Translate did not suck in 2006, it was a big leap over the competitors at the time. It was really the first time a company had launched productized statistical machine translation at all. Before that there was the Altavista Babelfish which was rules based.

Going neural definitely made a big difference but Translate was called "statml" by the infrastructure internally for years, maybe still is, because just using statistics and having training at all was a big deal back then.