Modeling cultural evolution. A parallel investigation of changes in bird song and human language

Results 2004

The project consists of several research strands, described in the project proposal that will become integrated as the project continues. Each of these main strands has proceeded according to plan in the first full year of the project, the numbering below matches that in section 12.2 of our project proposal:

1. How do multiple cultural traits coevolve together? And how do different patterns of transmission affect this? (Landsbergen & Lachlan)

A paper will be submitted for publication within the next month. Lachlan has trained Landsbergen (who as a linguist had no evolutionary or biological background) in evolutionary biology, and computational biology to the extent where Landsbergen is now capable of independent research. In doing so, they collaborated in producing the first research results of the project.
The paper deals with how the meanings of words changes over time in a population. As stated in the proposal, we have used a cultural evolutionary framework to model this process. The model borrows some structural concepts from earlier work by Lachlan on bird song. A word’s meaning is modeled as some region in an abstract semantic space, and we follow how the boundaries of this region evolve over time. Individuals learn their boundaries from failed communication attempts with others, creating a type of stabilizing selection. In the model, we demonstrate that a very realistic (compared with historical linguistic dogma) pattern of word-mean ing change can be generated with a very simple model, without any external causal motivation for change. Specifically, within any one population at one point in time, all individuals agree very closely on the meaning of a word, but over a realistic period of time (500 years), random changes cause the meaning to diverge considerably. A notable aspect of our results is that eve n when individuals have to construct their cultural “traits” out of individual instances that provide only limited clues, we still find reliable cultural transmission, and thus cultural evolution. We think this is an important result, because it addresses in a direct way one of the principle weaknesses attributed to cultural evolution by its critics in cultural anthropology.

The work also addresses more specific hypotheses about how different patterns of word usage affect linguistic evolution. We confirm that various aspects of usage: communication patterns, conventionality, frequency etc. have dramatic influences on the coherency of word meaning within a group, the rate at which meaning changes, and the breadth of the word’s meaning. Particularly relevant to our future investigations, we show how selection, which may be caused by competition between words, may cause directional change in word meaning.

We will follow this paper by examining a multi-word model, in which different words compete for semantic space. Landsbergen also plans to test the specific results of the paper in his empirical work (see below).

2, 3. How does a hierarchically organized set of grammatical rules evolve from a less structured predecessor? And may such rules ‘emerge’ from ‘usage based’ implicit transmission?

This project seeks to provide hypotheses to explain how apparent evolution in syntax, as demonstrated by the song of macronesian chaffinch populations can be explained. Our approach is to develop cultural evolutionary and gene-culture coevolutionary models.

Less progress has been made on this project. The principle behind the model is that birds apply some kind of “grammar induction” to learn aspects of syntax during development, while other aspects of syntax are encoded by genes. Lachlan has therefore made a survey of existing grammar induction techniques developed by computational linguists to search for simple, biologically plausible techniques. We have also engaged in collaboration with scientists from Edinburgh University who are following a related line of research, and have considerable strength in computational linguistics. The result of this is that we are implementing the “Minimum Description Length” theory of grammar induction: a technique in which grammars are learned based on how they minimize the amount of memory required to store them.

4a. Empirical measurements of vocal evolution in Macronesian chaffinch populations

This research project aims to tap linguistic concepts of syntax, and computational linguistic methods of grammar induction to explore how the complex syntax of chaffinch song is evolving in several reproductively isolated populations.
The first steps in this research project are to acquire enough recordings from the relevant populations to make generalizations about the shared and variant properties of the populations’ songs. We already have recordings of Canary Island populations made by Lachlan. In May 2004, Lachlan carried out a successful field trip to Madeira, recording the repertoires of 95 males (and around 500 different song types). Cataloging and preliminary analysis of these recordings is complete, and indicates that, as hoped, Madeira songs are intermediate in organization between Canary Island and mainland European songs.
Planning is nearly complete for the 2005 expedition to the Azores to complete the sampling of these populations.
Further analysis of these songs requires a more in-d epth analysis of song structure than has been carried out before in studies of bird song. To understand the rules of syntax, the sequencing of elements, we first have to be able to classify these elements into types. Using pre-existing techniques from bird song studies, this is a laborious task (given the large number of recordings), and relies on ad hoc, rather primitive methods of analysis. These problems are a central concern of studies of human speech, however. Therefore, Lachlan has developed a new computer program to facilitate and improve the analysis of field recordings of bird song. The program uses modern techniques borrowed from digital signal processing (edge identification and echo removal algorithms applied to spectrograms) to identify elements and from speech analysis (dynamic time warping) to categorize song elements. The program is placed within a database framework, which should greatly improve the archiving and standardization of such data . As of writing, the program is complete, and will be used this spring in earnest for our project.

4b.  Changes in grammatical constructions in Germanic languages

Landsbergen will carry out a linguistic analysis of grammatical language change. In order to do this, he has identified a suitable candidate grammatical construction to examine. He will collect corpus occurrences of the Dutch verb krijgen from the earliest sources in Early Middle Dutch (~A.D. 1200) to the present, and the German counterpart kriegen. The idea behind this empirical work is to analyze the changes that have taken place in the environment of the verb and to take those as a measure for its semantic change. This involves tagging of the types of subjects and adjuncts the verb is combined with, whether the verb is used transitively or intransitively, and whether the subject of the verb is agent or recipient. Currently, the methodology of the corpus analysis has been determined, and the work of the analysis is just beginning.
After collecting large amount of data, Landsbergen, together with Lachlan, will perform multivariate analysis, and subsequently construct an agent-based model of semantic ch ange that is capable of simulating this exact change (adapted from 1, above).

5. Interdisciplinair karakter

This projects is a collaboration between linguists and biologists. The two subprojects (on language change and birdsong, respectively) share a common basis in the modelling work. This is carried out in intensive collaboration between Lachlan and Landsbergen. They both meet at least once every week and Landsbergen received training on the modelling from Lachlan. They jointly constructed the first model to emerge out of the project. The 4 researchers involved meet on a monthly basis to discuss progress of the projects, next steps, etc.. On top of that Lachlan and ten Cate received an intensive training on linguistic theories from Verhagen and Landsbergen for several months in weekly meetings. All this has led to an intensive exchange of knowledge and ideas which all involved appreciate as very stimulating and beneficial to the project.