This page was automatically generated by NetLogo 3.1.4. Questions, problems? Contact feedback@ccl.northwestern.edu.
The applet requires Java 1.4.1 or higher. It will not run on Windows 95 or Mac OS 8 or 9. Mac users must have OS X 10.2.6 or higher and use a browser that supports Java 1.4. (Safari works, IE does not. Mac OS X comes with Safari. Open Safari and set it as your default web browser under Safari/Preferences/General.) On other operating systems, you may obtain the latest Java plugin from Sun's Java site.
created with NetLogo
view/download model file: lw-model.nlogo
This is a simple model of loanword transmission through a speech community, focusing on the nativization over time of non-native phonological segments or sequences (such as [ti] in Japanese), as described in Crawford (2007).
The agents are embedded in a social network generated using the algorithm in Davidsen, Ebel, and Bornholdt (2002). Each agent represents a given loanword using a variable prob-t, which represents the probability that the agent will produce the loanword with [ti] instead of [chi]. When the model is first setup, a certain fraction of the agents (controlled using the fraction-bilinguals slider) is randomly chosen to be "bilingual", meaning their prob-t is set to the value of the initial-bilingual-prob-t slider (which is 1 by default). The other agents are "monolingual" and have their prob-t set to the value of initial-prob-t (0 by default).
On each time step, the agents which know the loanword (which are initially just the bilingual speakers) will pick one of their neighbors in the network to talk to, and produce a token of the loanword with either [ti] or [chi] depending on their values of prob-t. The neighbor will then update its value of prob-t using the following formula:
prob-t = learning-rate * input + (1 - learning-rate) * prob-t
where input = 0 if the token has [chi], and 1 if it has [ti]. Then, with probability prob-learn-word, the neighbor "learns" the loanword and will begin producing it on the next time step.
The setup button will create a new network with the number of agents specified by the num-nodes slider. They will initially be placed randomly on the screen, which makes it hard to see the structure of the social network; use the layout button to fix that (but beware that it runs rather slowly, which is why setup doesn't automatically layout the nodes). The "Distance from bilingual" histogram, on the right, will show how many nodes in the resulting network are a given distance from a bilingual speaker (if a node is connected to two or more bilinguals, then the shortest distance is used).
The go button will then run the model as described above. You will also see new links being added at every time step, and nodes occasionally being removed from the graph as well, based on the network algorithm in Davidsen et al. (2002). Agents are colored gray if they haven't learned the loanword yet, and a shade ranging from blue to red based on their value of prob-t (0=blue, 1=red) if they have learned the loanword. The "Nativization by distance" and "Nativization per word" graphs will be updated on each time step based on the values of prob-t among nodes which have learned the loanword.
Click on the highlight button and then hover the mouse over a node in the display to examine its properties, such as its value of prob-t, its distance from the nearest bilingual node, how many times it has been exposed to the loanword, and so forth.
Models with more than a few hundred nodes will probably run very slowly if the node display is on; in that case, turn off the "display?" switch before clicking the setup button.
The "update-after-learned?" switch controls whether nodes will continue to update prob-t after they have "learned" the loanword.
"initial-prob-from-lex?" controls whether the initial value of prob-t when a node is exposed to a new loanword is based on the node's values of prob-t for other loanwords it has already been exposed to. If off, then prob-t will initially be set to the value of the initial-prob-t slider for all new loanwords; if on, then prob-t will be set to the mean prob-t among all other loanwords, or initial-prob-t if no other loanwords are known yet.
Finally, the prob-replace-node slider controls the probability that one of the nodes (and all of its links) will be deleted and replaced by a new node which is randomly linked to one of the existing nodes. This is a parameter used in the network model of Davidsen et al. (2002).
You will see in the "Nativization by distance" graph that nodes which are very close to a bilingual speaker will tend to have less nativization than those that are far away. Also, as you watch the node display, you will see that, for small values of fraction-bilinguals (around 0.01), generally only those nodes that are within a highly-connected clique containing a bilingual speaker will have a prob-t > 0.5, whereas nodes that are only weakly connected to a bilingual-containing clique will tend to have a much lower value of prob-t.
Try running the model using various values for fraction-bilinguals. As fraction-bilinguals becomes larger (around 0.25), the network structure changes so that most nodes tend to be relatively near one of the bilinguals, making nativization less likely.
Also note the effects of changing the learning-rate or prob-learn-word (both of which are 0.1 by default); larger values of these result in more variation in values of prob-t among all of the nodes.
In future versions of this model, I plan on using more realistic lexical representations for loanwords, to simulate the phonological neighborhood effects discussed in Crawford (2007).
Also, increasing the fraction of bilinguals is not an accurate representation of what has actually occurred in the Japanese speech community during the period of contact with English. While English education has increased such that nearly all Japanese students nowadays study English in school, actual bilingualism in English remains relatively rare. One way to simulate this would be to have the few bilingual nodes in the network connected to many more nodes than they would be on average, to represent the effect of English teachers in the network.
Crawford, C. (2007). An evolutionary account of loanword-induced sound change in Japanese. Presentation given at 31st Penn Linguistics Colloquium, Philadelphia, PA. http://www.people.cornell.edu/pages/cjc26/penn-handout.pdf
Davidsen, J., Ebel, H., and Bornholdt, S. (2002). Emergence of a small world from local interactions: Modeling acquaintance networks. Physical Review Letters 88 (12), 128701.