difference in form without difference in meaning

T. Florian Jaeger tiflo at csli.stanford.edu
Sat Aug 6 00:51:57 UTC 2011

Hi Fritz,

I've recently spent more time thinking about the very same question. I am,
however, not even sure that it is a well-formed question. At least if we're
willing to base our decision about the correct answer on data from actual
language understanding (I am not sure that meaning can be meaningfully
defined if we don't commit to this assumption).

The mapping from perceptual input to meaning is noisy, so that two different
forms can most certainly lead to the same set of inferences. This might seem
irrelevant to your question, but I think it might affect the answer. Meaning
differences that are associated with linguistic forms that are very likely
to lead to overlapping perceptual inputs are unlikely to be learnable.

You were asking about syntactic alternatives (or syntactically related forms
that share the same meaning). But even for those, there are some that differ
very little in perceivable linguistic form (e.g. that-omission, which you
mentioned; or to-deletion after *help* in English). I think there are
reasons to suspect that such difficult to perceive differences (in
conversational speech either of these two words is often going to reduced to
some co-articulatory information on the surrounding words) are unlikely to
be associated with strong meaning differences. This, of course, hasn't kept
people from claiming such meaning differences (e.g. Yaguchi, 2001; Dor, 2005
for that-omission). However, those meaning differences that seem so apparent
when we look at written language offline seem to be hard to confirm in
studies. Some years ago, Rafe Kinsey (back then an undergrad at Stanford)
conducted a study (together with Tom Wasow and me) on alleged meaning
differences between complement clauses with "that" and those without. We
didn't find any evidence for meaning differences. This, of course, doesn't
mean that there are none. What I thought was interesting is that I used to
bug some of my fellow students about whether they felt that complement
clauses with "that" were different from those without "that". Almost all of
them felt that there was a meaning difference. However, none of them agreed
on what the difference was and several of them even had the exact opposite
opinion! I find that example, though anecdotal in nature, quite instructive:
perhaps we can't help thinking that there are meaning differences, but that
doesn't mean that they are stable enough to become successfully associated
with one of the two forms.

I've been fascinated by the fact that most of my fellow psycholinguists
simply assume that there are no (relevant) meaning differences between
syntactic alternatives. They are quite fine running active vs. passive
experiments where effects of animacy or givenness of the agent or theme on
the preferred choice between the two structures are interpreted as evidence
about the underlying structure of the production system, rather than as
evidence for meaning differences. Arguably, they have one thing on their
side: these and other factors have the predicted effects across many
structural alternations across many languages (cf. e.g, Branigan et al 2009;
Jaeger and Norcliffe, 2009 for overviews).

I agree with the other comments that differences in form often end up
becoming associated with differences in meaning, but I think that for many
alternations, at any given point in time, differences in meaning **are just
one of several factors* *that determine speakers' preference between the two
forms. For example, there is evidence from heavy NP shift that sometimes the
only reason why it happens is that the heavy NP was not yet ready for
articulation when the speaker had to make a choice as to how to maintain
fluency (Wasow, 1997). Also, would we really want to claim that the same
speakers describing the same pictures reliably choose their argument order
(e.g. in the ditransitive structure) based on the number of words in the
theme/recipient constituent because that affects how likely they are to
think of the picture one way or another, thereby affecting what subtle
meaning difference they want to convey? It's possible, but I wouldn't bet my
money on it. Do we want to attribute the fact that more predictable relative
and complement clauses are less likely to have a relativizer/complementizer
"that" to meaning differences (same of passive RCs, to-omission,
contraction, etc.; Jaeger, 2006; 2010, 2011; Wasow et al., 2011; Levy and
Jaeger, 2007; Frank and Jaeger, 2008)? From a processing-perspective this
makes perfect sense, whereas the meaning theories that have been evoked
differ for each of those cases.

All of this is not to say that comprehenders aren't incredibly sensitive to
the motivations behind speakers' preferences. Actually, there's plenty of
evidence for that. For example, Arnold et al show that comprehenders know
that speakers are more disfluent before difficult words and that knowledge
allows them to process words that are a priori more difficult much faster
after a disfluency. Similarly, comprehenders expect difficult material after
a "that" at the onset of a complement or relative clause and if they don't
get it this slows comprehension (relatively speaking; Race and MacDonald,
2003). I think it's perceivable that these processing-based expectations can
easily create the 'illusion' of a meaning difference. They are also likely
to 'cause' meaning differences in the long run, but it seems to me (from the
data I have seen in experiments) that these meaning differences can be quite
fickle for a long time and can be overriden by processing preferences. One
of my students, Judith Degen, recently started looking into the possibility
that such processing preferences might even affect the choice between two
rather meaning-different forms (she's focusing on "some X" vs. "some of the
X"; recently presented at XPRAG 2011).

So my current best-bet-speculation (see also my thesis, Chapter 6.2.2) is
that speakers, when they encode their intended meaning into linguistic
forms, probabilistically select between different forms and that this
selection is affected by the strength of connections between different
meanings and that form as well as processing considerations (such as the
well-documented preference to avoid speech suspension; for refs see, e.g.
Clark and Fox-Tree, 2002; Fox-Tree and Clark, 1997; V. Ferreira and Dell,
2000; V. Feirreira 1996; Bock, 1987).

so in this sense (if my argument makes sense), it would be misleading to
think that most alternatives in syntactic alternations are meaning distinct
unless you're willing to accept any difference in the probability
distribution over inferred meanings given a linguistic form as evidence for
difference meanings -- in that case, it would probably hold that no two
forms are the same (including no two actual acoustic realizations of the
same syntactic structure, since they will differ in speech rate, etc., which
will affect some inferences the comprehender might draw).

I think for any stronger claim about meaning differences there would need to
be testable (and preferably quantifiable) theories about those meaning
differences, so that they could be pitched against well-established theories
of speakers' preferences during incremental language production.

I hope some of this is useful? This would be an awefully long email if it
turned out to be completely incomprehensible ;).


One final thought - didn't Bresnan et al (2007) also discuss alleged meaning
differences for the ditransitive alternation?

> Message: 1
> Date: Thu, 4 Aug 2011 14:17:27 -0700 (PDT)
> From: Frederick J Newmeyer <fjn at u.washington.edu>
> Subject: [FUNKNET] difference in form without difference in meaning
> To: Funknet <funknet at mailman.rice.edu>
> Message-ID:
>        <alpine.LRH.2.01.1108041417270.26399 at hymn33.u.washington.edu>
> Content-Type: TEXT/PLAIN; charset=US-ASCII
> Dear Funknetters,
> I am looking for convincing examples of where two syntactically-related
> sentence-types manifest clearly identical meanings, where 'meaning' is taken
> in its broadest sense, including discourse-pragmatic aspects. Another way of
> putting it is to say that I am looking for two sentence types that in early
> TG would have been related by 'optional rules', but which absolutely do not
> differ in meaning. It's not so easy to come up with good examples, once
> differences in topicality and focus are allowed as meaning differences. One
> possible example that comes to mind are sentences with or without
> complementizer-deletion, such as 'I knew that he'd be on time', vs. 'I knew
> he'd be on time'. But even here there have been argued to be meaning
> differences.
> One possibility that has been suggested to me is from Early Modern English,
> when many speakers could say both 'Saw you the bird?' and 'Did you see the
> bird?' Does anybody have evidence that there were subtle meaning differences
> here?
> I had always been quite skeptical of Dwight Bolinger's idea that
> differences in (lexical and syntactic) form always correlate with meaning
> differences. But I have become less skeptical recently.
> Thanks,
> --fritz
> Frederick J. Newmeyer
> Professor Emeritus, University of Washington
> Adjunct Professor, University of British Columbia and Simon Fraser
> University
> [for my postal address, please contact me by e-mail]

More information about the Funknet mailing list