speeds, or analyzed by machine, turn out to be a series of pulses or rapidly sequenced, discrete tones. In such cases, we will call each discrete pulse or tone a "subunit.") A series of units is called a "phrase." An unbroken sequence of similar phrases is a "theme," and several distinct themes combine to form a "song." Finally, a series of songs within which there is no pause longer than 1 minute is termed a "song session." Some sessions last for hours.

In summary: subunit < unit < phrase theme song song session. A diagram is shown in Fig. 1.

The shortest complete humpback song we have yet timed lasts 7 minutes, and the longest more than 30 minutes. Since these two songs are quite different in form and were recorded in different years, they are probably from different whales. But even the same whale, repeating its own song, will show different cycle lengths during a song session. For example, two successive songs of whale III (Fig. 4) last 7 minutes and 121⁄2 minutes, respectively. In spite of such variations in length, we call these vocalizations "songs" because they differ primarily in the number of times the phrases of a given theme are repeated. In our sample, the sequence of themes is invariable, and no new themes are introduced or familiar ones dropped during a song session. Except for the precise configuration of some units and the number of phrases in a theme, there is relatively little variation in successive renditions of any individual humpback's song. Yet, although we will not deal with it in detail here, we must not overlook this variation, for it is obviously an important feature of the songs.

Besides the variation among successive songs of one individual, there are often large differences among songs sung by different individuals. This raises the question of whether there is a single, species-specific song pattern, or whether each humpback sings its own pattern. All songs of humpback whales that we have heard consist of the following three main sections: (i) trains of rapidly repeated pulses that often alternate with sustained tones; (ii) many short, high-frequency units, most of which abruptly rise in frequency; and (iii) lower, more sustained notes that are monotonously repetitious in rhythm and frequency and contain many units that fall in frequency. We feel that these three sections constitute

Fig. 6. Theme 1. In this figure, as in Figs. 7 to 9, the enlarged top view shows diagram. matically a typical example of the phrases of which the theme is composed. Beneath, at reduced scale, are tracings of all phrases in each of two songs of whales I, II, and III. No phrases or other sounds are omitted or changed in order, and no time is left out. Observe, by referring to the enlarged top view, the slow evolution of this phrase. Note that the short pulse train becomes longer and higher in frequency with each repetition. We have aligned each phrase so that each moment of onset of sound is brought out to the left edge of the column. The sequence of phrases within the theme reads from top to bottom. Note that in song 1 of whale III, theme 1 contains more than twice as many phrases as does theme 1 of song 2 of the same whale.

A clear example of song type A is shown in Fig. 2 (whale 1). This figure shows sound spectrograms, at a greatly reduced scale, for a complete pair of songs. The pair of songs was lifted from the middle of a continuous song session that contained at least seven songs (there may have been more, but Watlington's tape ran out). The frequency scale in all spectrograms shown here is logarithmic, since that is roughly the way in which the human ear interprets frequency. Had we wished to define precisely the frequency components of each unit, we might have favored a linear frequency display, but our chief purpose was to facilitate comparisons between spectrograms and the sounds they depict. The spectrograms in Figs. 2 to 5 were made by the exceedingly tedious process of extracting successive 9.6-second segments of tape-recorded sounds and analyzing them on a Kay spectrograph (model 6061B), being careful that no time on

the tapes was omitted. The effective filter bandwidth was 60 hertz, and the highest "real time" frequency seen by the spectrograph under these circumstances, 2000 hertz.

Once the spectrograms were made, the hundreds required for each song were carefully matched, glued to large sheets of paper, and photographically reduced. Because the ocean is a noisy place (some of the tapes include wave noise, ship noise, dynamite blasts, and distant whales), and because loud sounds in the water are invariably followed by trains of echoes, we have isolated the whale sounds by tracing them on a separate sheet, thus omitting echoes and noise. Our tracings also exclude harmonics, since they are of little or no consequence in establishing sequences and are often spuriously generated by the spectrograph (12). The only sounds for which we have traced harmonics are brief, pulsive sounds that, to a human listener, are atonal (even when slowed down) and must therefore have a rich harmonic structure. The tracings, of course, fail to show variations in darkness of line -the only indication of intensity that spectrograph machines provide. However, darkness of line, even on the spectrograms shown here, can only be

relied on to reveal relative intensities within any particular 9.6-second sample, since the level of input to the spectrograph machine was often adjusted between successive samples in order to obtain the least distorted rendition. However, a thick line indicates ocean reverberations, in some cases, thus denoting a loud sound.

From inspection of the spectrograms, it is immediately obvious that phrases in most themes are repeated several times before the whale moves on to the next theme. As mentioned above, we find it true of all song types in our sample that, although the number of phrases in a theme is not constant, the sequence of themes is. (For example, the ordering of themes is A,B,C,D,E . . . and not A,B,D,C,E

.). We have no samples in which a theme is not represented by at least one phrase in every song, although in rare cases a phrase may be uttered incompletely or in highly modified form (compare phrase 4 in the first and second songs of whale III, Fig. 8). Figures 6 to 9 compare themes in song type A as sung by what we assume to be different whales, since the songs were recorded at different times. In this analysis, all phrases within any given theme are shown for each of

two songs sung by each of three whales (I, II, and III). No sounds have been omitted, changed in order, or shifted in time (although we have cut our traced record into pieces, we present the pieces in their original sequence and have omitted none of them).

Before discussing song type A, we must make one final point. Some phrases in humpback songs are further complicated by being inexact replicas within a theme. That is, the phrases systematically change, or "evolve," with each successive repetition during a theme. The phrases in theme 1 of song A are a good example of this. The six themes of song type A are as follows.

Theme 1: Figure 6 shows that each phrase in this theme consists of three principal units-unit 1, a sound like a motor running [a pulse train in which the number of pulses (subunits) per second changes during the train], followed by units 2 and 3, a pair of sustained or wavering tones. (Sometimes each sustained tone is followed by a faint pulse, and sometimes the phrase ends with a very low grunt.) With each repetition of this phrase, unit 1 becomes longer, and, toward the end of the theme, each sustained tone tends to occur at a higher frequency than the previous one. (In song type B, the pulse trains are given without sustained tones between them or with just a brief tone tacked on to the end of the train of pulses-see whale IV.) Thus, in song type A,

theme 1 consists of a series of sequentially modified phrases changing in a predictable fashion.

Theme 2: We arbitrarily designate the onset of theme 2 as the moment when sustained tones give way to shorter tones, or rapidly ascending frequency sweeps. Theme 2 is one of the most variable in our sample. It may consist of a great variety of sounds, but all, or most, of them are ascending frequency sweeps or brief (less than 1 second) high-frequency squeaks or chirps. The theme has no obvious subsets that repeat predictably. Theme 2 is quicker in tempo and less rhythmic than the rest of the song, which is composed of more measured, drawn-out phrases.

Theme 3: The phrases in theme 3 have as their first unit a loud, ascending frequency sweep, which is emitted from one to three times and is followed by several fainter, high-frequency harmonic arpeggios. In some cases, the arpeggios become more elaborate with each repetition of this phrase, until, by the end of the theme, the component notes become more sustained. (See Fig. 10 for an expanded, scale view of arpeggios from the songs of whale II and whale IV.) The loud, ascending frequency sweep, marking the beginning of each phrase 3, usually occurs twice (only once in case of whale III) at the outset, but in the course of the theme changes to three repetitions and stays at three. When there is a transition from two to three,

it is as irreversible as the basic sequence of phrases in a song.

Theme 4: Theme 4 is one phrase long. It starts with a loud, ascending sweep that is similar to unit 1 in theme 3. Each successive unit in this theme rises and falls in frequency, but returns each time to a lower frequency than the preceding unit. (This gives the impression of a deepening series of roars or bellows.) The last units of theme 4 are low-frequency grunts (which may be descending pulse trains).

Theme 5: The phrases of this theme typically have six units. Unit 1 is a downward sweeping note; unit 2, a loud, low, rising or slightly warbling note (followed sometimes by one or more brief, descending, pulsive notes that are hard to detect in poor recordings). Units 3 and 4 are very like units 1 and 2. Units 5 and 6 are grunts. Since there are usually two such grunts, each phrase in theme 5 usually consists of six units. However, there may be three (rarely more) grunts, and sometimes one or even none. This whole phrase is usually repeated once; thus, theme 5 is usually two phrases long.

Theme 6: The components of this theme are unit 1, a loud, sustained tone; unit 2, a descending note (sometimes in the form of a slowly falling warble); unit 3, another sustained tone similar to unit 1; unit 4, a descending note similar to unit 2 (sometimes there are faint pulsive notes after units 2 or 4, or both); and a series of one to

four grunts (units 5, 6, and so on). This phrase is usually repeated quite exactly ten or more times. An interesting sound that is sometimes superimposed on the grunts may represent the audible component of a train of ultrasonic pulses, but this possibility must await recordings on equipment sensitive to ultrasound.

Song type B (as shown by whale IV) resembles type A in themes 1, 2, and 3, but by theme 4 the parallel is less clear. The series of descending

notes shown on line 4 of Fig. 5 is not comparable to anything in song type A, unless it could be thought of as a series of units similar to those in theme 4, but with the duration of the units, and the silences between them, being much greater. The subsequent very short and rapidly delivered notes are certainly unlike type A songs. At the end of its first and second songs, whale IV emits sustained notes that are followed by a low grunt. In a broad sense, this is quite similar to the basic

phrase of theme 6 in song type A. However, in type A the sustained notes differ markedly in pitch, whereas they are delivered at roughly constant pitch in song type B. In some ways, type B is intermediate between type A and the very different song type C. (We have various examples of type C from 1969, but they will not be discussed here.)

Since variability within a fixed general pattern is such a prominent feature of humpback songs, it may be worth taking a detailed look at some of the

WHALE III

ས

highest frequency units in the songthe arpeggios following the loud rising units in theme 3 (Fig. 10 shows such units from whales II and IV, song types A and B, respectively). The complexity of the arpeggios and their variation from song to song are obvious.

Beginnings of Songs

Following the last phrase of the final theme in either song type A or B, the whale starts the first sound in the next song (which we have labeled the first, and shortest, pulse train of theme 1) without any noticeable break in the rhythm of singing. The pause between any two phrases of the last theme is, if anything, longer than the pause between the last phrase of one song and the first phrase of the succeeding song. This fact should make it apparent that not only is our designation of phrase beginnings and ends arbitrary, but so is our choice of the start of a song. It is also possible that what we have designated a "song" may, in fact, be a long segment of a still longer song, but that possibility must await further analysis.

The criterion we have used here to determine a starting point is influenced by our large sample of recorded songs from the years 1969 and 1970. In these songs, the overwhelming majority of starts, stops, and blows occurs at a point corresponding roughly to the starting point we have chosen here. Watlington's tapes include very few records of whales beginning to sing and only slightly more of the ends of song sessions. (Usually the tape ran out, the noise level from passing ships rose to mask the whale sounds, or recording was stopped.) Thus the Watlington sample is really too small to warrant any conclusion on this point. It is clear, however, that, regardless of where a song may begin, the whale continues the sequence of themes in the same irreversible order (that is, 3, 4, 5, 6, 1, 2, 3, 4, 5...).

Fig. 10. Linear, wide-band spectrograms (upper frequency limit 2 kilohertz, effective filter bandwidth 200 hertz) show the complex and variable structure of the highest notes in phrase 3. Columns 1 and 2 are from songs 1 and 2 of whale II. Columns 3 and 4 are from songs 1 and 2 of whale IV. Read sequence of sounds from top to bottom. (The intermediate lower and louder frequency sweeps are omitted here.)