Work and Play: Simulating Language Contact
Roland M ¨uhlenbernd
1& Jason Quinley
11
Seminar f ¨ur Sprachwissenschaft, Universit ¨at T ¨ubingen
Goals
I examine language contact on a social network
I what social factors are important?
I are conventions inevitable or contingent?
I extend signaling game accounts of language change
I investigate theories of social conditions and convention adoption
I how does context affect signaling strategies?
I does connectivity increase convention adoption?
Motivation: Systematic Divisions in English
I Q: Why are French cognates in English more prestigious?
I A: Historians cite social conditions after the Norman invasion personal abstract
freedom liberty knowledge science belief faith
brotherly fraternal
animal meat swine pork cow beef sheep mutton deer venison
Table: Systematic division of meaning space between words of Germanic and French origin.
Context Signaling Games: Structure
Q: How do conventions arise?
Informally:
I Senders and Receivers
I Nature chooses a context
I Context dictates possible topics
I Senders observe a topic
I Senders send a message
I Receivers interpret it with action
I Success upon coordination
I Failure otherwise
Formally:
I Sender S and Receiver R
I PC(c) and PT(t|c)
I C = {dinner, farm}
I T = {meat, animal}
I M = {swine, pork}
I A = {meat, animal}
I U(ti, aj) =
1 if i = j
0 else
Context Signaling Games: Extensive Form
N
N
S
R
1 0
R
1 0
S
R
0 1
R
0 1
N
S
R
1 0
R
1 0
S
R
0 1
R
0 1
.9 .1 .1 .9
dinner farm
’meat’ ’animal’ ’meat’ ’animal’
”swine” ”pork” ”swine” ”pork” ”swine” ”pork” ”swine” ”pork”
’m.’ ’a.’ ’m.’ ’a.’ ’m. ’a.’ ’m.’ ’a.’ ’m.’ ’a.’ ’m.’ ’a.’ ’m.’ ’a.’ ’m.’ ’a.’
Figure: (N.B. dashed lines connect situations the recipient cannot distinguish; leaves resulting in 1 depict successful communication.)
Social Network Structure
I Each node is an agent
I Each edge represents a CSG
I Size
represents social class
I Black dots represent invaders
Figure: Scale-free network with 300 agents.
Experimental Settings
I social network with scale-free properties
I 300 agents with a randomly chosen social status
I agents communicate with neighbors by CSG
I agents switch between sender and receiver
I agents learn by learning dynamics
belief learning + best response dynamics
I when all agents have learned the same strategy → Norman invasion by replacing 30 agents
Result 1: Context-Dependent Strategy
Dinner: ’meat’
’animal’ ”swine” Farm: ’meat’
’animal’ ”swine”
Result: Before the Norman Invasion, all agents learn this context-dependent strategy
The Norman Invasion: Different Settings
Experiment 1: Randomly chosen agents are replaced by Norman invaders
Experiment 2: Only agents with a social status higher than a given lower limit are candidates to be replaced
Experiment 3: Agents must have a high social status paired with a high degree of centrality to be candidates to be replaced
Result 2: Emergence of two Strategies
S1: ’meat’ ”pork”
’animal’ ”swine” S2: ’meat’
”swine”
’animal’
”pork”
Results:
I the whole society learns either S1 or S2
I both strategies are context-independent
I the emergence of each strategy is equiprobable in Experiment 1, but nor for Experiment 2 or 3
Result 3: Local properties
perc.S 1-trials
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8 9 β
perc.S1-trials
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8 9 β Results:
I by increasing the threshold β of the replaced agents, the probability for a society-wide spread of strategy S1 raises (Experiment 2)
I if the social status σ is also correlated with a higher degree d and
therefore a higher local influence, the increasing of threshold β of the replaced agents leads to an accelerated raising of the probability for a society-wide spread of expected strategy S1 (Experiment 3)
Conclusion & Outlook
I language conventions depend on social factors
I context and connectivity accelerate and ensure certain conventions emerging
I without uniform social factors, patterns like those seen in modern English are left to chance
I social networks are a promising ground for treatments of language variation and change
Jason Quinley & Roland M ¨uhlenbernd Work and Play: Simulating Language Contact Universit ¨at T ¨ubingen