In Chinese, words can have different meanings if their tones are changed, e.g. 是 (shì) and 十 (shí). In Italian, words can have different meanings if a consonant is geminated, e.g. sete and sette.
My question is: is there any language whose words can have different meanings based only on the loudness of the utterance? If so, please provide examples.
Intensity is the physical correlate of loudness, and is also a correlate of stress in some languages. Moreover, stress can create differences in meaning in some languages (e.g. PRO-test vs. pro-TEST); we say that such languages have lexical stress, or that stress is lexically distinctive. It follows that if we can find a language where intensity is a good correlate of stress, and stress is lexically distinctive, that would answer your question.
One example is Papiamentu (Remijsen and van Heuven, 2005), and I'm sure there are others.
Note: I'm responding to the question in your title. The question in your main text adds the qualifier only, which makes such cases a bit harder to find, because intensity usually isn't the only cue for stress. Although I doubt pitch is the only cue for tone in your Mandarin example either, since tone also affects duration...
Remijsen, Bert & Vincent J. Van Heuven. 2005. Stress, tone and discourse prominence in the Curaçao dialect of Papiamentu. Phonology 22(2). 205–235.
@ me when replying.
In Russian the word stress is marked purely by the loudness/force of the syllable (contrary to other Slavic languages where the stress may be also marked by the length or the pitch.) This stress may alter meaning of words, with the most notorious example being verbs "писать" and "писать" meaning respectively "to write" and "to pee". This is a subject of many jokes, such as "Tchaikovsky wrote a lot, and wrote the 'Swan lake'." ;)
A contested and debated case with different opinions is the case of final obstruent devoicing in many languages; it has been claimed by a lot of research, and contested by others, that in spite of these languages devoicing voiced obstruents in the coda of syllables that the loudness difference between the original fortis and lenis pairs remains, and that this difference can in fact be used by native speakers to differentiate between minimal pairs with higher than chance accuracy.
It's also contested, in this case, just how much of it is purely the loudness and whether it's also the length or the minimal length difference of the vowel that comes before it that cues native speakers. Historically, the German pairs of "Rat" and "Rad" were thought to be pronounced identically, but careful accoustic analysis often finds that the /t/ in the former is realized louder than the /d/ in the latter, despite both being devoiced.
