Introduction whose producing utterances from meaning space along

Introduction

Language is one of our defining characteristics
that separates us from non-human animals. So, how this special communication
system evolved and adapted by humans? Finding the origins of language and
explaining its evolution considered to be one of the hardest topics in science.
Over the past few decades, interest and research in this field accelerated by
the advancements of computer technologies and the vast amount of laboratory
experiments. However, the question “How
did language evolved?” is not yet answered completely and requires further examination.
When we scrutinize this question, we get two major interpretations; first the biological
evolution and second the cultural evolution. On one side, the
biological evolution relates to humans with their cognitive capacity and affinities
to its lineage, on the other side the cultural evolution discloses the language
itself and the mechanisms that pressures the change in its structure. This
paper investigates some of the structures that emerges through the cultural evolution
and shapes our language. I will try to justify these structures by giving
examples from both experimental and computational studies.

 

Cultural Transmission

For tracing the origins of these structural
features, we need to comprehend one of the key facets of the cultural evolution
that is the cultural transmission. Every language speaker whom once a language
learner back in early days of his life were exposed to the utterances that was
generated by other speakers (Kirby, Griffiths, Smith, 2014). Kirby (2001) proposed
the iterated learning model for
showing this transmission. He built an agent-based model that consists of
meaning and learning spaces accompanied with agents whose producing utterances
from meaning space along with the agents that induce those utterances and re-produce
them for the next generations. Later, this model studied
more with the experimental setups (Galantucci, 2005; Kirby, Cornish, &
Smith, 2008; Horner, Whiten, Flynn,
& de Waal, 2006) and it was fruitful for understanding it’s
underlying consequences.

 

Compositional
Structure

We as language speakers are capable of merging
meaningful words in structured ways to construct utterances whose meanings are
constituted from the same words. This structure is called compositionality.
Exploring and tracing the origins of this feature is a vital one for
understanding the evolution of language (Kirby, 2001). Kirby et al. (2008)
designed an experimental setup with adult human participants whose learnt a
generated artificial language composed from a set of 27 randomly generated
strings that to be transmitted to new learners. In their first experiment
without any pressure for unambiguity, participants easily transmitted strings
to next generations. Language was highly learnable but not with respect to
expressivity. Simplest kind of structure would emerge by this transmission and
later Kirby et al. (2015) named this as degenerate
because of its loss of meaning distinctions. In second experiment, by adding an
adaptive pressure for expressivity, experiment showed that compositional
structure arises and rules for combination of substrings passes through the next
generations. Language became both learnable and expressive. Thus, just like the
biological evolution, we can see emergence of an adaptive structure without an
intentional design.

 

Categorical Structure

Regardless of seeing these successful demonstrations,
an important question arises by these experiments. How realistic are these
assumptions when we compare them with natural languages? An important factor seems
to be leading us to this question that is the limits of the meaning spaces and
the imposition of the structure (naturally categorizable) which may be given by
the experimenter. In a recent study Kirby et al. (2017) investigated this
important question by extending the meaning spaces with open-endedness (was
only continuous before). In such a way, experimenters did not give any -naturally
categorizable- training stimuli to be tested. They started with basic
transmission experiment without applying any pressure and continued with two
other experiments including an artificial expressivity pressure to the
transmission chain. First one was the same as the Kirby et al. (2008) first experiment,
except this time, experiment done with 20 participants and 10 generations but crucially,
meaning space was also open-ended. Participants had to learn words that
assigned to the triangles applied by the previous generation and produce new
words for the novel ones. There was no prescribed structure given by the
experimenters because of the randomly generated triangles. As a result,
participants were arbitrarily categorizing the triangles. Without giving any a
priori information of the space and category, they observed the emergence of
categorical structure. With this structure language became more compressed and
easy to learn. Passing the words to the next generations was simplified. Obviously,
less amount of words was assigned to the triangles and loss in expressivity was
observed. Continuing the second and third experiments by including expressivity
pressure resulted just like the Kirby et al. (2008) second experiment in terms
of emergent compositional structure alongside with the categorical one.
Respectively, assumptions that made for the emergent structures and the results
were relatively realistic when we compare them to the previous studies.

 

Combinatorial Structure

Another unique design feature of language is combinatoriality.
Hockett (1960) identified this feature with ‘duality
of patterning’ along with the other twelve design features of language. He
argues that a combinatorial system becomes more beneficial when that language
has too many divergent non-combinatorial forms. In 2012 Tessa Verhoef
introduced an alternative hypothesis by designing a transmission chain. She
argued that combinatorial structure would emerge through the cultural
transmission with a tendency for simplicity.