I’ll answer this question for English, rather than cross-linguistically; I’ve A2A’d users who are more across the right typological databases.
Markedness (the linguistic notion of what is the default value between two alternatives) is a confluence of several factors, and in all of them, voiceless wins.
- In frequency within the lexicon (frequency of types), θ is by far more frequent. ð is very frequent in tokens, because of its prevalence at the start of very common function words; but if you pick a random word of English with a <th>, it will almost always be voiceless.
- If you look at the synchronic rules for how <th> is pronounced, in both the Stack Exchange and Wikipedia links, the “else” rule is the voiceless. That makes the voiceless the default value in speakers’ internalised rule system.
- For what it’s worth, θ diachronically was also the unmarked value: ð was restricted to occurring between vowels.
- This means that in peoples’ intuitions of English, θ is the unmarked reading of <th>. If they are confronted with a new random word with <th> in it, θ is how they will pronounce it by default.
- In collaboration of that, look at how Modern Greek δ is transliterated into English. You will occasionally see the spelling dolmathes for ντολμάδες, but you almost always see the spelling dolmades instead. And there is a straightforward reason for that: because ð is so marked in English, no one would assume it is the pronunciation of a novel loanword with a <th> in it.
, Linguistics PhD candidate at Edinburgh. Has lived in USA, Sweden, Italy, UK.