SV: [FUNKNET] Rule-List asymmetry
macw at cmu.edu
Fri Jun 20 20:13:09 UTC 2008
Dear Peter et al.,
I think one gets out of this "commitment-to-a-formalism" quandary
by thinking in terms of "combination" rather than "rules". Let me
As I noted in my earlier posting on this thread (as I argued in
my 1978 monograph on the acquisition of morphophonology) the three
crucial processes in a learning-theoretic account of language learning
and processing are rote, combination, and analogy. Rote is what
people have been calling "lists" in the context of this discussion.
In formulations from Derwing, Bybee, PDP, and even perhaps de
Saussure, you can hook up rote to analogy and get productivity and
generativity in this way. In that system, rules are completely and
thoroughly embedded in the between-item similarity matrix. The other
option hooks up rote with combination. In that framework, rules
emerge during learning as methods of adjusting combinations for
semantic, phonological, prosodic, or positional patterns.
In both of these accounts, rules are not really on center stage,
so the issue of how to "talk about rules" is no longer the key issue.
We can view linguistic processing as an ongoing horse race between
rote and combination where the winner is rote whenever the relevant
rote form has high frequency and easy retrieval. The processes that
"patch up" combinations are called analogy in one tradition and rules
in another, but as long as we are talking about learned modification
processes based on evidence in the inventory of stored rote forms,
then there is no empirical difference between rules and analogy. I am
not saying that pattern or rule learning is trivial or uninteresting.
Far from it. However, I am saying that the core contrast is between
rote and combination and that rule or pattern learning is triggered by
the fact that we have decided to build up utterances at least in part
The generative view is different from this three-process account.
Unlike the 3-process account, minimalism passes everything through the
MERGE operation. So everything is combination. Moreover, rules are
much more on center stage during the process of MERGE because they are
often innate and not learned. And, if learned, they are acquired
through an innate system of parameterized options. None of this,
except for the core operation of MERGE, is involved in the learning-
theoretic 3-process account.
-- Brian MacWhinney
On Jun 20, 2008, at 10:36 AM, Peter Harder wrote:
> There is a sense in which it is fairly obvious that rules exist.
> This is already implicit in Martin's reference to systems that
> consist only of lists, such as traffic signs - which invites the
> inference that human languages are not of that kind.
> If you do not subscribe to a generative view of linguistic rules,
> you have the problem that there is no agreed precise alternative for
> what a rule then is. However, consider a messy informal 'direction-
> for-use' rule such as "after verbs like say or think you can put a
> that-clauses, indicating the propositional content of what was said
> or thought". This can only be an empirically misguided description
> of English if speakers never combine verbs and complement clauses
> freely, and invariably limit themselves to repeating whole
> statements that they have heard previously and added to their lists
> of possible utterances, like lists of traffic signs in driver's
> manuals. Not all speakers do that, so it follows that English has
> rules of that kind.
> This does not answer the question of the psychological reality of
> rules, or of the less exalted semi-general combination strategies
> that functionalists might be satisfied with instead of fully general
> rules. But if we pursue the same abstract agenda as Martin, we can
> say that this is an extra issue that is external to the semiotic
> system as a set of social options available to speakers of English.
> You cannot make full use of the options offered by the English
> language, if your cognitive system works only by listing. Therefore
> listing as a theory of how English works (as a public code) does not
> cover all there is to say about it. This is, of course, a very
> modest level of abstraction to be working at, but the way I see it,
> it gets us just a little beyond total agnosticism with respect to
> the existence of anything beyond lists.
> Peter Harder
> Fra: funknet-bounces at mailman.rice.edu på vegne af dryer at buffalo.edu
> Sendt: fr 20-06-2008 13:38
> Til: Martin Haspelmath; funknet at mailman.rice.edu
> Emne: Re: [FUNKNET] Rule-List asymmetry
> I realize that I have probably misrepresented Bruce Derwing's
> While it is true that he argued that the burden of proof on those
> who claim
> that there are rules in particular cases is to provide empirical
> that there are rules rather than lists, he did so only because the
> prevalent practice was (and is) to assume rules without any evidence
> such, not because he considered lists the default. If there had been
> people claiming in particular instances that people only stored
> lists and
> saw no reason to provide any empirical evidence for that claim, I
> Derwing would have raised the same objection, in effect "You have to
> provide empirical evidence; you cannot simply claim that on a priori
> In other words, in the absence of empirical evidence in a particular
> instance whether there is a list or a rule, rather than saying "In the
> absence of empirical evidence, we should assume there is just a
> list" we
> should be saying instead "In the absence of empirical evidence, we
> shouldn't assume anything; we should look for empirical evidence."
> So while Martin may be right that at some abstract level, lists are
> logically more basic than rules, this is ultimately beside the
> point. It
> does not justify assuming in any particular instance in the absence of
> empirical evidence that there is a list rather than a rule. To do
> so runs
> the risk of committing the error that Derwing argued against, namely
> of deciding an empirical issue on a priori grounds.
> And that is why I think Tom was right when he said that in arguing
> was the default, we were gravitating towards an empirical vacuum.
> --On Thursday, June 19, 2008 9:10 AM +0200 Martin Haspelmath
> <haspelmath at eva.mpg.de> wrote:
>> It seems that some people still have not seen my original point,
>> surprises me.
>> Thus, Matthew Dryer wrote:
>>> I think Tom's quite right about the issue of what is the default
>>> something of a red herring.
>> And Aya Katz wrote:
>>> But in scientific enquiry, why do we need a burden of proof?
>>> Wouldn't it
>>> be better to avoid bias altogether?
>> My claim was and is that in the study of semiotic systems such as
>> language, the burden of proof is on those who want to claim that
>> exist. Those who claim that lists exist do not have this burden of
>> because lists are the default, and there is an inherent bias in
>> favour of
>> Rules and lists simply do not have the same status. Semiotic
>> systems with
>> lists and no rules are perfectly possible (and widely attested, e.g.
>> traffic signs), but semiotic systems with rules and no lists are
>> logically impossible. Rules have to range over a certain domain,
>> and the
>> domain must be defined in terms of lists. There is no symmetry here.
>> Sorry, Tom, this is not an empirical argument, it's a logical
>> Perhaps it's not particularly interesting, because everyone of course
>> agrees that languages have lists. I just made this argument because
>> Newmeyer had claimed that Sandy Thompson was guilty of the rule-list
>> fallacy: Assuming that there are no rules just because everything
>> can be
>> explained by lists. But this is not true: If speaker behaviour can be
>> fully explained by assuming only lists, it is indeed reasonable to
>> that no rules exist. (Conversely, it is never possible to claim that
>> speaker behaviour is fully explained by rules and no lists are
>> Some lists are always needed.)
>> I think this point is important because as Matthew Dryer pointed out
>> (citing Bruce Derwing), linguists have generally been very eager to
>> rules, which is OK, because that is their job. But they have often
>> often) jumped to the conclusion that speakers also have these rules,
>> without any evidence. It has been widely assumed that speakers are as
>> eager to extract generalizations from the data as linguists, but
>> this is
>> not reasonable. Speakers just want to talk. If they can get away with
>> just lists, they might well do without rules. Any claim about
>> that goes beyond lists needs additional evidence.
>> The reason why linguists generally agree that there are both lists
>> rules is that this evidence is often easy to come by: Speakers
>> abundant productivity in different areas of language structure:
>> especially in syntax and phonology, but often also in morphology.
>> As soon
>> as we observe productivity, we have knock-down evidence for rules.
>> only question is what we say in the absence of productivity, and of
>> course how exactly we diagnose productivity; if we rule out
>> evidence and try to rely exclusively on natural discourse data, as
>> Thompson seems to be doing, it's not so easy to find evidence for
>> productivity. In the absence of productivity, providing a proof for
>> is a very heavy burden indeed.
>> Martin Haspelmath (haspelmath at eva.mpg.de)
>> Max-Planck-Institut fuer evolutionaere Anthropologie, Deutscher
>> Platz 6
>> D-04103 Leipzig Tel. (MPI) +49-341-3550 307, (priv.)
>> +49-341-980 1616
>> Glottopedia - the free encyclopedia of linguistics
>> (http://www.glottopedia.org <http://www.glottopedia.org/> )
More information about the Funknet