, make those changes, and you'll have a much more successful application.

If you have any more questions on localizing, feel free to contact LumenVox Support. We'll be happy to talk to you about it. You can email us. Also be sure to check out the other resources we have here at LumenVox.com. We have all sorts of white papers, tips, articles, and videos on developing and localizing speech applications.

© 2018 LumenVox, LLC. All rights reserved.