I think that is just a technology showcase that this technique is so good that for some small-mid size pronunciation dictionary, it can work quite well without

I think that is just a technology showcase that this technique is so good that for some small-mid size pronunciation dictionary, it can work quite well without using a language model. Similar works were reported recently (ASR systems without language models or with character models). But seems like real system is still using a graph generated with a language model and context dependent phones.