Re: [tied] The minimal-pair test

From: Harald Hammarstrom
Message: 14342
Date: 2002-08-16

Thanks for your answer Piotr,

> > Forgive me for asking but is there a requirement in the minimal pair tests that both pair words be "in the language"?
>
> Essentially, yes, at least in the pure form of the method, because the test
> is then independent of any individual speaker's subjective opinion.

? You have to rely on speakers' subjective opinion on what is an
English word and what isn't anyway. The only difference I can see
is that if something is considered a word, then it has a meaning. If
something isn't, it doesn't have a meaning. Is that the crucial difference
are you referring to?

> In
> your example, some speakers might (hypothetically) object to [gæZ] as
> un-English by pointing out that there is no English word to rhyme with
> it,

So?

It's still a different word and minimal to /gæN/

> and that final [Z] seems to require a long vowel or diphthong (at
> least in RP). Others might disagree, since [æZ] may occur in a medial
> context (as in <casual> and <azure>).
>
> > For example for /N/ and /Z/, any english speaker will tell you that 'gang' /gaN/ is a different word from 'gazh' /gaZ/ even though gazh isn't a word in English. Any english speaker would also say that for instance 'gang' with say an uvular N instead of velar, would simply be a funky pronounciation of gang, regardless of the existence of a word gang with uvular N in English.
>
> In practice, linguists often resort to "near-minimal" pairs, where the words
> differ in more than one segment, but the immediate environment of the
> phones being contrasted is identical (we don't normally expect
> allophonic
> conditioning to operate at a distance, except in some special cases such
> as umlaut). Thus, <hanger> and <azure> should in principle do, since the
> presence or absence of an initial [h] is not known to have any effect on
> medial consonants in English, and the crucial environment is the æ_&(r)#
> part.

And why not make this into theory by avisioning the minimal-pair test
such as suggested or something similar because it conforms better to
the intuition that /N/, /Z/ etc are phonemes in English. Moreover it
is independent of the completely (?) random existence of survived/
formed pair words with meaning out of the set of phonotactically possible
phoneme concatenations in a language. Or do we know (that I don't know)
about this subset (set of words with meaning in a language out of the
possible phoneme collocations) ?

best wishes

Harald