Rules vs. Lists

[Rob Freeman]

Maybe you are right Jouni. Maybe the distinction example-class does not matter.

Perhaps a better way to make my point is to say we need to focus on
ways of breaking things down, rather than ways of putting things
together.

We can take it to another level and accept that "utterances" too need
to be thought of as classes if you like. Accept we need to segment
them from some global speech act. Or for convenience we can assume a
level of phonemic or lexical abstraction and think only about how
rules can be abstracted from sequences of those. What is important for
my point is that we think about ways combinations of elements can be
abstracted from wholes at any level.

I wasn't sure if what you wanted to do was define X phonemes and then
argue there can be >> X rules governing their combinations. Trivially
that is so. The problem is we can't find any actual set of rules which
completely explain all the combinations of phonemes we get. I want to
turn that around. The interesting question for me is given Y
combinations of phonemes: abcd.., aabc.., acdb.... how many
generalizations/rules can we find between those combinations? More
than Y?

-Rob

On Sun, Jul 6, 2008 at 9:04 PM, Jouni Maho <vch468d at tninet.se> wrote:
> Rob Freeman wrote:
>>
>> We have to be careful what we regard as "examples". As I
>> said to Jouni phonemes should be thought of as classes not
>> examples. Similarly "roots", "lexemes", "morphemes" etc.
>
> Well, you have to convince me why example-class is an important distinction
> to make here.
>
> I'm sorry if I seem to be running off on a tangent, but I understood the
> more-rules-less-examples thing as being about lists of items and the rules
> that apply to them, but perhaps you're actually talking about something
> else.
>
> Still, let me try to retract a bit, just to try to clarify to (for?) myself.
>
> When a language user extracts rules (generalisations) from a series of
> utterances, that assumes that the rule-extractor has analysed the utterances
> into an abstract list, so that each uttered "Hi!" is analysed as belonging
> to a set.
>
> Each generalisation (phones to a phoneme, many uttered "Hi!" to one abtract
> 'Hi!') is a rule, of course, but the abstract entities /a/ and "Hi!"
> themselves become units of a list on which other rules can apply. Hence also
> the rules themselves become members of lists. (Perhaps my earlier
> hypothetical example was not 5 items plus 6 rules, but rather 11 items
> including 6 rules.)
>
> Anyway, is "example" equal to the member of an abstract list ("Hi!" counts
> as one) or each uttered word ("Hi!" counts as many)? As a language user I
> make generalisations on various levels of abstraction. I can establish
> lexemes and phonemes from utterances, but I can also generalise syntactic ad
> morphological rules that apply to only certain classes of words or phonemes
> (which requires that I have made the example>class analysis first). So, does
> the distinction example-class really matter here?
>
> ---
> jouni maho