Each construction represents an one-to-one mapping between a noun and its gender. In the case of gender acquisition for a word, the agent adopts a new construction for the respective noun, such that all the semantic properties of the word are kept, while the score is initialized and the gender takes the new acquired value. The old and new constructions now become gender competitors for the same noun.
Computer simulations of language change notes
This website collects my personal notes on Computer simulations of language change. These notes are provided to bring full transparency to my research process. Of course, since they are only notes, they do not reflect my final thoughts on a topic, and should not be interpreted as such. To read finished papers, please consult my website. Do not use these notes as a basis for your own scientific research. Start from high-quality, peer-reviewed scientific literature instead.
Modelling pronominal gender agreement in Dutch
p. 219
p. 221
agent-based model
- reorganisation of gender in Dutch
p. 221
Gender in Dutch
(De Vogelaer re-iteration)
p. 224
Anaphoric reference game
language game
- yes
p. 225
Vocabulary
vocabulary
- 60 words, each with a gender
- also have values for tangibility, countability and animacy
p. 226
p. 227
Example construction for ‘hond’ in FCG
Learning operators
mapping mechanism
- allows agents to infer the gender of a word if they don’t know
- works through exemplars
↳ exemplars?
- each gender has a word with a very high score for gender
- that word serves as the basis for comparison (using feature vectors)
p. 228
The final step of the language game is the so-called alignment, performed by the hearer. The agent that plays the role of the speaker does not perform any adjustment on his internal model, as this would mean that half of the time the agent will boost his own preferences, making him less adaptive overall (Wellens 2012).
lateral inhibition
- widely adopted alignment method in the field of language games
- in case of a successful interaction, an agent…
- …decreases the scores for competing constructions
- …increases the score for the correct solution
- when a game fails, the wrongly chosen construction is decreased in score
A systematic overview of alternative updating functions is contained in Oliphant (1997).
Evaluation
Measures that will be used to evaluate the outcomes of the language model
communicative success
- determine if an interaction was successful or not
- can happen in two cases:
- the speaker and the hearer have the same preferred gender for a word
- the hearer has the speaker’s solution in its inventory, but he did not choose it as its preference
alignment success
- measures the degree to which the agents have the same preferences over the genders for the elements of the world
- alignment success is reached at the end of an interaction only if the agents have the same choice of gender for the uttered noun
cluster purity
(Manning et al. 2008, Chapter 16)
- a measurement of the quality of our final gender mapping
p. 228-229
We split our semantic space into 5 clusters (represented by all the feasible combinations of the features: animate-concrete-countable, inanimate-concrete-countable, inanimate-abstract-countable, inanimate-concrete-uncountable, inanimate-abstract-uncountable). For each cluster we assign a class, i.e. the gender that is most frequently encountered inside the cluster. The purity is then computed by dividing the number of words belonging to this majority class by the total number of words in the cluster:
TODO equation
p. 229
Results and discussion
Three types of experiments
- analyse the semantic systems that can form from our simulation when there is no pre-existing syntactic gender system in place
- should offer us an indication regarding how well our chosen strategies can perform in terms of identifying discriminating dimensions and clusters
- while starting from the syntactic gender system, we introduce a gradual population replacement
- under various substitution and frequency rates
- which conditions we still have a shift from the initial syntactic agreement system towards the semantic strategy?
- we explore the effect of various knowledge transmission rates on the stability and purity of the obtained clusters
For all our experiments we use a closed population of size 20.
Weighing semantic features
weighting
- used to make a certain feature more important
p. 240
population turnover rate
- percentage of population to be replaced
knowledge transmission rate
- percentage of gender knowledge to be transmitted between subsequent agents
turnover frequency
- population replacement interval in terms of number of interactions
p. 241
Population turnover
p. 244
Knowledge transmission
Computer simulations of language change notes
This website collects my personal notes on Computer simulations of language change. These notes are provided to bring full transparency to my research process. Of course, since they are only notes, they do not reflect my final thoughts on a topic, and should not be interpreted as such. To read finished papers, please consult my website. Do not use these notes as a basis for your own scientific research. Start from high-quality, peer-reviewed scientific literature instead.