Our species, homo sapiens, is unique in many ways. We are a highly intelligent, extremely social species, possessing a singularly complex linguistic communication system and the ability to craft specialized tools allowing us to survive in almost any environment. We create music and art, build cities, seek to explain the world around us, and even show a willingness to sacrifice ourselves in the name of abstract ideas. Given our range of peculiar traits, it is important that any account of our evolutionary origins explain not only what we have in common with other species, but also what gave rise to our uniqueness.

One recent strand of evolutionary thinking provides such an explanation, it argues that in our lineage the emergence of cultural learning fundamentally altered the way evolution operates in our lineage. Rather than being solely the product of biological evolution, our recent evolution is the result of interactions between genes and culture, and can only be understood in terms of this combined process of gene-culture coevolution (Blackmore, 1999; Boyd and Richerson, 1985; Richerson and Boyd, 2006; Dawkins, 1976; Deacon, 1997; Durham, 1991).

While their are strong arguments supporting this theory, currently our poor understanding of the process of cultural change restricts its applications. Although a large number of theories of cultural evolution have been proposed, the majority depend on ad hoc, post factum reasoning that does not offer the sort of predictive capacity needed to derive coevolutionary explanations. What is needed is a theory cultural evolution which is, both, applicable in the general case, and which can make predict how culture will change in response to particular biological environments.

The potential foundation for such a theory can be found in Dawkins' (1976) concept of a meme; a replicating unit of cultural information. Several authors have attempted to build darwinian theories of cultural evolution based on this

idea (e.g. Blackmore 1999; Hull 1988; Brodie 1996; Gabora 1997; Lynch 1998 etc.), but they have generally failed to reach any sort of consensus on what such a theory should look like. But perhaps more significantly, their attempts have been met with widespread criticism for a variety of reasons and from a range of sources (e.g. Sperber 1996; Rose 1998; Blooch 2000).

In this thesis I argue that the main source of disagreement between existing theories, and also the cause of much of the criticism, stems from the overzealous application of biological ideas to memetic theories of cultural evolution. This occurs both on the part of proponents of these theories, and on the part of those seeking to criticize them. I identify several criticisms from the literature which I argue are examples of this over-application. I then show that in many cases, even though these criticism are biologically inspired, they seek to place constraints that even biological evolutionary theory cannot live up to. As evidence of this, I present examples of evolutionary processes in biology that would run afoul of the objections raised against memetics.

I then make the more significant contribution of showing that if we discard the single-stage biological model of replication in our definition of memes that much of the disagreement between authors in the memetic literature can be resolved. I do this by introducing a two stage life-cycle model of memetic replication, which links the brain-internal products of culture, i-memes, with the brain-external products, e-meme, as distinct stages in the meme replication process. This definition has the advantage of making the cultural germ line explicit, which prevents misleading claims of Lamarckian inheritance, and it also makes it possible to capture Blackmore's (1999) distinction between copy-the-product and copy-the-instructions memes which has been problematic in earlier models.

By separating the memetic lifecycle into two stages, the replication process is also separated into two distinct steps; production and learning. This is significant as it is often the case that memetic traits which make replication at one step easier, impede replication at the other. For example, variations that make the production of an e-meme more desirable, often make the associated i-meme more difficult to learn. When working with a traditional single stage model of meme replication, it is hard to separate these opposing effects, making the discussion of the adaptive benefit of particular cultural properties difficult. The two-stage model suffers from no such limitation.

Further, by allowing the quantification of replicative success at the two steps in isolation, the two-stage life-cycle model makes it possible to categorized meme replication strategies according to their average replication frequency at each step. This adds an extra dimension to the discussion of the life history strategies adopted by memes, and allows predications to be made as to the most successful type of memes in particular kinds of socio-cultural environments.

A final advantage of the two-stage model is that it provides a new perspective on the debate surrounding the relevance of a replicator/vehicle distinction to memetics. Originally proposed in biology to clarify arguments about the level at which selection operates, the distinction has been widely borrowed into memetic theories of cultural evolution, where it is used in several contradictory ways. From the perspective of the two stage life-cycle model I argue that regardless of whether such a distinction can be made with regards to culture, the current uses of the term 'vehicle' in the memetics literature completely miss the point that lead to the term's introduction in biology.

In relation to questions concerning the size of memes, and their replication fidelity, I address several common misconceptions, and argue that there is no theoretical requirement for memes to be defined as clearly delimited discrete entities as some seem to believe (Bloch, 2000). This is important as I also argue that no such discretization is actually possible, and that a key property of cultural information is the continuous degrees of linkage possible between concepts. Unlike in biology, linkage is not determined by the physical arrangement of replicators, but as a result of their specific properties. This is shown to have important affects on the ways in which memes can coevolve with each other, which in turn has important implication for the selection pressures they are subject to.

The arguments presented in so far are intended to put memetic theories of cultural evolution on a stronger theoretical foundation, not to provide a full theory of memetic evolution. It is thought that this stronger foundation should give us more confidence in the claims made by memetic theories of cultural evolution. First that cultures are not unitary entities but temporary aggregates, second that the components of these aggregates evolve purely to enhance their own replication, regardless of the effects this may have on each other, or us, their hosts. In the remainder of the I applies these and other memetic ideas to the study of the evolution of human language. In doing this, it is my aim to show that memetic theories are not just possible, but also have practical applications.

Before proceeding to practical applications, I first explain the core methodological difficulties involved in the study of the evolution of language; the limited availability of direct historical evidence and the necessity to understand the evolutionary interactions of biology and culture. It is argued that these difficulties are responsible for the heavy use of computational simulation by researchers working in this area. The perceived benefits and some of the potential pitfalls of this approach are discussed. A software framework designed to make modeling results more accessible to non-programmers is introduced, which will be used for modeling in all subsequent sections.

To demonstrate the way modeling is used in practice I give an in-depth analysis of one recent coevolutionary model from the language evolution literature, which was designed to investigate the coevolutionary emergence of linguistic learning biases (Yamauchi and Hashimoto, 2010). It is shown that several key behaviors reportedly displayed by this model are the result of arbitrary decisions made during its design, and are not the result of emergent properties in the underlying coevolutionary system being studied. This is shown to invalidate the central arguments made in the original paper, and serves as a stark illustration of the care that needs to be taken when utilizing computational models.

I then apply the memetic perspective to the evolutionary development of the modern human lexicon. It is argued that the lexicon is larger than is strictly biologically necessary, and that its size imposes significant biological costs. This has previously been explained as the result of sexual selection (Miller, 2000), but I suggest an alternative explanation in terms of an evolutionary arms race between our ancestors and their culturally evolving lexicons. To investigate this hypothesis I develop and analyze a computational simulation of lexicon evolution, which provides evidence as to the plausibility of the coevolutionary scenario I propose given certain easily justifiable assumptions.

Next I apply the memetic to the explanation of observed patterns of lexical diffusion between languages in show certain word classes are more frequently borrowed (Haugen, 1950). I argue that the Complexity Hypothesis (Jain et al., 1999) from molecular genetics, which explains an analogous pattern of horizontal transfer in biology, can be generalized to apply to borrowing between any coadapted darwinian system, including languages. I argue that this generalized complexity hypothesis can explain the observed patterns of borrowing, and may also play a role in determining rates of language internal change. I construct a computational model of populations of lexical learners in which coadaptation affects learnability and demonstrate that in this model the complexity hypothesis produces the predicted effects.

Having spent the last few years studying language from a darwinian perspective I have come to believe that what Dobzhansky (1973) said about biology, applies equally to linguistics:

"Nothing in linguistics makes sense except in the light of evolution."

It has been my goal in writing this thesis to add to the case in support of this statement. If I have been successful, I will have convinced readers of both the possibility, and the importance, of developing a darwinian science of language.

Dawkinsによって提唱された文化進化のミーム理論(memetic theory)は、以後、多くに研究者の手により発展したが、その基本概念においてすら統一性に欠け、数多くの批判を浴び続けてきた。本論文は、二段階の複製によるライフサイクルのモデルを提案し、ミーム理論における多くの問題を解消することに成功した。また、言語進化の計算論的なモデリングによって、ミーム理論が文化進化の理解に役立つことを示した。

本論文は9章から成り立っている。第1章は進化論に関する解説であり、Darwin以前、Darwin自身、そしてDarwin以後の進化論が著者の視点からまとめられている。第2章では、本論文の主題である文化進化に関して紹介されている。一般化されたDarwin進化論が述べられた後、文化進化の理論は、単に生命進化とのアナロジーに触発された結果ではなく、文化と生命がどちらもDarwin的な進化システムの具体例であることに依拠しており、したがって、文化進化と生命進化は類似しているが相違点もあることが、Campbellの規則として述べられている。第2章では、その後、文化進化のミーム理論とその問題が詳細に分析されている。特に、ミームはどこにあるのか、ミームの表現型とは何か、Lamarck的な進化の可能性、などの問題が分析されている。(第一部)

次の第3章においては、本論文の以降の部分の構成について述べられている。すなわち、第4章および第5章(第二部)において著者のミーム理論が展開されている。第6章から第8章まで(第三部)では、言語進化にミーム理論が適用され、具体的には、語彙の巨大化の問題が分析されている。本論文の二つの部分(第二部と第三部)には直接的な関係はないが、第二部がミーム理論から導かれる予想に高い信頼性を与えたことにより、第三部のもとになる仮設を形成することが可能となり、第三部の議論が正当なものとなった。逆に、第三部の語彙の巨大化の議論は、第二部におけるミーム理論の哲学的な探求の動機となっており、ミーム理論が間違っていないだけでなく、役に立つことを示唆している。

続く第4章では、上述したように、二段階の複製によるライフサイクルのモデルが提案されている。脳の中のミーム(i-meme)と外界のミーム(e-meme)が導入され、これらが互いを複製するという進化のモデルが提案され、このモデルにより、第2章で述べられたミーム理論の多くの問題が解消されることが示される。そして、第5章では、このモデルを用いて、文化進化の各種の問題が議論されている。

第6章では、以後の章で用いられる計算論的なモデルとエンド・ユーザに易しいモデリングのフレームワークが紹介されている。第7章では、このモデルとフレームワークを用いて、言語進化に関する既存研究の分析が詳細に行われている。これらは次の第8章の導入という位置づけである。そして第8章では、語彙の巨大化の問題がこのモデルとフレームワークを用いて分析され、二種類の利己性(selfishness)が同定されている。一つは語彙を増そうとする文化的な利己性であり、もう一つは種を保存させようとする生物的な利己性であり、これらが軍備競争(arms race)を演じることにより、全体として非最適な結果に至ることが示されている。すなわち、軍備競争により、生物的な適応性(fitness)は低下し、語彙は巨大化するものの語彙中の同意語が増加する。

第9章では、第8章までの成果がまとめられるとともに、将来の研究の可能性が示唆されている。特に、第二部のミーム理論を直接的に適用することによる言語進化の研究の可能性について言及されている。

結論として、本論文は、二段階の複製に基づく新たなミーム理論によって、将来の言語進化研究の方向付けを与えた。よって本論文は博士(情報理工学)の学位請求論文として合格と認められる。