![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
On Explaining Language Change
Mar. 29th, 2014 03:21 pmD-N: Deductive-nomological
"I do not think there can be deductive answers to 'why' questions in linguistics."
cultures are norm-governed and rule-governed, not law-governed
"I suggest that it is wrong to say that 'Jones was injected with penicillin' is ever a satisfactory answer to 'Why did Jones get a rash?'"
"The fact that 'explanations' like [Jim was exposed to the measles and there was a high probability of his getting the measles, which is why he got the measles] give us only inductive likelihood, not deductive certainty, makes them not only 'weaker' than D-N explanations, but makes them non-explanatory"
While any explanation entails prediction, prediction does not entail explanation.
probabilistic
"the 'probabilistic' explanation, which is often taken to be useful in the social sciences, biology, and other slightly woolly domains, is not in fact 'explanatory' in any useful sense.'
If you posit "ease of articulation", any counterexamples are extremely problematic. On the one hand, singular counterexamples do not falsify probabilistic theories. This is an epistemological difficulty that threatens to reduce all 'tendency' explanations to vacuity (Lass believes it succeeds).
It is a necessary property of inductive probabilistic explanations that they are strictly speaking non-empirical and don't explain anything. They don't predict for individual members.
A probabilistic explanation has no empirical content, because no individual failure of the "law" can ever be a counter instance. cf. Popper 1968a 146
Many people still insist that probabilistic explanations are both useful and genuinely explanatory.
Markedness: 85% of languages have the unmarked form, 15% have the marked form. How can it be a property of something that N languages have it? How can it be a property of languages that only N of them have it?
"notions like markedness and naturalness have no explanatory significance, but are merely taxonomic cover-terms for observed distributions, or instances of the 'naming fallacy' (virtus dormativa?)"
stochastic laws could exist in languages, just not in the pan-chronic set of all languages, and they are not of the 'natural' or 'markedness' type.
Lass says that if the 15% remained constant in different language families, i.e. some language families lost them and others gained them, that would support a stochastic law (though it would still not be an explanation, just an observation of a distribution).
We should continue to attempt to formulate D-N explanations, because it's a prereq to forming a serious explanation, and gives us the ability to distinguish the deterministic components from the non-deterministic components.
It is not possible to establish any sufficient or even necessary conditions for homorganic nasal assimilation.
What is the status of a "problem" if some languages appear to consider it not to be one? They adopt the "null strategy" approach.
"According to Lightfoot only the fact of change has to be predicted...not the specific change. I find this unsatisfactory."
"Even the naturalness of assimilation becomes either boringly analytic or mysterious."
The acoustic effect could be misinterpreted. But there are more common ways to misinterpret it, so why this one?
The multiple strategy problem is not of course that anything goes, but they operate within pretty wide latitudes, i.e. they are rather loosely determined by physiological or perceptual boundary conditions.
Spectographic similarity or acoustic similarity (to a phonetician) does not necessarily equal perceptual similarity.
Confusion of why with how: specifying that a segment arose because of properties of the airstream, vocal tract, etc. still doesn't invoke genuinely causal factors that might have led to these properties being capitalized on in particular cases in particular ways.
No phonetic explanation which is not nomically necessary is an explanation.
Reconstructions are not testable [(but cf. laryngeals). Also, extrapolations from data sets should be testable. Lass acknowledges that it gives us hope that our techniques are reliable.]
Reconstruction must be justified by methodological criteria (purely conventionalist, like Occam's Razor) and partly empirical ones like uniformitarian bias.
We can develop empirical and quasi-empirical checks on our procedures, but historical linguistics will never be empirical/falsifiable in a Popperian sense. No history is empirical science.
The uniformitarian principle (the future is like the past) is irrational, but either we adopt it or we give up.
[I'm not sure that we need to invoke "the past is like the present" as a special principle. It seems to me that it holds for a general rule of unknowns vs knowns--don't adduce hypotheses unless they explain things better. It doesn't need to be induction except in a psychological sense.]
If someone reconstructs a language with the vowel system [e a o] we can reject it on methodological grounds because no living language has been observed with such a system. [I disagree. This is a problem of attestation and paucity of evidence. If we have *reason* to believe that no such language exists for a reason and that this reason is a principle of human psychology/cognition--which he doesn't even believe in, I think--then we can rule it out. Otherwise, it's a question of "evidence must be strong." Because what if all languages died out, but left us with attestations of only analytic entities? I suppose he's saying it may have been the case, but it's epistemologically shut off from us as a "configuration of entities" that was the case only in the past. Since we reconstruct the Big Bang, I still disagree. Maybe it was unstable and it was easy to get from [e a o] to [i a u] but not vice versa. Maybe one-way/unstable features are dying out. The principle of reconstructing surprising things is that the evidence must be correspondingly greater.] Lass treats [e a o] as a universal--i.e. never found.
Well, he then goes on to say "in the absence of strong counter-evidence", for neutral cases.
And also that the emergence of new phenomena contradicts earlier laws, but is still allowable "in type".
The natural properties of languages, no matter how inexplicable or unintelligible they are (and some are not, but some are), can serve as a basis for certain reconstructive activities.
Lass says little emergence in the timespan of languages we have available to us [--but that's not the same thing as little loss.]
Teleology: functional language change
The null strategy problem: if this tendency is explanatory, there shouldn't be counterexamples. [I disagree, but I'm willing to accept the difficulty of making a falsifiable hypothesis. Yes, if he takes explanatory = nomological.]
What does it mean for a dialect to be "more efficient" than it was before, or than another dialect?
Explanation on perceptual grounds is dubious.
Non-applications of principles are common, so how can an instance of an application be distinguished from a random event? What is the status of tendencies when they are also invoked as causes? Lass thinks we have no respectable notion either of 'explanation' or of 'causality'.
All explanations of "avoidance of homophony" seem to be ad hoc.
Avoidance of homophony would require speakers to calculate the effects of their sound changes in advance, which Lass says is absurd.
Functional explanations seem to be of the type "X would lead to Y, Y happened, therefore X." Fallacy of affirming the consequent.
The alternative is that people produce the offending form, realize what they've done, and self-correct. Lass asks how they would know what the original form is. [Free variation? Presumably there must be someone who is familiar with both forms.]
Appeals to natural tendencies are non-explanatory unless there is and could be evidence independent of their alleged manifestations. Goal might just be a de facto terminus.
Teleology of purpose is admissible in the realm of human behavior, but not language change.
In the absence of a quantification of tolerability of homophony, "intolerable" because identical with "not having been produced", and "tolerable" with "having been produced." This is circular.
In order for us to be able to identify something as the function of a particular behavior, we must be able to say that the behavior "characteristically and normally" contributes to the achievement of a particular goal.
1) The fact that an organism is living is evidence that it is fit to survive.
2) In the case of organisms that have failed to survive, the cause of extinction, where we know it precisely, usually has nothing to do with maladaptation.
The passenger pigeon died because it was not bullet-proof and because it was edible.
When languages cease to be spoken, it is not because of maladaptation to speakers' needs.
Unless we attribute all adaptive change to precognition, the alternative is massive communication failure.
Affirmation of the consequent means identification of functions is limited only by the cleverness of the investigator. Since there are no cases of language death being caused by failure to implement a function, half the argument is missing.
We can't distinguish linguistic functions from epiphenomena. In biology, pumping blood is a function, whereas making whooshing noises is an epiphenomenon. We distinguish by saying that contributing to survival is a good.
i) There are no D-N explanations, because 'laws' of the appropriate kind do not exist.
ii) Probabilistic explanations are not explanations in any reasonable sense. They are post hoc recognitions of generalizations, but they fail to account in any principled way for cases that don't fit. [If they did, wouldn't they be D-N?]
iii) There are no functional explanations, because we have no principled definition of dysfunction.
Causality and the nature of language
Dominant paradigm of expanding knowledge in the natural sciences = "systems of laws + initial conditions = predictions"
This unjustifiably devalues many important kinds of human knowledge
The idea that everything has a cause is metaphysical, not falsifiable. Popper replaces this with a methodological rule that we must not stop looking for causes, because if we do, we've abandoned science.
Lass questions whether language change is an "event" that takes place "in the world" and therefore is subject to this rule.
The actions of living beings are not nomically caused, because there are individuals involved, and individuals transcend cause. [See Dennett on this.]
Causality in living systems
There is a gradient of predictability in living systems, from nearly 100 to 0.
Things that contribute to indeterminacy:
1) randomness of an event with respect to the significance of the event, e.g. mutations
2) uniqueness of individuals
3) complexity of systems
4) emergence
Mayr conclusions:
1) causality is only partly predictive
2) there are sets of causes, many of which are unanalyzed and unanalyzable
Lass: Both the conviction that the universe is causal and the conviction that it can never be shown to be so are metaphysical doctrines.
Scriven argues that you can have explanations that are only partially predictive, and are still explanatory. Lass says no: as far as he can see we are still no further forward than being able to say that if the likely happens, then it was likely.
We could easily enough predict the death of a man hit by an automobile, given enough information about the injuries and some laws about what causes death. [But cf. Popper on the epistemology of probability.]
Necessary and sufficient conditions: X causes Y just in case factors A...N are present. X is necessary but not sufficient.
Radical autonomy
Linguists have traditionally treated language as if it were an autonomous natural object.
Lass does not think that there is any very strong evidence that existing speakers are a necessary assumption.
Parallel to Popper's "epistemology without a knowing subject" there could be a "language without a using/knowing speaker."
If it is legitimate to exclude the speaker, then functional explanation is beside the point.
If the speaker is implicated, this brings language into the non-nomic domain of the psychological and cultural, ruling out all D-N explanations except the trivially physical.
Then Lass talks about how change would be explained in his autonomous model. Change would have to be introduced as an axiom. So would random mutability. The most obvious way to introduce mutability would be to consider language a self-regulating open system that interacts with the extralinguistic world, like an organism and its environment.
Concludes by saying that if languages are totally autonomous systems, there is no reason for them to change, but if they're not, it is still debatable if there are such reasons.
Lass limits the neurophysiological domain, saying that it's responsible only for boring universals. The social and individual control aspect is unpredictable, which is not to say stochastic. It is a function of the neurophysiological domain that it is "easier" to produce a homorganic cluster, but a matter of "choice" (not conscious, human choice) whether to capitalize on this. No change is ever necessary.
Lass rejects the notion that languages have to adapt to anything. Languages are equifunctional.
Language shares some of the characteristics of play.
"Instance of a law" and "unique object or event" are mutually exclusive. History =/= science.
A good discussion of what constitutes individuality. How can we group like events? What is the demarcation between individuals and the types of which they are tokens?
Emergence poses difficulties for reductionism. [Cf. Dennett.]
"I do not think there can be deductive answers to 'why' questions in linguistics."
cultures are norm-governed and rule-governed, not law-governed
"I suggest that it is wrong to say that 'Jones was injected with penicillin' is ever a satisfactory answer to 'Why did Jones get a rash?'"
"The fact that 'explanations' like [Jim was exposed to the measles and there was a high probability of his getting the measles, which is why he got the measles] give us only inductive likelihood, not deductive certainty, makes them not only 'weaker' than D-N explanations, but makes them non-explanatory"
While any explanation entails prediction, prediction does not entail explanation.
probabilistic
"the 'probabilistic' explanation, which is often taken to be useful in the social sciences, biology, and other slightly woolly domains, is not in fact 'explanatory' in any useful sense.'
If you posit "ease of articulation", any counterexamples are extremely problematic. On the one hand, singular counterexamples do not falsify probabilistic theories. This is an epistemological difficulty that threatens to reduce all 'tendency' explanations to vacuity (Lass believes it succeeds).
It is a necessary property of inductive probabilistic explanations that they are strictly speaking non-empirical and don't explain anything. They don't predict for individual members.
A probabilistic explanation has no empirical content, because no individual failure of the "law" can ever be a counter instance. cf. Popper 1968a 146
Many people still insist that probabilistic explanations are both useful and genuinely explanatory.
Markedness: 85% of languages have the unmarked form, 15% have the marked form. How can it be a property of something that N languages have it? How can it be a property of languages that only N of them have it?
"notions like markedness and naturalness have no explanatory significance, but are merely taxonomic cover-terms for observed distributions, or instances of the 'naming fallacy' (virtus dormativa?)"
stochastic laws could exist in languages, just not in the pan-chronic set of all languages, and they are not of the 'natural' or 'markedness' type.
Lass says that if the 15% remained constant in different language families, i.e. some language families lost them and others gained them, that would support a stochastic law (though it would still not be an explanation, just an observation of a distribution).
We should continue to attempt to formulate D-N explanations, because it's a prereq to forming a serious explanation, and gives us the ability to distinguish the deterministic components from the non-deterministic components.
It is not possible to establish any sufficient or even necessary conditions for homorganic nasal assimilation.
What is the status of a "problem" if some languages appear to consider it not to be one? They adopt the "null strategy" approach.
"According to Lightfoot only the fact of change has to be predicted...not the specific change. I find this unsatisfactory."
"Even the naturalness of assimilation becomes either boringly analytic or mysterious."
The acoustic effect could be misinterpreted. But there are more common ways to misinterpret it, so why this one?
The multiple strategy problem is not of course that anything goes, but they operate within pretty wide latitudes, i.e. they are rather loosely determined by physiological or perceptual boundary conditions.
Spectographic similarity or acoustic similarity (to a phonetician) does not necessarily equal perceptual similarity.
Confusion of why with how: specifying that a segment arose because of properties of the airstream, vocal tract, etc. still doesn't invoke genuinely causal factors that might have led to these properties being capitalized on in particular cases in particular ways.
No phonetic explanation which is not nomically necessary is an explanation.
Reconstructions are not testable [(but cf. laryngeals). Also, extrapolations from data sets should be testable. Lass acknowledges that it gives us hope that our techniques are reliable.]
Reconstruction must be justified by methodological criteria (purely conventionalist, like Occam's Razor) and partly empirical ones like uniformitarian bias.
We can develop empirical and quasi-empirical checks on our procedures, but historical linguistics will never be empirical/falsifiable in a Popperian sense. No history is empirical science.
The uniformitarian principle (the future is like the past) is irrational, but either we adopt it or we give up.
[I'm not sure that we need to invoke "the past is like the present" as a special principle. It seems to me that it holds for a general rule of unknowns vs knowns--don't adduce hypotheses unless they explain things better. It doesn't need to be induction except in a psychological sense.]
If someone reconstructs a language with the vowel system [e a o] we can reject it on methodological grounds because no living language has been observed with such a system. [I disagree. This is a problem of attestation and paucity of evidence. If we have *reason* to believe that no such language exists for a reason and that this reason is a principle of human psychology/cognition--which he doesn't even believe in, I think--then we can rule it out. Otherwise, it's a question of "evidence must be strong." Because what if all languages died out, but left us with attestations of only analytic entities? I suppose he's saying it may have been the case, but it's epistemologically shut off from us as a "configuration of entities" that was the case only in the past. Since we reconstruct the Big Bang, I still disagree. Maybe it was unstable and it was easy to get from [e a o] to [i a u] but not vice versa. Maybe one-way/unstable features are dying out. The principle of reconstructing surprising things is that the evidence must be correspondingly greater.] Lass treats [e a o] as a universal--i.e. never found.
Well, he then goes on to say "in the absence of strong counter-evidence", for neutral cases.
And also that the emergence of new phenomena contradicts earlier laws, but is still allowable "in type".
The natural properties of languages, no matter how inexplicable or unintelligible they are (and some are not, but some are), can serve as a basis for certain reconstructive activities.
Lass says little emergence in the timespan of languages we have available to us [--but that's not the same thing as little loss.]
Teleology: functional language change
The null strategy problem: if this tendency is explanatory, there shouldn't be counterexamples. [I disagree, but I'm willing to accept the difficulty of making a falsifiable hypothesis. Yes, if he takes explanatory = nomological.]
What does it mean for a dialect to be "more efficient" than it was before, or than another dialect?
Explanation on perceptual grounds is dubious.
Non-applications of principles are common, so how can an instance of an application be distinguished from a random event? What is the status of tendencies when they are also invoked as causes? Lass thinks we have no respectable notion either of 'explanation' or of 'causality'.
All explanations of "avoidance of homophony" seem to be ad hoc.
Avoidance of homophony would require speakers to calculate the effects of their sound changes in advance, which Lass says is absurd.
Functional explanations seem to be of the type "X would lead to Y, Y happened, therefore X." Fallacy of affirming the consequent.
The alternative is that people produce the offending form, realize what they've done, and self-correct. Lass asks how they would know what the original form is. [Free variation? Presumably there must be someone who is familiar with both forms.]
Appeals to natural tendencies are non-explanatory unless there is and could be evidence independent of their alleged manifestations. Goal might just be a de facto terminus.
Teleology of purpose is admissible in the realm of human behavior, but not language change.
In the absence of a quantification of tolerability of homophony, "intolerable" because identical with "not having been produced", and "tolerable" with "having been produced." This is circular.
In order for us to be able to identify something as the function of a particular behavior, we must be able to say that the behavior "characteristically and normally" contributes to the achievement of a particular goal.
1) The fact that an organism is living is evidence that it is fit to survive.
2) In the case of organisms that have failed to survive, the cause of extinction, where we know it precisely, usually has nothing to do with maladaptation.
The passenger pigeon died because it was not bullet-proof and because it was edible.
When languages cease to be spoken, it is not because of maladaptation to speakers' needs.
Unless we attribute all adaptive change to precognition, the alternative is massive communication failure.
Affirmation of the consequent means identification of functions is limited only by the cleverness of the investigator. Since there are no cases of language death being caused by failure to implement a function, half the argument is missing.
We can't distinguish linguistic functions from epiphenomena. In biology, pumping blood is a function, whereas making whooshing noises is an epiphenomenon. We distinguish by saying that contributing to survival is a good.
i) There are no D-N explanations, because 'laws' of the appropriate kind do not exist.
ii) Probabilistic explanations are not explanations in any reasonable sense. They are post hoc recognitions of generalizations, but they fail to account in any principled way for cases that don't fit. [If they did, wouldn't they be D-N?]
iii) There are no functional explanations, because we have no principled definition of dysfunction.
Causality and the nature of language
Dominant paradigm of expanding knowledge in the natural sciences = "systems of laws + initial conditions = predictions"
This unjustifiably devalues many important kinds of human knowledge
The idea that everything has a cause is metaphysical, not falsifiable. Popper replaces this with a methodological rule that we must not stop looking for causes, because if we do, we've abandoned science.
Lass questions whether language change is an "event" that takes place "in the world" and therefore is subject to this rule.
The actions of living beings are not nomically caused, because there are individuals involved, and individuals transcend cause. [See Dennett on this.]
Causality in living systems
There is a gradient of predictability in living systems, from nearly 100 to 0.
Things that contribute to indeterminacy:
1) randomness of an event with respect to the significance of the event, e.g. mutations
2) uniqueness of individuals
3) complexity of systems
4) emergence
Mayr conclusions:
1) causality is only partly predictive
2) there are sets of causes, many of which are unanalyzed and unanalyzable
Lass: Both the conviction that the universe is causal and the conviction that it can never be shown to be so are metaphysical doctrines.
Scriven argues that you can have explanations that are only partially predictive, and are still explanatory. Lass says no: as far as he can see we are still no further forward than being able to say that if the likely happens, then it was likely.
We could easily enough predict the death of a man hit by an automobile, given enough information about the injuries and some laws about what causes death. [But cf. Popper on the epistemology of probability.]
Necessary and sufficient conditions: X causes Y just in case factors A...N are present. X is necessary but not sufficient.
Radical autonomy
Linguists have traditionally treated language as if it were an autonomous natural object.
Lass does not think that there is any very strong evidence that existing speakers are a necessary assumption.
Parallel to Popper's "epistemology without a knowing subject" there could be a "language without a using/knowing speaker."
If it is legitimate to exclude the speaker, then functional explanation is beside the point.
If the speaker is implicated, this brings language into the non-nomic domain of the psychological and cultural, ruling out all D-N explanations except the trivially physical.
Then Lass talks about how change would be explained in his autonomous model. Change would have to be introduced as an axiom. So would random mutability. The most obvious way to introduce mutability would be to consider language a self-regulating open system that interacts with the extralinguistic world, like an organism and its environment.
Concludes by saying that if languages are totally autonomous systems, there is no reason for them to change, but if they're not, it is still debatable if there are such reasons.
Lass limits the neurophysiological domain, saying that it's responsible only for boring universals. The social and individual control aspect is unpredictable, which is not to say stochastic. It is a function of the neurophysiological domain that it is "easier" to produce a homorganic cluster, but a matter of "choice" (not conscious, human choice) whether to capitalize on this. No change is ever necessary.
Lass rejects the notion that languages have to adapt to anything. Languages are equifunctional.
Language shares some of the characteristics of play.
"Instance of a law" and "unique object or event" are mutually exclusive. History =/= science.
A good discussion of what constitutes individuality. How can we group like events? What is the demarcation between individuals and the types of which they are tokens?
Emergence poses difficulties for reductionism. [Cf. Dennett.]
