Speech recognition is really hard for computers to do even under ideal situations. Accounting for people with multiple, distinct accents speaking multiple, distinct languages is no easy task. To understand why a computer might have difficulties, look no further than this comment thread and you'll see that even humans have a hard time!
You're probably misinterpreting the peek.
If you have the sentence "Tá sé" and you hover over "tá", you'll get a hover with "tá sé", then under both of those words "he is", then under the left word only "is", "am", "are". Then if you hover over "sé", you'll get a hover with "tá sé" again, with "he is" under both words, then under the right word only, "he", "it", "six".
So it's telling you that "tá sé" is "he is", and that this phrase is made up of "tá" = "is" (in this case) and "sé" = "he" (in this case).
It's not the case that tá by itself means "he is" and sé by itself also means "he is".
I think it's fairly useful that Duolingo's peek function glosses not only individual words when you peek at them but also combines words from a sentence and tells you how those words, taken together, may have a translation that is not immediately obvious.
(For example, "Tá úll agat" - "Is - apple - at you" may not be enough to enable you to translate it as "you have an apple", but if hovering over the "tá" also tells you that the combination "tá ... agat" can be translated as "you have", that can help.)