Baker • SILVF.REYES SONG EVOLUTION 
457 
eyes' songs. I found there are some stereotyped 
syllables repeated now and then in an ongoing 
stream of songs from an individual, and syllable 
sequences recur occasionally, but the syllables did 
not constitute a finite repertoire expressed by a 
given bird. Extreme complexity of song struc¬ 
tures. including hundreds of syllable types in the 
songs of an individual, and individual song 
repertoires of hundreds or thousands (some likely 
unbounded), have been described in songbird 
species (Wiidenthall 1965, Kroodsma and Parker 
1977, Kroodsma 1980, Kroodsma et al. 1999, 
Huntsman and Ritehison 2002, Price and Yuan 
2011). This vocal variation can be limiting or too 
often effectively preclude the study of geographic 
variation in specific birds. The Eastern Bluebird 
(Sialiu sialis ), an example of these kinds of 
songbirds (Huntsman and Ritehison 2002), dem¬ 
onstrated the more songs recorded, the more song 
types are found, even after assaying many 
hundreds of songs of an individual (i.e.. plotting 
the cumulative number of song types identified 
versus the number of songs uttered did not reach 
an asymptote). 
My unsuccessful attempt to categorize the large 
variety of song elements of the Zosterops 
recorded into a relatively fixed repertoire of 
syllables, and my lack of confidence in the 
typological classification approach, led me to 
consider an alternative method of analysis. This 
alternative is based upon the use of linear 
discriminant analysis to create a way to compare 
the song syllables of individual birds and of 
population samples of birds' songs. I used a 
syllable-level analysis for two reasons. First, I 
assumed my results would be most comparable to 
the only other detailed examination of Zosterops 
songs (Slater 1993), which used the syllable in 
analyses, and second because previous studies of 
several songbird species (although not Zosterops) 
had demonstrated that syllables w'ere the units of 
transmission during the song learning process 
(Marler and Peters 1977, 1988; Kroodsma and 
Picker! 1984). 
Statistical Analyses .—Measurements of the 
acoustic features of song syllables were examined 
by linear discriminant analysis (LDA) using 
STAT1STICA software (Version 5.1. StatSoft 
Inc. 1995) or MINITAB (Version 13. Minitab 
Inc. 2000), Post-hoc assignment probabilities 
from LDA were obtained by jack-knifed cross 
validation. Correlation matrices lor LDA were 
judged for sampling adequacy with the Kaiser- 
Meyer-Olkin (KMO) test, and their ‘condition’ 
was examined with Bartlett’s Sphericity Test 
using PARANAL (Budaev 1997). Basic statistics 
(means, etc.) were calculated with MINITAB. 
RESULTS 
I recorded 10 birds in the Esperance (mainland) 
population and obtained 69 good quality songs 
(mean = 6.9 songs/bi id) and on Woody Island, 19 
birds and 102 songs (mean = 5.4 songs/bird). The 
acoustic features of the constituent song syllables 
0/ 1.728: i.e., those not overlapped) w'ere 
measured with Sound Analysis Pro. All songs 
were prepared as sound spectrograms and exam¬ 
ined visually. Initially, at least, this visual 
appraisal began with the goal of sorting the 
constituent syllables into categories (types) for a 
given population. The potential use of such a 
lexicon of syllable types would allow calculation 
of sharing among birds in a population and. for 
my primary use, to compare song characteristics 
between island and mainland samples. This 
approach has often been used and works well 
for many songbird species in which individual 
vocabularies arc of small or modest size and 
consist of stereotyped syllables. 1 found it to be 
impossible to apply in my song samples, as 
indicated in my initial analysis. 
I selected the 13 songs recorded from a single 
Woody Island bird for the initial analysis by 
visual classification of song syllables and exam¬ 
ined the constituent syllables sequentially. This 
bird was selected because it was recorded with 
none of its syllables overlapped by another singer, 
and it had the longest sequence of such songs in 
the samples. The first four of the 13 songs of this 
bird are illustrated (Fig. I). indicating by numbers 
the 55 different kinds of syllables I categorized by 
visual examination in the total 102 syllables, and 
the cases I judged to be recurrence of certain of the 
syllable types. Many syllabic types appeared to be 
discretely different but. in some cases, shapes alone 
were misleading. Syllable types 10, 12, and 34 for 
example, were grouped in my first pass through the 
songs, but closer examination and subsequent 
measurements revealed they differed consistently 
on the frequency scale (decreasing -400-500 Hz 
from type 10 to 12. and u similar decrease from 
type 12 to 34). Similar problems arose in different 
renditions of syllable 47, which covered a range of 
frequencies, or the small but apparent differences 
between syllables 35 and 53. Completing the 
syllable-by-syllable examination of the 13 songs 
