Male Bengalese finches sing complex songs that are expressed by finite-state automata. Females of the same species prefer syntactically complex songs to simple, linear songs. We hypothesized that song complexity arouse from sexual selection, because complex song could be a handicap trait. Behavioral experiments supported this notion. A finite state syntax thus could evolve through sexual selection, independently from the semantics. Conventional theories of language evolution state that the symbolic competence could somehow elaborate to produce syntax, but how this could be done is not specified. Our Bengalese finch studies inspire that the form and content of human language could have developed independently.

Key Words: Birdsong, Generative Syntax, Finite State-Syntax, Hierarchical Structure, Evolution

In passerine birds, song is a learned behavior used by males to attract females (Catchpole & Slater, 1996). Generally, a song is composed of several types of song notes and the order of these notes is stereotyped. However, unlike most songbirds studied to date, Bengalese finches (Lonchura striata var. domestica) sing non-deterministic songs that can be described by a finite-state syntax.

Bengalese finches critically depend upon auditory feedback when singing (Okanoya & Yamaguchi, 1997). When deafened, temporal organizations of the song immediately deteriorate. This is another special aspect of the Bengalese finch song, but critical dependence to auditory feedback and song complexity may be closely related: Bengalese finches are probably listening their own songs in real time to produce immediate varieties.

Each male Bengalese finch sings a unique song with unique set of song elements and individual specific finite-state syntax. Two to five song elements are organized into a "chunk." Several chunks are further organized into phrases by passing through a finite-state syntax. Several phrases are, in turn, arranged based on the finite-state syntax. A Bengalese finch can produce infinite varieties by taking different paths on the finite-state automaton. Why do Bengalese finches sing such complex songs? We shall try to answer to this question from four perspectives as suggested by Tinbergen (1963): its mechanism, development, function, and evolution.

The forebrain song control system in songbirds consists of a set of discrete nuclei including NIf, HVc, and RA (Margoliash, 1997). We hypothesized that each of these nuclei might be responsible for the control of the hierarchical structure of the Bengalese finch song.

When NIf was bilaterally lesioned, the song lost phrase level variability: that is, the complex song changed into simple, stereotyped ordering of linear syntax (Hoshino & Okanoya, submitted). When HVc was partially lesioned, a particular state transition disappeared but all song elements were preserved (Uno & Okanoya, in preparation). Thus HVc is responsible for chunk level variability. When RA was partially lesioned, a certain song note dropped from the song but overall finite-state syntax remained unchanged (Hirata & Okanoya, in preparation). RA, thus, is responsible to produce each note. Taken together, we found that the finite-state syntax is expressed in hierarchically organized brain nuclei.

Next, to answer from the functional point of view, we examined reproductive behavior of females when stimulated with complex or simple song syntax. Number of nesting material carried each day when stimulated with complex or simple song syntax was counted and compared. Female finches carried more nesting materials and had higher levels of estrogen when stimulated with the complex song.

We thus showed that a song with complex syntax could effectively stimulate the reproductive system of females. From this result we postulate that a complex song patterning should be more attractive to female birds, and therefore the song syntax in Bengalese finches may have evolved through sexual selection (Darwin, 1871).

Why do females prefer song complexity? An ability to sing a complex song may be an honest signal that can advertise the potency of the singer (Zahavi and Zahavi, 1997). Since singing a complex song may require 1) higher level of testosterone, 2) more cognitive load, 3) and more brain spaces. We tested one of the above hypotheses by perturbating an ongoing song by a flashlight. Birds with lower song linearity (i.e., more complex song) were less prone to stop singing (Okanoya & Nakamura, in preparation). The data support that singing a complex song probably requires more cognitive load and make the animal less careful about potentially dangerous situations.

Bengalese finches are domesticated strain of the wild white-backed munia. White-backed munias were imported in Japan about 250 years ago and domesticated since then. Aviculturists selected white-backed munias based on their parental abilities, but never by their songs. Nevertheless, we found that the complex song syntax was absent in the ancestor species (Honda & Okanoya, 1999). The finite-state song syntax in Bengalese finches must have evolved during 250 yeas.

We hypothesized that the sensory-exploitation process (Ryan, et al., 1990) worked upon the evolution of complex song syntax in Bengalese finches. That is, females of the ancestor species had a preference for complex songs, although males of the ancestor species were not able to sing complex songs because of constraints in the nature including predation risk, foraging cost.

We tested this hypothesis again by examining reproductive behavior of the ancestor species. A simple, linear song of a white-backed munia male was spliced into song notes and a finite-state syntax from a Bengalese finch was applied on the song notes of the munias. We thus composed a new song that had the phonology of the white-backed munia but had the syntax of the Bengalese finch. We played either the original simple munia song, or the newly composed hybrid song. The string-carrying behavior was much more active when stimulated with the hybrid song.

Based on these results, we discuss the evolution of complex behavior and associated changes in the brain. Bengalese finches developed finite-state syntax presumably through the process of sexual selection. However, each token in the Bengalese finch song does not have any associated semantics, and the finite-state syntax in Bengalese finch song does not produce any meanings. Thus, a finite state syntax could evolve without meaning.

Most authors arguing the origin of language assume that proto-language that had symbolic contents but no syntactical structure evolved first, then the proto-language somehow evolved into true language with the syntax (i. e., Bickerton, 1995; Pinker, 1994). In this scheme, ability for state-dependent signaling could evolve into an ability to represent symbols through natural selection (i.e., Munn, 1986). Then, the ability to develop symbolic representation could evolve into the competence to produce a context free grammar through social elaboration. However, how the proto-language acquired the syntax is always the hardest question to answer when considering the origin of language (but see Calvin & Bickerton, 2000). Through the analyses of Bengalese finch songs, we reached to a hypothesis that a syntactical behavior could evolve without a need of semantics.

We propose that the symbolic aspect and the syntactical aspect of human language could have evolved independently. As in Bengalese finch songs, rudiment of syntax might be evolved thought sexual selection in humans; sexual display between males and females might be a pre-adaptation to the syntax.

This could happen if an animal that use courtship display with a serial pattern generator evolved into an animal that use a finite-state automaton for courtship display through sexual selection. To produce complex sequences of behavior, a random number generator would be more appropriate. However, the nervous system would rather develop a finite-state machine than a random number generator. Using the vehicle of sexual display as a rudiment of syntax, symbolized tokens were presumably arranged to form a primitive, but true language.

When the animal had already established a competence for the symbolic representations, an elaboration between the syntactical module and the symbolic module by social and sexual selection pressures would reach to a context free grammar (Pereira & Wright, 1998). This could happen by making an external indexing system to bootstrap the finite-state syntax.

Research supported by PRESTO, Japan Science and Technology Corporation.

 Conference site: http://www.infres.enst.fr/confs/evolang/