PresentationPDF Available

Agent-based simulations of affix change: Interacting mechanisms under social dynamics

Authors:

Abstract

Affixes change over time, especially under social dynamics, such as adult language contact (Lupyan & Dale, 2010; Sinnemäki & Di Garbo, 2018) and dialect contact (Trudgill, 1986). To study the mechanisms behind affix change, we propose to simulate communication between speakers using agent-based computer simulations (Smith, 2014) in conjunction with real-world data. We present two case studies, to illustrate how agent-based models allow studying complex interactions between mechanisms in affix change, particularly for conditions which are infeasible to manipulate in the real word. These case studies show that agent-based computer simulations can shed light on interacting mechanisms behind change of affixes: phonological processes playing a role in morphological simplification and conversational mechanisms influencing spread of innovations.
Peter Dekker AI Lab, Vrije Universiteit Brussel
Sonja Gipper Institute for Linguistics, University of Cologne
Marian Klamer Leiden University Centre for Linguistics (LUCL)
Bart de Boer AI Lab, Vrije Universiteit Brussel
Affixes symposium, Turku, Finland
17-18 August 2023
Agent-based simulations of affix change:
Interacting mechanisms under social dynamics
1
Affix change under social dynamics
Affix change through social dynamics: situation with at least two heterogeneous
groups (e.g. language contact, dialect contact, groups within community)
Often interactions between mechanisms
Agent-based models: Computer simulations of populations of speakers
Individual is focal point, study population behaviour
Unintrusive, testing in reality not always feasible
Modelling makes hypotheses explicit
Two case studies:
Phonotactic mechanisms in contact-induced morphological simplification in Alorese
Conversational priming and spread of innovations (Lithuanian dialects)
2
Phonological mechanisms in contact-induced
morphological simplification in Alorese
3
Contact-induced morphological simplification
Language contact involving adult speakers could lead to morphological simplification (Wray & Grace, 2007; Lupyan &
Dale, 2010; Trudgill, 2011)
But what are actuating factors (Weinreich et al., 1968) on contact-induced change? Why does it happen in one
language and not in another?
Specific sociodemographics of contact scenario (Sankoff, 2004; Ross, 2013)
Could phonotactic factors play a role?
1. Phonology itself can be affected by contact (Napoleão de Souza & Sinnemäki, 2022; Blaxter, 2017;
Blevins, 2017)
2. Phonological and morphological complexity show positive correlation (Easterday et al., 2021)
3. Phonology of language (e.g. Germanic stress shift for Scandinavian) can be pre-condition for
contact-induced morphological simplification (Kusters, 2003)
4. Consonant clusters are cross-linguistically difficult to learn (Carlisle et al., 2001)
Our hypothesis: The phonotactics of a language, specifically syllable structure (and more specifically
avoidance of consonant clusters) could be a factor in the morphological simplification of that language under
contact.
4
Case study: Alorese
Alor & Pantar islands, Eastern Indonesia
Alorese (Austronesian)
Contact with Papuan Alor-Pantar languages (Timor-Alor-Pantar family)
Alorese lost verbal subject marking compared to sister language Lamaholot (Klamer, 2020)
Some verbs express subject using prefix, others using suffix
Alorese lost suffixes, while prefixes are retained
Could adult language contact (Lupyan & Dale, 2010), in combination with phonotactic mechanisms, have
caused morphological simplification?
(map: Owen Edwards and UBB)
(photo: Yunus Sulistyono)
5
Moro (2019)
Alorese: Phonotactic factors?
6
prefix suffix
n-enung
3SG-drink
'she drinks'
hitun-na
count-3SG
'he counts'
Lewoingu Lamaholot (Nishiyama & Kelen, 2007)
n-enung
3SG-drink
'she drinks'
hitun-Ø
count-3SG
'he counts'
Alorese (Klamer, 2011)
All prefixing verbs are vowel-initial: phonotactic factor at play?
Some suffixing verbs+suffix create consonant cluster (⅓ of verbs in grammar)
Consonant clusters disencouraged in Alorese (Klamer, 2011; Nagaya, 2011) and
contact languages (Schapper, 2014)
Could avoidance of consonant clusters during incomplete L2 transmission be a
factor in contact-induced morphological simplification?
Agent-based model
Agent-based model of intergenerational transmission
(cf. Kusters, 2003):
L1 speakers initialised with full morphology: get language
faithfully transmitted
L2 speakers learns language through interaction with
previous generation (L1 & L2)
Language game (Steels, 1998)
Meanings: lexical concepts of verbs + person
(e.g. to go-2SG, to have-3SG)
Signals: verb affixes (e.g. k-, t-, -ko)
Listener saves affix when communicative success
Test mechanisms:
Phonotactic reduction mechanism: L2 speakers drop full affix
in production when consonant cluster arises
hitun-na
CVCVC-CV
Generalisation mechanism (affix prior): use distribution over
affixes from all concepts during production, instead of just this
concept
Evaluate model for different proportions of L2 speakers:
find relationship between adult language contact,
phonotactic reduction and generalisation 7
Results: Phonotactic reduction and generalisation
No generalisation Generalisation 10%
With generalisation, phonotactic reduction gives morphological simplification with increasing L2
8
Intermediate conclusions
Some evidence for simplifying effect of phonotactic reduction in Alorese,
but not very strong
Model surprisingly resilient to strong reduction mechanism (through
meaning in model)
Generalisation needed to spread empty affix from verbs with consonant
clusters to verbs without consonant clusters
9
Conversational priming in Lithuanian dialects
In the context of the project Conversational priming in language change (Universität zu Köln)
Thanks to Eugen Hill, Pascal Coenen.
Thanks to:
The Yurakaré Nation
Jeremías Ballivián Torrico
University of Cologne Excellent Research Support Programme, FORUM
University of Cologne Cluster Development Programme
DFG Netzwerk Interaktionale Linguistik (413161127)
Volkswagenstiftung (81821 & 83448)
Research Foundation – Flanders (FWO) PhD Fellowship (11A2821N)
10
Hypothesis
💬 Not much research on the role of conversation in language change
🚊As the prime habitat for language, conversation must provide infrastructures
for linguistic innovations to spread from one speaker to another
HYPOTHESIS:
💡 Conversational priming in repetitional responses leads to faster spread of
innovative forms
11
Conversational priming: in repetitional responses ("Do you want coffee?" -"I
do."), some linguistic items are primed and predicted to change faster (Gipper,
2020)
Subject marking as clear test case with contrasting predictions:
Conversational priming works only on third person: other persons as contrast
Frequency makes opposite prediction:
3SG most conservative (Diessel, 2007) vs 3SG most innovative
(conversational priming)
12
Conversational priming: Asymmetries in repetitional responses
(Yurakaré, Isolate, Bolivia)
(1) A: adojla balip
bali-p
go -2PL.SBJ
‘Did you (pl.) walk?’
(2) A: dulawla
dula -w=la
make -3PL.SBJ=COMM
‘Did they build it?’
B: adojla balitu
a-dojjo=la bali -tu
go -1PL.SBJ
‘We walked.’
B: dulaw
dula-w
make-3PL.SBJ
‘They built it.’
data from: van Gijn, Hirtzel, Gipper & Ballivián Torrico 2011: Conversation-NL, YURGVDP08oct06-01
Priming
Repeat may force B to copy form from A.
Priming facilitates reuse, repetition boosts
priming (Gipper 2020)
No priming
Deictic shift, no reuse of person marker
13
Spread of innovations through priming?
Inspiration for model: Lithuanian dialects Zietela and Lazūnai
3SG/PL is only form that changes in innovative dialect, caused by priming?
eĩti ‘to go’ Zietela
(conservative)
Lazūnai
(innovative)
inf. eĩti eĩti
1SG eimù eimù
2SG eimì eimì
3SG=PL eĩti eĩma
1PL eĩmam eĩmam
2PL eĩmat eĩmat
Vidugiris 2014: 198–200
Rozwadowski 1995: 136, thanks to Eugen Hill
Zietela, Belarus (Wikipedia)
PREDICTION In languages with person marking, if an innovation occurs, third person
markers will diffuse faster than first and second person markers.
14
Agent-based model
Computer model of repeats in conversations (cf. general models
of innovation spread: Pierrehumbert et al. 2014; Josserand et al.
2021)
Interaction in population of agents: conservator (0% innovative
form) vs. innovator (90% innovative form) agents
Meanings: 1SG, 2SG, 3SG
Forms: conservative vs. innovative
Probability increase of form during both production and perception
Priming:
If person different (1/2SG) Sample form from own distribution
If person same (3SG) Use same form as questioner
3SG
3SG
1SG
2SG
15
Interaction
Speaker A Speaker B
Randomly sample person (1/2/3SG) to talk about
Sample form to use from probability distribution
Send form
Increase sent form
Increase received form
Determine person to answer (1→2, 2→1, 3→3)
If person different (=1/2SG):
Sample form from probability distribution
If person same (=3SG):
Use same form as questioner
Send form
Increase sent form
Increase received form
new = (old+increase)/(1+increase)
other = old/(1+increase)
16
Results basic model
Priming (3SG) converges faster than non-priming (1SG)
Model converges to population mean (differential equations)
Favouring the innovative form gives more realistic dynamic (S-curve; Blythe & Croft, 2012)
Basic model Basic model + favour innovative
17
Conversational priming versus frequency
Work in progress: contrast conversational priming with frequency
Most frequent 3SG (Seržant & Moroz, 2022) would be most conservative
Can conversational priming counter this?
Frequency implemented by forgetting mechanism: frequent concepts overcome forgetting
Tentative conclusion: Conversational priming does not counter frequency, but speeds up existing
processes
3SG freq 50%
No conv priming
3SG freq 50%
Conv priming 18
Conclusion
Agent-based models can shed light on interacting mechanisms behind affix
change in situations of social dynamics:
Phonotactic mechanisms, in combination with adult language contact,
could lead to morphological simplification
Conversational priming could lead to faster spread of innovations, once
invented by other mechanisms
Contact us if you have questions or ideas! peter.dekker@ai.vub.ac.be
Kiitos kaikille!
19
ResearchGate has not been able to resolve any citations for this publication.
Computer model of repeats in conversations (cf. general models of innovation spread
  • Pierrehumbert
Computer model of repeats in conversations (cf. general models of innovation spread: Pierrehumbert et al. 2014; Josserand et al.