2.1. Tagalog and Filipino

The terms ‘Tagalog language’ and ‘Filipino language’ are often used interchangeably. When a distinction is intended, ‘Tagalog’ generally refers to the language of the Tagalog ethnic group in the northern Philippines, whereas ‘Filipino’ is the national language of the Philippines, based on Tagalog and enriched with vocabulary from other Philippine languages, English, Spanish and elsewhere. ‘Filipino’ tends to refer to the language as spoken outside the Tagalog region, or in Philippine cities where Tagalog and non-Tagalog people interact; it can also refer to a prestige variety of the language used in national media (Nolasco Reference Nolasco2007). Filipino, especially as used outside of the Tagalog region, has grammatical differences from Tagalog (Rubrico Reference Rubrico2012; Demeterio & Dreisbach Reference Demeterio and Dreisbach2017).

We use the term ‘Filipino’ in this article, because the music we are analysing forms part of the Philippine national mass media; the linguistic sources we cite use the term ‘Tagalog’.

2.2. Word prosody in Filipino

2.2.1. Two types of word

The great majority of words in Filipino fall into two classes: penult-prominent and ultima-prominent. Table 1 lists the phonetic properties that have been observed in Gonzalez (Reference Gonzalez1970), Anderson (Reference Anderson2006) and Klimenko et al. (Reference Klimenko, Maria and Javier2010);Footnote ² examples are in (1). In the song corpus, there are about the same number of words of each kind (by type: 269 penult-prominent, 266 ultima-prominent; by token: 498 penult-prominent, 475 ultima-prominent). To avoid committing prematurely to a phonological analysis, we avoid the IPA stress notation (International Phonetic Association 1999) and instead place an acute accent over the prominent vowel.

Table 1 Two types of word in Filipino.

• 

While lexical words are generally at least disyllabic, there are several monosyllabic function words, for example, [ʔat] ‘and’, [ba] ‘(question particle)’, [din] ‘also’. There are loanwords with antepenultimate prominence, such as [ʔágila] ‘eagle’, from Spanish águila, which was the only such word in the song corpus we used. And there are loans with prominence on a closed penult, such as [bɾiljánte] ‘diamond’, from Spanish brillante.

The spectrograms in Figure 1, made with Praat (Boersma & Weenink Reference Boersma and Weenink2017) from recordings in the online dictionary of Tagalog.com, illustrate a minimal pair in citation form.Footnote ³ The top example is [ʔábot] ‘power, capacity’, with penult and ultima vowels similarly long, penult louder than ultima (as can be seen by comparing the heights of the waveforms for [a] and [o]) and high pitch on the penult followed by a pitch fall between the two vowels, as shown by the pitch track (thick black trace overlaid on the spectrogram). The bottom example is [ʔabót] ‘arrival’, with a short penult vowel and long ultima vowel, penult and ultima similarly loud and a pitch fall late in the ultima.

Figure 1 Citation-form disyllables: penult-prominent [ʔábot] ‘power, capacity’ (top) and ultima-prominent [ʔabót] ‘arrival’ (bottom).

Further evidence that final prominence is not associated with longer duration (prominent ultimas are not longer than non-prominent ultimas) comes from Reed’s (Reference Reed2022) acoustic study of reduplication. Tagalog has semi-productive copying of a stem’s first two syllables, as in ka-sáma ‘companion’ vs. ka-sáma-sáma ‘constant companion’. Impressionistically, both copies often have word-level prominence. Reed finds that when penult-prominent roots undergo copying of their first two syllables, the copied penult is also long, as in d[aː]mi-dámi ‘quite a lot’, from dámi ‘quantity’. But when ultima-prominent roots are copied, the copy of the ultima is not long: an[o]-anó ‘what-pl’, from anó ‘what’. Reed takes this difference as evidence for the underlying-length analysis, but it is also consistent with final stress not causing additional lengthening beyond word-final lengthening.

The facts for multi-word utterances are less clear. Schachter & Otanes (Reference Schachter and Otanes1972) describe most Tagalog intonation patterns as including a phrase-final pitch-accent; for some patterns, this accent’s location depends on vowel length. The behaviour of these ‘fixed-P2 patterns’ (Schachter & Otanes Reference Schachter and Otanes1972: 32) is illustrated in (2): for penult-prominent words, the pitch accent can optionally move to the phrase-final syllable when an enclitic is added; for ultima-prominent words, the pitch accent must move to the phrase-final syllable.Footnote ⁴

• 

The optionality for penult-prominent words, as well as the variety of intonation types and their overlapping semantics or usage, makes it difficult to say whether any given utterance falsifies these claims.

2.2.2. Previous analyses

Previous authors have fallen into two broad camps, summarised in Table 2.Footnote ⁵ For some (Constantino Reference Constantino1965; Schachter & Otanes Reference Schachter and Otanes1972; Soberano Reference Soberano1980; Himmelmann & Kaufman Reference Himmelmann, Kaufman, Gussenhoven and Chen2020), there is a phonemic length contrast in open penults. This explains the phonetic duration facts: long vowels are pronounced with greater duration, as are all final syllables (leaving the [ʔa] of [ʔabót] short). This analysis must stipulate that long vowels attract pitch accent away from its default phrase-final location, and that a syllable following the pitch accent has lower amplitude. For these authors, stress, if it exists at all in the language, is the surface result of prominence due to vowel length or intonational pitch-accent. (Constantino Reference Constantino1965 uses underlying stress to encode exceptions such as /bɾilˈjante/ → [bɾil.ján.te] ‘diamond’, which has prominence on a closed penult.)

Table 2 Approaches to word prominence in Filipino.

For other authors (Bloomfield Reference Bloomfield1917; Blake Reference Blake1925; Ramos Reference Ramos1981; French Reference French1988, Reference French1991; Avery & Lamontagne Reference Avery and Lamontagne1995; Sabbagh Reference Sabbagh2014; Richards Reference Richards2017), there is a phonemic stress contrast. This analysis must stipulate that both stressed syllables and word-final syllables are lengthened, but not additively, so that stressed, word-final syllables are not extra-long. Like the vowel-length analysis, the stress analysis stipulates that a syllable following the pitch accent has lower amplitude.

We now review three phonological phenomena relevant to length or stress, concluding that they can be analysed under either approach.Footnote ⁶ Thus, new data are needed in order to distinguish between the two approaches.

The first phenomenon is that closed penults cannot bear prominence (with exceptions, mainly in loans): *hágkan is not allowed. Under the length analysis, this is straightforward: as in many languages, a syllable can have either a long vowel or a final consonant, but not both, ruling out *[haːg.kan]. A stress analysis can stipulate that stress must fall on one of the last two moras of the word (with final consonants extrametrical – that is, not bearing a mora). As shown in (3), stress (the bottom-most × mark on the grids) cannot fall on a consonant mora (ruling out (3d) *[haǵ.kan]), nor can it fall on a mora that is not one of the last two (ruling out (3e) *[hág.kan]).Footnote ⁷ This is reminiscent of many trochaic languages’ ban on words that end with a heavy penult and a light ultima (see Hayes Reference Hayes1995 on trochaic shortening; Zuraw Reference Zuraw2018).

• 

The second phenomenon is prominence shift in verbs. As shown in (4), prominence shifts one syllable to the right when a suffix is added. Exceptional loan verbs with prominence on a closed penult retain that prominence and gain another on the final syllable.

• 

Under a length analysis, ultima-prominent words require no explanation; they have no long vowel underlyingly, and continue to have no long vowel after suffixation. For penult-prominent words, some form of prosodic faithfulness can be invoked, whereby length moves in order to remain penultimate; see Shryock (Reference Shryock1993) and Crosswhite (Reference Crosswhite1998) for analyses of similar phenomena in other languages. Prosodic faithfulness has to be overridden by some mechanism that anchors exceptional vowel length in a closed syllable, as in [kweːntuhan]; the final syllable is far enough from the long vowel to get phrase-final prominence.

Under a stress analysis, a similar prosodic faithfulness mechanism is needed to keep stress penultimate when a penult-prominent root is suffixed, and the same mechanism applies to ultima-prominent roots, keeping stress final (see French Reference French1988; Sabbagh Reference Sabbagh2004). For the exceptional words, again some mechanism keeps the exceptional stress in place, and an additional stress is added to the final syllable to avoid stress lapse.

The third phenomenon is prominence shift in some suffixed nouns. Depending on the morphology, several patterns are possible (see Schachter & Otanes Reference Schachter and Otanes1972: 98–102). Both penult-prominent and ultima-prominent words can, when suffixed, become either penult-prominent or ultima-prominent, with or without lengthening of the root-initial vowel, and possibly other vowels. A sampling is given in (5). Under any account, there must be morpheme-specific prosodic requirements and optionality. A length account requires length to be deleted, moved, or added; a stress account requires stress to be moved or not moved and pre-tonic length to sometimes be added (see Sabbagh Reference Sabbagh2004 and Hagberg Reference Hagberg2006 for stress-based accounts of part of the pattern).

• 

While these length-based and stressed-based accounts have their strengths and weaknesses, both are workable, and these primary phonological data are thus not decisive.

2.3. Original Pilipino Music (OPM)

OPM stands for Original Pilipino Music. The term can refer to any Philippine pop music, but there is a stylistic core of music that is most likely to be labelled OPM. Arceo-Dumlao (Reference Arceo-Dumlao2017) is a rich collection of interviews with OPM songwriters, focusing on topics such as how an individual got into the music business, the inspiration for a song, or how a singer and songwriter met. There is little discussion of songwriting mechanics such as word choice, but the interviews do provide insight into the songwriting process. Some famous songs were written, music and lyrics, by one person in one day. In other cases, lyrics were written for a melody that had already been composed by someone else. And in yet other cases, the lyrics were written first and then another songwriter composed the melody.

Other literature on OPM includes engineering projects to classify songs into sub-genres, identify mood, distinguish OPM from K-pop, recommend songs automatically or predict hits (Deja et al. Reference Deja, Blanquera, Carabeo, Copiaco, Nishizaki, Numao, Caro and Suarez2016; Mital et al. Reference Mital, Tobias, Bandala, Billones, Dadios, Niranjan, Rana and Khurana2019; Abisado et al. Reference Abisado, Yongson and Los Trinos2021; Sulit Reference Sulit2022; Monterola et al. Reference Monterola, Abundo, Tugaff and Venturina2009); humanistic studies of the history, culture and politics of OPM (Lockard Reference Lockard1996; Maceda Reference Maceda2007; Gabrillo Reference Gabrillo2018; Domingo Reference Domingo2021; Peña Reference Peña2021; Cayabyab Reference Cayabyab, Johan and Santaella2021; Prudente Reference Prudente2021; Shunwei & Jia Reference Shunwei and Jia2022; Nagai Reference Nagai, Hamza, Chan and Chin2022; Gaillard Reference Gaillard2022); a study of cover performances on social media (Anacin et al. Reference Anacin, Baker and Bennett2021); and social-science studies of music preference (Boer et al. Reference Boer, Fischer, Atilano, Hernández, Garcia, Mendoza, Gouveia, Lam and Lo2013; Limjuco et al. Reference Limjuco, Ticudo and Pregua2014) and of attitudes towards code-switching in OPM (Bareng Reference Bareng2019). We have identified two linguistic studies of OPM. Alegado et al. (Reference Alegado, Labaya, Lirio and Rivera2021) analyses instances of English in OPM according to their length and syntax (the corpus we examine here did not happen to contain any code-switches into English). Sumalinog et al. (Reference Sumalinog, Salid, Sarino and Amante2021) discuss OPM lyrics’ use of Swardspeak, ‘the vernacular language or code used by Filipino gay men in the Philippines and in the diaspora’ (Manalansan Reference Manalansan2003: 46). Strikingly, more than half of the literature we identified was published since 2020.

We failed to find linguistic analyses of text-setting – the relationship between text and notes – in music of any Philippine language. The closest publication we could find was Anderson (Reference Anderson2015), a guide to performing Tagalog Kundiman songs, which notes several instances where it is musically effective for duration, beat strength and/or pitch to correlate with stress.

3.1. Song corpus

We found sheet music online for 19 usable songs. Nine songs were purchased from the composer Aldy Santos’s Web site, Aldy Sheet Music (aldysheetmusic.com), and ten songs from MuseScore (musescore.com), a site that allows users to upload their own transcriptions. (We found but excluded another seven songs: one was in 3/4 time; one was translated from Visayan, and song translators are working under different constraints; and five were each in a markedly different style from the core 19 songs.) The Appendix Table A1 lists the songs.

All songs were in the 4/4 time signature, which means that each measure has four beats. In this time signature, each beat is a quarter note (or crotchet). For those unfamiliar with musical notation, what is important about this time signature is that one can count along to the music in a repeating pattern of 1–2–3–4, 1–2–3–4, 1–2–3–4, ….

3.2. Coding

After listening to recordings and correcting the sheet music where necessary (this was rare), we hand-converted each piece of sheet music into a spreadsheet. In Figure 2, we show a fragment from ‘Akin ka na lang’ by Francis Salazar, as performed by Morisette (with accent marks added). The opening words are Bákit hindíɁ mo maramdamán… (‘Why don’t you feel…’). Each column in the spreadsheet stands for one sixteenth-note of duration, with sixteen columns per ‘measure’ of music. The rows include a repeating metrical grid (the rows with × marks; Liberman Reference Liberman1975; Lerdahl & Jackendoff Reference Lerdahl and Jackendoff1981, [1983] Reference Lerdahl and Jackendoff1996), to guide us to the correct cell for data entry, and rows to enter information about each syllable.

Figure 2 Transcription fragment, from ‘Akin ka na lang’.

As mentioned above, each measure is counted as 1–2–3–4; these numbers appear on the ‘beat’ row of the spreadsheet. We assume, following the usual convention for music in the 4/4 time signature, that the strongest position in the measure is the 1, or downbeat; we show this by giving the downbeats the tallest columns in the metrical grid, with five × marks. The second-strongest position is the 3, which we give four × marks. The next-strongest positions are the 2 and the 4, with three × marks each. If a musician wants to count to the music more finely, dividing the measure into eight equal parts, they can count 1–and–2–and–3–and–4–and, with an and falling in the middle of each beat. These ands are numbered 1.5, 2.5, 3.5 and 4.5 in the ‘beat’ row, and are the next-strongest positions, with two × marks each. A musician can count even more finely, dividing the measure into 16 equal parts, often as 1–ee–and–a–2–ee–and–a–3–ee–and–a–4–ee–and–a. These ees and as, numbered 1.25, 1.75, 2.25, 2.75, etc., are the weakest positions, with one × mark each.

The other rows encode linguistic information. Ba, for example, is entered in the ‘text’ row, in the column where it begins (the downbeat, identified as 1 in the ‘beat’ row). Because Ba is set to an eighth note, it extends over two columns; the underscore in the next cell indicates continuation. The 1 in the ‘stress’ row indicates that Ba is prominent.Footnote ⁸ The L in the ‘length’ row indicates that this syllable has a long vowel (in this case, predictable from being a stressed open penult); closed syllables are coded as C, and open syllables with short vowels as S. The ‘syll_position’ row shows 2, indicating that Ba is a penult (second from the end of the word). The ‘line_number’ row shows that all the syllables depicted here are in the first line of the song. The line is not a repeat, and we made no special notes (so the ‘repeat’ and ‘notes’ rows are blank). Filipino spelling does not indicate prominence, pre-tonic length, or word-final glottal stops; we relied on a combination of dictionary entries and one author’s native-speaker knowledge for the ‘stress’ and ‘length’ rows, and for the Qs indicating word-final glottal stops in the ‘text’ row.

An R script (R Core Team 2021) reads and processes these spreadsheets.

3.3. Exclusions

We excluded 223 syllables because they belonged to a word that fell entirely or in a part on a triplet (a division of a note into three equal parts instead of two or four), because we had no principled way to classify the prominence of the second and third sub-beats of a triplet. All the songs we coded had four beats per measure, but some had short passages with a different number of beats per measure, and we excluded 11 syllables because they belonged to words that fell partly or wholly within such passages. We excluded repeated lines, so that our data set would not appear (in plots and statistical analyses) to be bigger and more consistent than it really is. Finally, we excluded the one word in the corpus with antepenultimate stress, ágila ‘eagle’, a Spanish loan.

3.4. Predictions

We are comparing four syllable types: prominent and non-prominent penults and ultimas. Our absolute null hypothesis is that all four types are assigned to notes of similar length, and beats of similar strength.

The next closest to a null hypothesis is that OPM text-setting purely reflects the phonetics of word prosody, and tells us nothing about its phonology. In that case, non-prominent penults should be assigned to shorter notes than all the rest, as illustrated by the arrows on the left in (6), which start from the syllable type predicted to be set to longer notes, and point to the syllable type predicted to be set to shorter notes. The musical equivalent of loudness is less direct, but there is a tendency for loudness to signal beat strength (Lerdahl & Jackendoff [1983] Reference Lerdahl and Jackendoff1996: 17–18, 78–79). Non-prominent ultimas should then be assigned to weaker beats than all the rest, as illustrated on the right in (6), but with dashed arrows to show that the predictions are less direct, starting from the syllable type predicted to be set to stronger beats, and pointing to the syllable type predicted to be set to weaker beats.

There are two non-null hypotheses: that generated by the vowel-length theory, and that generated by the stress theory. As illustrated in (7), the vowel-length theory predicts that prominent penults, which contain a phonemically long vowel, should be set to longer notes than all other syllable types. Versions of the vowel-length theory that assign predictable stress to those phonemically long vowels also predict that prominent penults should be assigned to stronger beats than the rest (since cross-linguistically, stressed syllables tend to fall on strong beats, as stated in §1). Dashed lines are again used for the beat-strength predictions, to reflect that they are made by only one version of the vowel-length theory.

The hypothesis generated by the stress theory is that prominent syllables, because they are stressed, should be assigned to stronger beats than non-prominent syllables are, as shown in (8). There is less research on note length in uncontroversial stress languages, with some evidence that German stressed syllables are set to longer notes (Girardi & Plag Reference Girardi and Plag2022). The arrows for note length are dashed to show that this prediction is less clear.

4.1. Quantifying duration

Duration was quantified in quarter-note beats: a whole note (or semibreve, a note that lasts one measure) has a duration of 4 beats, a half note (minim) has a duration of 2, a quarter note (crotchet) 1, an eighth note (quaver) 0.5 and a sixteenth note (semiquaver) 0.25.

4.2. Duration results

There were 498 penult-prominent words and 475 ultima-prominent words analysed.

4.2.1. All final two syllables

The bean plots in Figure 3, made in R using the beanplot package (Kampstra Reference Kampstra2008), show results for the final two syllables of all usable words. On the left are the penult-prominent words, like ʔábot, and on the right are the ultima-prominent words, like ʔabót. The left side of each pair, coloured orange, represents the smoothed distribution of duration for the penult in each type of word, and the right side, coloured sky blue, represents the distribution of duration for the ultima. The four horizontal line segments show the mean of each distribution.Footnote ⁹

Figure 3 Duration of final two syllables of all words.

In the penult-prominent words, on the left, the penult tends to have a shorter duration (0.7 beats on average) than the ultima (1.0 beats). This might seem unexpected, but recall that in speech, final syllables tend to be long regardless of stress. In ultima-prominent words, the gap is bigger: penults have an average duration of 0.4 beats and ultimas 1.5. Overall then, ultimas are long, but less so in penult-prominent words.

Another way of looking at this plot is to compare the two orange distributions to each other: penults are slightly longer when prominent (0.7 on the left > 0.4 on the right). And comparing the two sky-blue distributions to each other, ultimas are longer when prominent (1.5 on the right > 1.0 on the left).

The plot in Figure 4 shows that syllable shape has little consistent effect. Whether a penult-prominent word has a closed or an open penult, ultimas tend to be slightly longer than penults. And regardless of the syllable shapes in an ultima-prominent word, the ultima is substantially longer than the penult.

Figure 4 Duration, broken down by syllable shape.

4.2.2. Separating out phrase- and line-final words

The very long durations seen for some ultimas are mostly line-final syllables, reflecting the musical tendency to place a long note at the end of a line. Splitting up the results into line-final versus line-medial words will allow us to see whether prominent ultimas are still long even when not line-final.

Furthermore, given various authors’ observations about how intonational prominence can track the end of the phrase, rather than the end of the word, we should further split line-medial words into phrase-medial and phrase-final, to see whether prominent ultimas are still long even when phrase-medial. (We assume that line-final words are always phrase-final.)

To determine phrase boundaries, we follow Schachter & Otanes’s (Reference Schachter and Otanes1972: 36) description of where optional pauses may occur; we take these optional pause locations to represent phrase boundaries. Schachter & Otanes state that phrase boundaries never occur after a proclitic or before an enclitic, and are more likely to occur at large syntactic breaks than at small ones, such as between a modifier and the word it modifies. They give the following examples:

• 

We operationalised phrasehood by having a look-up list of enclitics, based mainly on Kaufman (Reference Kaufman2010).Footnote ¹⁰ If a word was followed by one of these enclitics, it was considered phrase-medial.Footnote ¹¹ If not, then it was considered phrase-final. In the examples above, our procedure correctly codes all the boldface words, but leaves bagong miscoded as phrase-final. (Monosyllabic words are ignored here because they are excluded from our results.) Following this example and Schachter & Otanes’s description, we decided that two-word modifier–modified pairs connected by the linker suffix -ng, like bagong paaralan, should be hand-coded as belonging to the same phrase; we identified 19 such sequences in the song corpus.

We assumed that line breaks occurred where displayed on the lyrics website Musixmatch (https://www.musixmatch.com/). Some of the songs rhyme, providing further support for the line-break locations, but we did not use rhyming or other criteria to make changes from Musixmatch’s line breaks.

Figure 5 illustrates this three-way breakdown. As we move from phrase-medial to phrase-final to line-final, durations get longer, especially for ultimas. But the key result still holds in all three environments: prominent penults are longer than non-prominent penults; and prominent ultimas are longer than non-prominent ultimas.

Figure 5 Note duration broken down by position.

4.3. Duration summary

We have seen that, as predicted by both the vowel-length analysis and the stress analysis, prominent penults are set to longer notes than non-prominent penults. As predicted by only the stress analysis, prominent ultimas are also set to longer notes than non-prominent ultimas. This is true even for phrase-medial position, where, in speech, ultima-prominent words may lose their intonational prominence.

We defer discussion of statistical significance to §6. A regression model, included in the Supplementary Material, finds that ultimas are longer, prominent syllables are longer and prominent ultimas are the longest of all. In §6, we argue that regression may not be conservative enough, and provide an alternative method for assessing significance.

5.1. Measuring beat strength: syncopation

OPM has extensive anticipatory syncopation (Temperley Reference Temperley1999; Tan et al. Reference Tan, Lustig and Temperley2019), meaning that beats that, according to various expectations, should count as strong, begin slightly early. Two examples are shown in Figure 6: the syllable ram begins on the ‘and’ (second half) of a beat, the second-weakest position, but musically it behaves as though it began on the following measure’s first beat, which is the strongest position; the syllable man begins on the last sixteenth note of a beat, the weakest position, but musically behaves as though it began on the third beat of the measure, the second-strongest position. Temperley’s (Reference Temperley1999) solution is to move each syncopated note forward, so that it counts as beginning later than it really does.

Figure 6 Examples of syncopation in first line of ‘Akin ka na lang’.

To operationalise this, we identified, for each syllable, the strongest beat that it contains: for ram, beat 1 of a measure, and for man beat 3 of a measure. One danger is that if a note is very long, it will always end up counting as strong, because it eventually goes on long enough to include a strong position. Therefore, we only looked for the strongest beat contained in the first 1.25 beats of the note. This is enough for a note that begins on the last beat of the measure (or even one sixteenth-note earlier) to count as beginning on a downbeat, if it lasts long enough.Footnote ¹² Our procedure ‘corrected’ 49% of sixteenth notes, 27% of eighth notes and 11% of quarter notes to a stronger position.

5.2. Beat strength results

As with duration, there were 498 penult-prominent words and 475 ultima-prominent words analysed.

5.2.1. All final two syllables

Results are plotted in Figure 7. The plot is the same as in Figure 3, except that the vertical axis measures beat strength. The strongest value, 5, is for a note that starts on or contains a downbeat (beat 1 of the measure); the next-strongest, 4, is for a note that starts on or contains beat 3 of a measure; 3 is for a note that starts on or contains beat 2 or 4; 2 is for a note that starts on or contains at most the second half of a beat, and 1 is for a note that starts on and contains at most the second or fourth quarter of a beat.

Figure 7 Beat strength (correcting for syncopation) of final two syllables of all words.

On the left, we see that in penult-prominent words, average strength is somewhat higher in penults than in ultimas; the modal penult of such words is set to a downbeat, while their modal ultima is set to the second half of a beat (the peak of the distribution is at 2). On the right, for ultima-prominent words, ultimas are on average much stronger (modally downbeats) than penults (modally second half of a beat). Thus, beat strength matches prominence, especially for ultima-prominent words.

Just as we did for duration, we can also divide the results by syllable shape, as shown in Figure 8, with little effect: regardless of syllable shape, penult-prominent words have a stronger penult, and ultima-prominent words have a stronger ultima.

Figure 8 Beat strength broken down by syllable shape.

5.2.2. Separating out phrase- and line-final words

Just as with duration, we separate out phrase- and line-final words. As shown in Figure 9, for phrase-medial words, penult-prominent and ultima-prominent words look symmetrical, but moving to phrase-final and then line-final, penult-prominent words show less and less of a strength difference between penults and ultimas. Nevertheless, in all three environments, prominent syllables are stronger than non-prominent syllables within the same word type; prominent penults are stronger than non-prominent penults; and prominent ultimas are stronger than non-prominent ultimas.

Figure 9 Beat strength broken down by position.

5.3. Beat strength summary

The results again support the stress analysis. As both the vowel-length analysis and the stress analysis predict, prominent penults are set to stronger beats than both non-prominent penults and non-prominent ultimas. But as predicted only by the stress analysis, prominent ultimas are also set to stronger beats than both non-prominent ultimas and non-prominent penults. This holds even phrase-medially, where, in speech, ultima-prominent words may lose their intonational prominence.

A regression model, included in the Supplementary Material, finds that prominent ultimas and penults are set to the strongest beats, and non-prominent penults to the weakest. A more conservative method of assessing statistical significance is given now.

8.2. Pre-tonic duration

The bean plots in Figures 14 and 15 show the distribution of note duration for pre-tonic syllables that are short-vowelled and open, short-vowelled and closed, or long-vowelled and open. (There were no long closed tokens.) In fourth-to-last syllables (Figure 14), we see that songwriters have assigned the short open syllables to the shortest note durations. In contrast to the duration data for penults and ultimas above in Figure 4, syllable shape does matter here: the closed syllables pattern with the long-vowel syllables – although, not surprisingly given the small amount of data, the differences are not significant.Footnote ¹⁴ This is consistent with French’s (Reference French1991) contention that both closed syllables and long-vowelled syllables (aspect reduplicants) attract secondary stress, and goes against the otherwise appealing notion that the reason a closed penult cannot be prominent is that it can’t have a long vowel: even though these closed pre-tonic syllables have short vowels, they appear to be receiving prominence.Footnote ¹⁵

Figure 14 Note duration in fourth-to-last syllables.

Figure 15 Note duration in third-to-last syllables.

For third-to-last syllables, in Figure 15, the data are further divided according to whether the following syllable is prominent (penult-prominent word, as in liːlípas ‘will elapse’) or not (ultima-prominent, as in puːputíʔ ‘will turn white’). The only syllables to receive longer note duration are closed and long-vowelled syllables that are not followed by a prominent penult, and thus not subject to stress clash. The difference between on the one hand long-vowel syllables not subject to stress clash and on the other hand short-vowel syllables was significant ( $p < 0.001$ ).

8.3. Pre-tonic strength

The results for beat strength are less perfectly in line with our secondary-stress predictions, but still broadly support them. In fourth-to-last position, long-vowelled syllables – but not closed syllables – trend towards being set to stronger beats, as shown in Figure 16.

Figure 16 Beat strength in fourth-to-last syllables.

In third-to-last syllables (Figure 17), the syllables set to the strongest beats are those that we predict to have secondary stress: closed and long-vowelled syllables in ultima-prominent words (no stress clash). There also appears to be a difference within the short open syllables between those that are followed by a prominent syllable and those that are not. It could be that short open syllables prefer to bear secondary stress if followed by an unstressed syllable (Blake Reference Blake1925; Avery & Lamontagne Reference Avery and Lamontagne1995). There is also a plausible musical explanation for this: unlike note length, beat strength alternates in the underlying musical structure. Because prominent penults and ultimas tend to be assigned to strong beats, there will thus be a tendency for an antepenult preceding a prominent penult to be weak, and for an antepenult preceding a prominent ultima to be strong. In the case of short-vowelled open syllables, this musical tendency creates a small difference; in the closed and long-vowelled syllables the musical tendency combines with stress clash avoidance to create a bigger difference. The difference between, on the one hand, long-vowelled syllables not subject to stress clash and, on the other, short-vowelled syllables was significant ( $p < 0.001$ ).

Figure 17 Beat strength in third-to-last syllables.

8.4. Pre-tonic syllables summary

We have seen that closed or long-vowelled pre-tonic syllables are set to longer notes and stronger beats, as long as the following syllable is not the tonic (as in a word like liːlípas ‘will elapse (time)’). This supports French’s (Reference French1991) contention that pre-tonic closed syllables and long vowels attract secondary stress, subject to some stress clash avoidance. We saw earlier that open versus closed syllable shape in penults and ultimas, which does not affect stress (except that closed penults may not bear stress), was not important for note length and beat strength. Thus, syllable shape seems to matter for text-setting only where it has been claimed to matter for stress.