Hi there,
although our bot is far away from even intending to master (audio) speech output, we thought it would be a nice idea to teach it phonetics. Ok - I admit - the idea was not to teach the bot, but to provide a nice means of generating IPA output for our dictionaries, but as the bot can make use of this functionality - why not exploit it?
Now why I’m writing this: we made the discourse engine “aware” of the phonetic markup and there are unexpected side effects…
As you can see in the subject, it can convert normal english text to IPA. It’s not perfect - of course - because there are many exceptions it hasn’t learned yet, but it starts to get usable. The side effects come from the fact, that the bot is inherently multilingual. It can transcode e.g. german or czech too:
Velmi zajímavá úloha = [vɛlmɪ zajɪːmavaː uːloɦa]
Schönes Wetter heute = [ʃønes veter hɔʏte]
But the fun starts, when you ask to do the transcoding “AS IF you were interpreting the text in another language than it is written in”.
How would a person who knows only english phonetics probably read the german word “Dach” [dax]? The bot knows the answer: [dɑːtʃ]
Which - if spoken out - for the german listener is like “Dahtsch” and therefore uncomprehensible.
Imagine having a reverse transcoding function which would take [dɑːtʃ] and do the eng->deu transcoding and receiving [dax]. Perfectly possible and we might implement that also in the near future. Then you’d finally know what the anglais meant when he said that funny word.
Of course, the spelling problems are not unidirectional. Imagine the poor 40+y man from the czech rep. who never had english in school. He comes to a CD shop in NY and would like to buy a CD from his favourite band [θɛ kjʊr], but all he manages to say is: [tɦɛ tsurɛ]
Lucky the CD dealer who can perform language context sensitive IPA transcriptions back and forth in his head.
Richard