Just to take a few points from diamondage and a response to Guilianna:
The truth is that children's spoken vocabulary far, far exceeds their reading vocabulary until well into secondary school so it's not so common for a child to come across a word, decode it and for it not to trigger recognition.
However, it does happen, as it does for fully literate adults. So, what do you do if you happen across a word in, say, Scientific American, and you don't know how to pronounce it? You make an educated guess and then, if it's important to you, you look it up in a good dictionary, or online, or you ask someone how to say it.
With the word 'labyrinth', whether you say it as I do as lab uh rinth, or as you do as lab rinth doesn't matter. If it's already in your spoken vocabulary, you'll recognise it. On the other hand, if you want your pupil/child to be able to spell it, you split it into its syllables /l/ /a/ | /b/ /uh/ | /r/ /i/ /n/ /th/. The only difficulty now is the schwa or weak vowel sound frequently encountered in weak syllables of polysyllabic words like this. You tell the child that we spell the /uh/ like this . The you get them to write it sound by sound across the three syllables and read it back and ask them, if they had to spell the word tomorrow, what would be the difficult bit? And get them to focus on it. Actually, this spelling is really a bit of an anachronism. You'll see in Tyndale's version of the Bible that is a frequently spellyng for what has become in many modern spellings an . The other thing is that, because language is dynamic and pronunciation changes over time and with fashion, spelling can be a bit conservative and lag behind.
But, of course, you are quite right!: 'It just isn't possible to know from text alone'. That's because language is phonologically based: sounds (spoken language) come first and written language is a relatively recent invention. Furthermore, on account of accent differences, we will probably say all sorts of words slightly differently. That doesn't mean though that every single sound in every single word hasn't at some point been assigned a spelling.
Going back to the original question from Guilianna 'What would you say to a SLT convinced that 'sight reading' is as effective a strategy as phonics, and who advocates teaching mixed methods?' I'd say to the SLT that there is no scientific evidence to support their hypothesis. The scientific community is resolved that all the data converges on a consensus that teaching phonics results in significant gains in word recognition and spelling. and that teaching whole language is a 'very efficient way to generate large social class differences in reading achievement', to quote Keith Stanovich (from his essay 'Putting Children First by Putting Science First' in Progress in Understanding Reading: Scientific Foundations and New Frontiers.