*Introduction *Speakers can recognize inter-speaker variability in various
pragmatic phenomena (e.g., uncertainty expressions [1], or
under-specification of adjectives [2]) and to adapt to the speakers’
different preferences of language use, in cases where such distinctions
facilitate the derivation of meaning from the utterances of a specific
speaker. In this study, we asked whether speaker-specific learning can
occur when the language use of different individuals does not entail
different meanings, but instead is based on differences in stylistic
preferences (see [3] for an account of syntactic-stylistic adaptation). We
used the weak adjective ordering preferences in Hebrew (a post-nominal
language, where adjectives appear after nouns) [4], such that choosing to
use one order over another does not convey a meaning modification.
*Methods* *Experiment 1:* Native Hebrew speakers (N=60) took part in a
learning paradigm consisted of an exposure phase, where one speaker used a
certain order and the other a different order, and an explicit test phase
that tested whether the participants learned these speaker-specific
preferences. In the exposure phase, participants saw 96 images of shapes
which had 3 distinctive visual features: size, color and pattern (Figure
1), and had to judge whether they matched an auditory description. In half
of the cases, the descriptions matched the image, and in the other half,
they did not. The adjective orders varied based on the speaker, in 3
between-subject conditions (the 3 combinations of the two most common
orders (Noun-Size-Color-Pattern and Noun-Color-Size-Pattern), and the most
deviant one (Noun-Pattern-Size-Color), based on [4]). The auditory
descriptions were recorded by a male and a female to ease their
discrimination. The characters always used the same adjective order in
their 48 descriptions which were presented in 4 interleaved speaker blocks
(pseudo-randomized and counter-balanced across participants and
conditions). In the test phase, participants had to decide which speaker
could have uttered 24 written three-adjective phrases, similar to those
used in the exposure phase (Figure 2). Half of the descriptions included
the adjective order used by one speaker and half included the order used by
the other speaker, presented in a randomized sequence. *Experiment 2:*
Native Hebrew speakers (N=20) took part in the same paradigm as in
Experiment 1, but with a fourth condition in which both speakers produce
less common orders [4], similar in form (Noun-Pattern-Size-Color and
Noun-Size-Pattern-Color).
[image: image.png]
*Figure **1**.* An example of a stimulus in the training phase. Each of the
orders was produced either by Speaker A or by Speaker B. In half of the
trials in the training phase the description and the image mismatched (one
of the features was inappropriate for the image). Participants were
required to press F if the description matched the image or K if it did not.
[image: image.png]
*Figure **2**.* An example for a trial in the test phase. Participants were
instructed to choose who of the speakers could have uttered the written
descriptions. Originally, descriptions were presented in Hebrew with Hebrew
letters. *Naama is a common female name and Yoav is a common male name and
they are not used for both sexes.
*Results* Participants who performed at above-chance level (as measured by
individual-level binomial tests) were classified as ‘learners’. *Experiment
1:* The conditions in which one common order was presented with the most
deviant order (Noun-Size-Color-Pattern/Noun-Pattern-Size-Color &
Noun-Color-Size-Pattern/Noun-Pattern-Size-Color) yielded substantially
higher number of ‘learners’ (11/20 & 12/20, respectively; Figure 3) than
the condition where the two common orders
(Noun-Size-Color-Pattern/Noun-Color-Size-Pattern) were used (2/20; Figure
3). An equality of proportions test revealed that there were more
successful learners in both conditions in which one of the speakers used
the deviant order than in the condition in which both speakers produced a
common order (*p*s < 0.01). There was no significant difference between
these two conditions (*p* = 1). *Experiment 2:* The condition in which both
orders were uncommon but similar in form yielded higher rates of successful
learning (Number of ‘learners’: 9/20) than the condition where the two
common orders were used (*p* = 0.03).
[image: image.png]
*Figure **3**.* Correct answers in the trial phase, by condition. SCP =
Noun-Size-Color-Pattern; CSP = Noun-Color-Size-Pattern; PSC =
Noun-Pattern-Color-Size, SPC = Noun-Size-Pattern-Color. SCP and CSP are the
most common and natural adjective orders and PSC and SPC are uncommon.
*Discussion* When both speakers produced the most common adjective orders
in Hebrew (with color, size and pattern adjectives), participants performed
worse on the test phase than when a deviant order was included. This cannot
be attributed to the similarity in form in these two orders, given the more
successful learning in Experiment 2. The ability to distinguish between
speaker-specific styles was enhanced when at least one of the speakers
produced a deviant order. This suggests that listeners are better at
detecting speaker-specific language use when such use deviates from common
use, even when no change in meaning is associated with this variability.
These results are in line with evidence regarding surprisal-driven learning
[5].
*References*
[1] Schuster, S., & Degen, J. (2020). I know what you're probably going to
say: Listener adaptation to variable use of uncertainty expressions.
*Cognition*, *203*, 104285.
[2] Pogue, A., Kurumada, C., & Tanenhaus, M. K. (2016). Talker-specific
generalization of pragmatic inferences based on under-and over-informative
prenominal adjective use. *Frontiers in psychology*, *6*, 2035.
[3] Ostrand, R., & Ferreira, V. S. (2019). Repeat after us: Syntactic
alignment is not partner-specific. *Journal of memory and language*, *108*,
104037.
[4] Trainin, N., & Shetreet, E. (2020). It's a dotted blue big star: on
adjective ordering in a post-nominal language. *Language, Cognition and
Neuroscience*, 1-22.
[5] Lai, W., Rácz, P., and Roberts, G. (2020) Experience with a linguistic
variant affects the acquisition of its sociolinguistic meaning: An
alien-language-learning experiment. *Cognitive Science* 44(4): e12832.