LINGUIST List 13.1289

Wed May 8 2002

Disc: Falsifiability vs. Usefulness

Editor for this issue: Karen Milligan <karenlinguistlist.org>

Directory

1. Dan Everett, FW: 13.1279, Disc: New: Falsifiability vs. Usefulness
2. Kevin Gregg, Re: 13.1279, Disc: New: Falsifiability vs. Usefulness
3. Dan Everett, RE: 13.1288, Disc: Re: 13.1279:Falsifiability vs. Usefulness
4. dave odden, Re: 13.1279, Disc: New: Falsifiability vs. Usefulness

Geoffrey Sampson offers reasonable, well-known responses to my entirely
unoriginal claims about falsifiability vs. usefulness. I think the views
he offers in reply, however, are unhelpful and problematic. But before I
say why, let me just say why I posted on this bottomless pit of a topic.

Falsifiability is a very outdated perspective in the philosophy of
science. I would guess that the vast majority of epistemologists moved
beyond this years ago. Yet it lingers in linguistics with negative
effects. I agree with Hull (1988: 342) when he claims that "Except for
operationism, no other philosophical doctrine has been so thoroughly
misunderstood or caused more damage in science than falsifiability." It
really irks me to see linguists cite it as a sort of mantra.

But let's take up Sampson's remarks. The first one, given below as (S1),
refers to the well-known Popperian desire to distinguish metaphysics
from science. Consider what Sampson says:

(S1) Your rhetorical questin "What could falsify the statement
'Scientific theories 
should be falsifiable'?" would not be accepted by Popperians as a
reasonable criticism, because Popper's demarcation criterion was very
explicitly put forward as demarcating empirical science from other forms
of discourse; it was never claimed to demarcate sense from nonsense more
generally, and certainly does not. "Scientific theories should be
falsifiable" is not, and is not intended to be, an assertion of
empirical science; nor is, for instance, "It is wrong to kill people".

But it is not surprising that the attempt to maintain this kind of
distinction got Popper into trouble. For example, it forced him to label
evolutionary theory/Darwinism as metaphysics, rather than science at one
point. Later, however, Popper realized that this was a weird move and
reinstated Darwinism as science. But it is very hard to do this and
exclude, say, Marxism (which exclusion was an objective of Popper's - on
this more directly). 

Since Sampson refers to Imre Lakatos's work on 'Scientific Research
Programs', let us consider some Lakatos here. In Motterlini (1999,
90ff), Lakatos observes that it is absolutely crucial that Popper
distinguish metaphysics and science or "Newton and Marx are on a par."
But, Lakatos goes on to say, "the whole enterprise of the demarcation
criteria hinges on whether we can have this dichotomy, i.e., whether we
can have some criteria to decide about the goodies and the baddies."
Lakatos points out that one can draw a distinction between 'shelving'
and 'abandoning' certain problems for a theory. In fact, I think if one
looks at the history of modern linguistics, this is a very common
distinction. Chomsky has made much this same distinction repeatedly,
i.e., that it is often necessary to set problems aside ('to shelve'
them) to give the theory breathing space. But this is not to be taken as
an admission that the theory cannot solve them (i.e. that it must
'abandon' them). In the history of Relational Grammar, for example, many
syntactic problems and potential counterexamples were shelved in order
to focus exclusively on clear cases of grammatical relation-changing
operations as support for the theory. Or, in phonology, I have noticed
that most researchers ignore/shelve the data that in some languages
onsets contribute to stress-placement, because this doesn't fit anyone's
theory. Or, to refer to a piece that Peter Ladefoged and I wrote for
Language some years ago "The problem of phonetic rarities", the basic
response of segmental phonologists to sui generis segments is to ignore
them. However, as Lakatos (ibid) goes on to observe, "Once we accept the
shelving-abandoning equation, Marxism and Freudism are exactly on a par
with Newton, and therefore Popper's enterprise flounders completely."


Now let's take Sampson's second point:

(S2) The point you make about it being usually implausible that a
scientist will treat a pet theory as falsified because of apparently
contrary evidence is well understood and accepted among Popperian
philosophers of science. The man who has discussed this in the most
sophisticated way was Imre Lakatos, who argued that what we ought to
assess are not individual theories taken in isolation, almost all of
which will seem incompatible with some datum here or there, but
"problemshifts" -- sequences of successively modified theories, which
may react to data in ways that progressively enrich their testable
consequences, or may react by increasingly sealing themselves off 
from any logical possibility of refutation.

I agree about Lakatos, as my previous quote suggests. Though I do not
think that Popperians 'get it', not, for example, the problem mentioned
by Lakatos in my above quote. 

On the usefulness of falsifiability, though, let us consider a couple of
additional problems that many who espouse
falsificationism/falsifiability tend to miss. First an easy one from
Hull. Then a deadly problem from Hempel.

The easy one: "In any instance of apparent falsification, too many
alternative sources of error are not only possible but plausible. If one
could be absolutely sure that a particular observation is veridical and
that no modification elsewhere can save a particular hypothesis, then
single-minded attention to falsification would be justified." (Hull
1988: 342)

This is the EASY problem and I already think that it is fatal for any
form of falsificationism. But now consider the hard problem:

"If a sentence S is completely falsifiable whereas N is a sentence which
is not, then their conjunction S.N ... is completely falsifiable..."
(Hempel p74 - in Boyd, Gasper, and Trout, 1991).

This means in effect that I can make any statement falsifiable by merely
conjoining it to a falsifiable statement.

So, for example, "2+2=4 and God exists" is a falsifiable statement.
However, it is hardly what Popper had in mind.

An additional problem with falsifiability, a well-known one, is that it
can only tell us negative things. It allows us to say that this or that
statement is wrong. But not that this or that statement is worth
believing in, worth trying, worth following, etc. I think that this is a
hellacious omission. And I am not alone.

Falsifiability/falsificationism is old hat, unhelpful, and particularly
pernicious in some of the misguided applications made of it in
Linguistics. 

With that positive note I push the send button.

Dan Everett

Message 2: Re: 13.1279, Disc: New: Falsifiability vs. Usefulness

Dan Everett offers 'an old alternative to falsifiability, "usefulness".' I
don't see why the two criteria for hypotheses should be seen as
alternatives. Compare
	1) Football players should be strong.
	2) Football players should be fast.
A coach who restricts himself to just one of those two principles is going
to find himself with a losing team. The alternative to a falsifiable
hypothesis is an unfalsifiable one, and the alternative to a useful
hypothesis is a useless one, and given those alternatives it seems fairly
easy to choose.

It's worth keeping mind, too, that falsifiability does not apply to certain
classes of statements in the first place. One such class is exemplified by
Everett:
>
> As I read someplace recently "Needless to say, whether
>they are right or wrong, our theories should always make strong,
>falsifiable claims." But let's think about this. For example, what
>could falsify this very statement? If there is no way to falsify that
>statement, is it still worth making?

This is a normative statement, about what should be the case, not an
empirical hypothesis on which one can base testable predictions. The fact,
if it were a fact, that every time we make a falsifiable hypothesis we fail
to get uniform confirmation or disconfirmation would not itself be a
falsification of the claim that we *should* make falsifiable claims.

The other case is rather more important: Any empirical science,
linguistics included, makes existence claims, and these are in principle
unfalsifiable: there's no way to demonstrate a negative. This hardly
stops scientists from making existence claims, of course; nor does it mean
that such claims are never rejected.





Kevin R. Gregg
St. Andrew's University
(Momoyama Gakuin University)
1-1 Manabino, Izumi
Osaka, Japan 594-1198

Message 3: RE: 13.1288, Disc: Re: 13.1279:Falsifiability vs. Usefulness

Folks,


There are numerous misunderstandings in the set of replies by Godden,
Lass, and Ludden. I was very pleased to see the professional diversity
they represent, though (engineering, psychology, and linguistics).
Briefly, Godden's mistake is to confuse the way he does engineering and
the way he thinks other scientists operate with falsifiability; Lass
fails to understand that 'usefulness' to James *is* epistemological; and
Ludden thinks 'falsifiability' is a clearer notion than it is.

Let me begin with Godden. He says, "Yes, but all things being equal,
falsifiable claims are far better than non/unfalsibiable claims. The
problem with your examples is that all things are *not* equal."

Let me requote Hull here: "In any instance of apparent falsification,
too many alternative sources of error are not only possible but
plausible. If one could be absolutely sure that a particular observation
is veridical and that no modification elsewhere can save a particular
hypothesis, then single-minded attention to falsification would be
justified." (Hull 1988: 342). This means the following to me. (i) For
falsifiability to be put into operation we need to be sure that the
statement to be falsified is faced with a 'true fact'. This means in
turn that (a) we have a good grasp of what we mean by Truth and its
exemplars and (b) that all those concerned with the falsifiability of a
given statement agree that the counterexample is a fact that meets this
notion of Truth. (ii) The theory or statement is only going to be
acknowledged as falsified if it has no 'room to wiggle'. Practically,
this means, optimistically speaking, that it is going to be well-nigh
impossible to get a statement to sit still long enough to be falsified.
(One thinks of the history of c-command, government, etc.)

Now let me comment on Godden's belief that the following is obvious:
"Statements of theory should also be descriptive and predictive, among
other things." Yes, this would be nice, I agree. But the point is this:
how can you guarantee that the statement scientist 'x' asserts to be
descriptive and predictive will be so understood by scientist 'y'? The
answer is, you cannot. Therefore, Godden's statement here is merely a
restatement of his (and many others') heuristic operating procedure. 

Now let us examine Lass's objections. He says that I have misunderstood:
"I think there is some misunderstanding here. Falsifiability is an
*epistemological* criterion"

And so it is. But so is usefulness. That is the radical innovation of
James and most of the other American Pragmatists, especially John Dewey.
Consider a couple of statements: "The truth of an idea is not a stagnant
property inherent in it. Truth *happens* to an idea. It *becomes* true,
is *made* true by events." [James' emphasis, DLE] (James 1987, 574 -
Library of America Edition, "Pragmatism's Conception of Truth") James
goes on to explain that to be true is to be useful. Nothing more. And
certainly nothing less. Rorty (2000, p4) puts the point this way: "... a
corollary of James' principle that a difference has to make a difference
to practice before it is worth discussing." That is, how is it that what
you claim affects me practically as a linguist? *That* is the
epistemology of Pragmatism. It seems eminently more defendable than
Popperian falsifiability. However, I do not intend to defend Pragmatism
per se here, merely to question falsifiability (but I am working on a
monograph in this regard, to give a brief plug). And I would say that
most linguists, most scientists, most human beings base their actions
and their thoughts on this notion rather than falsifiability. (Not that
I am saying the matter is to be resolved by a vote!)

Lass also says: "But in many fields we don't know enough to make
falsifiable statements"

That is right. But the issue involves more than this. I do not think
that a falsifiable statement is particularly useful per se. Recall
Hempel's statement that *any* statement can be made falsifiable by
conjoining it to a falsifiable statement, to cite one reason for my
reticence.

Now on to Ludden. Ludden's error is to think that we can simply refine
the notion of falsifiability and get it to work. Or that we can have
some sort of statistical notion of it. But what he describes is a useful
way of going about the business of research, not a defense of
falsifiability. This is because what he describes is not falsifiability
but usefulness. 
What he seems to mean by falsifiable is not what Popper, Lakatos, Hull,
Rorty, or Hempel (among a multitude of others) take to be
falsifiability, but clarity and a certain degree of testability. These
are indeed useful notions. But they are not falsifiability.

Ultimately, the weird thing about substituting usefulness for
falsifiability - what makes people recoil from it - is what it does to
the notion of Truth. Truth disappears, except as a consensus on the
useful. To be true is, moreover, (to paraphrase Rorty), "to resist
revision by all future audiences." So linguistics cannot be trying to
get at the Truth. There is no 'ultimate essence' of grammar that we are
trying to discover. There is simply utility. But that is a lot.

Whoops. I let the cat out of the bag.

Dan Everett

Message 4: Re: 13.1279, Disc: New: Falsifiability vs. Usefulness

Dan Everett raises the question whether "usefulness" is as good as or better
than "falsifiability" as a criterion of theory selection. He notes that
"...each claim must be judged on a case-by-case basis according to our
goals". A crucial prerequisite for such evaluation is the identification of
a purpose. Some academic action is conducted for the purpose of
manufacturing a specific profitable product, some is for the purpose of
being employed. My goal is to gain objective knowledge of the nature of
reality, which means to be able to distinguish statements about the universe
which are true, from those which are false. My remarks are attuned to that
particular goal, and the reasoning may be less cogent to someone with a
different goal.

Dan suggests that a more useful statement is better than a more
falsifiable statement "because the statements of a theory are just
tools to perform a task". That is one view of the nature of
theoretical statements. Another view is that the statements of a
theory are representations of reality (which relates to the presumed
goal of a theory, to be a representation of reality). There needn't be
a conflict in these two perspectives, but there easily can be. The
danger of the "just a tool" perspective is encapsulated in the
pragmatics of "just", in that it conveys a sense of unimportance to
theoretical statements. If theoretical statements are "just tools",
what are they just tools to do? By separating theoretical statements
from the lofty but nebulous goal of "understanding the universe", I am
left with no idea what form this knowledge of the universe could
take. I would argue that the form in which "an understanding of
reality" is realised *is* in the form of theoretical statements. Thus,
theoretical statements are not tools, they are the actual goal.

Many of us recognise that the concepts "strong" and "useful" tend to
be used in this business as aesthetic labels, which is quite
regrettable. The concept "degree of falsifiability" (i.e. "strong") is
given substance in The Logic of Scientific Discovery. It is rare than
one actually sees computations of the relative falsifiability of
statements in linguistic theory, so the Popperian ideal remains
largely unrealised in linguistics, but we at least know what it means,
can often identify one theory as being less falsifiable than another
(especially in proper inclusion cases), and can use falsifiability as
a means to select theories. I know what it means for a theory to be
falsifiable (it means that specific empirical statements follow from
the theory, and these statements can be matched with observation to
determine if they are true or false).

In contrast, I don't know what "useful" means: it appears to mean that
someone has found some use for a theory, for some purpose. I can't
compute whether a theory is "useful". The standard for evaluating a
theory in terms of falsifiability is clean yes/no logic, which allows
me to say "This theory is false, because it makes a false
statement". What is the corresponding negative value from the
"usefulness" perspective? The scalar concept of "useful" is applicable
in some degree to all statements: can you imagine an argument being
published that goes "These results show that Smith's theory is
useless"? "Being useful" is not a useful philosophy of science
concept, because it does not partition the universe of theoretical
statements in any useful and objective fashion. On the contrary, it
provides a means of rendering personal subjective opinion more
powerful in deciding scientific debates.