Semantic Parallelism in Traditional

In the present article, I propose some initial topics for the typological-comparative study of semantic parallelism, which also serve as an analytical axis to study semantic parallelism in traditional kakataibo chants. These basic topics are the structure of the cotext, lexical and grammatical type of parallel units, type of semantic relations between parallel units, number of parallel units per pair of lines, and associated phenomena such as recurrent or fixed pairing and diphrasism. Following this initial proposal, I postulate that the cotext in Kakataibo semantic parallelism is a repeated morphosyntactic structure that does not exhibit ellipsis or increasing, the number of parallel units can be one to three slots and one pair is more productive, parallel units are related by semantic fields such as kinship terms and colors. Furthermore, I analyze the interaction of semantic parallelism in the continuous creation of lines with two other poetic forms, enjambment and repetition, and I postulate that semantic parallelism is a compositional strategy due to its high productivity for line composing.

I have organized the present article in the following way1. In (1), I offer a brief characterization of Kakataibo language and its speakers. In (2), I present my corpus and the data I employed in the present article. In (3), I discuss some theoretical definitions of semantic parallelism available in the literature and I have outlined an initial analytical framework for future typological-comparative studies, based on 14 languages (see Annex). In (4), I present ethnographic information about traditional kakataibo chants. In (5), I analyze Kakataibo semantic parallelism following (3). In (6), I study the combination of semantic parallelism with enjambment and repetition to create larger groups of lines and/or thematic passages. In (7), I propose that semantic parallelism is the most productive poetic form in tradition Kakataibo chants and can be classified as a compositional strategy (Frog, 2009). Finally, in (8) I offer some final remarks.

Some notes on Kakataibo language and its speakers
Kakataibo is a Panoan language spoken in the Peruvian amazon. Its speakers live in the departments of Huánuco and Ucayali. According to the last census of Instituto Nacional de Estadística e Informática (2017), 1553 people identified themselves as Kakataibo speakers; however, FENACOKA, the Native Federation of Kakataibo Communities, estimated 3500 Kakataibo around 2007 (Zariquiey, 2018). Regarding the dialectological distribution of the language, Zariquiey (2011b) proposed five dialects grouped into two main branches: Lower Aguaytía, Upper Aguaytía, Sungaroyacu, San Alejandro and Nokamán †. The first branch is Aguaytía/Sungaroyacu and groups Bajo Aguaytía (BA), in one subgroup, and Upper Aguaytía (AA) and Sungaroyacu (SU) in another subgroup. The second branch is San Alejandro/Nokamán and it is composed by San Alejandro (SA) and Nokamán †, a variety already extinct. The traditional kakataibo chants analyzed in the present article come entirely from the Lower Aguaytía variety and I recorded them in Mariscal Cáceres and Yamino. The phonological inventory of Kakataibo from Lower Aguaytía is composed of 15 consonants and 6 vowels. Table 1 and Table 2 summarize the phoneme inventory of the Lower Aguaytía variety (orthographic conventions are included in brackets).  With regards to its syntax, following Zariquiey (2018), Kakataibo is a postpositional and agglutinating language that shows high levels of synthetic verbal morphology. Word order is pragmatically oriented; however, verb-final sentences are the usual. Kakataibo distinguishes between closed and open word classes. The following are closed classes: pro-forms, postpositions, numerals and quantifiers, interjections and onomatopoeic words. Open classes are nouns, verbs, adjectives, and adverbs. Furthermore, Kakataibo has a rich switch-reference system and it uses nominalizations systematically in discourse.

Data and methodology
My data comes from two databases: Zariquiey (2014) and my fieldwork during 2015-2017. From Zariquiey (2014), I used five chants recorded in Yamino during the years 2010-2012, using a ZOOM-H4 recorder and in an uncompressed format WAV (stereo, digitized at a ratio of 44,100 Hz and 16 bits).
From my fieldwork, I used five chants documented in Yamino and Mariscal Cáceres, recorded with a Sony ICD-PX312. These ten chants are divided into five no bana 'iti sung by two kakataibos from different communities and in different years; two rarumati, different singer, different year, different community; two bana tuputi, same singer and same year; and only one kankiti, see Table 3. All my informants are over fiftyfive years old.
I transcribed and translated each song during fieldwork in a notebook with the help of an experienced Kakataibo singer and using the alphabet proposed in Zariquiey (2011a). Then, I segmented line by line the audio in ELAN, software from the Max Planck Institute for Psycholinguistics. Lines in kakataibo are defined by meter (Prieto Mendoza, 2015). In total, my corpus contains 840 lines (Table 3). See the following Table 3.  Total 840

Traditional Kakataibo chants
As for traditional Kakataibo chants, Erwin Frank affirms that "they are one of the main keys for the understanding of the life and culture of this group" (Frank 1994, 227). Due to this, it is not difficult to suspect that there is a long documentary tradition centered on chants (Brabec de Mori 2011; Frank 1983Frank , 1988; Prieto Mendoza 2015Winstrand-Robinson 1975, 1976. Considering the bibliography available and my own fieldwork itself, there has been documented a total of 13 different types of chants of which nine have available recordings. Kakataibo chants can be distinguished, first, by a marked gender distribution, and only one chant, rarumati 'funeral wailing', is sung collectively. Some men's chants are no bana 'iti, ño xakwati, and xunkati. No bana 'iti is about life experiences that have marked the life of the singer: for example, military service, some physical prowess or learning how to hunt. The singer highlights physical abilities, the aptitude of not being deceived, etc. The 'inka, a kakataibo mythological figure, plays an important role in no bana 'iti and this figure is usually related to metal tools, shotguns, airplanes, boats. It is usually sung at night (from 6pm) or very early in the morning (3-5am).
Ño xakwati is sung before hunting or during the last hours of the night to attract the ño 'peccary'. In addition, kakataibos, as other Panoan groups, raise peccaries and, in the past, when these peccaries where sufficiently grown to eat, families or neighboring groups gathered to kill the raised peccary on a festivity that is no longer practiced, during this festivity, ño xakwati was also sung.
Xunkati is about two main activities. The first is to stop storms, heavy rains or thunders; the second is to accompany healing while chanting. In both cases, kakataibos conceived xunkati as a very powerful song that not everyone can sing, but only experienced kakataibos that correctly know the kakataibo way of life, that is those who have practiced a very strong diet and follow complex systems of rules and taboos. Nowadays, xunkati is maybe the most almost forgotten men's song and the kakataibos that I have worked with declare that they do not know how to sing it as it was in the past. My recordings have very few lines in comparison to no bana 'iti or ño xakwati.
Women's chants, bana tuputi are about life experiences, but with special attention to the parents or grandparents of the signer raised her in childhood. Kakataibo women usually refer to this learning process, as "how one became señorita (young lady)". The singer not only tells her childhood but also instructs her offspring in the kakataibo way of life while singing. Other usual topics are the travels of her children or the nostalgia that the singer feels because of it. Kankiti is about arguments or fights with a member of the community; also, when someone has spoken badly of the singer or her relatives, or the singer suspects it. Finally, kakataibos sing rarumati when a relative or a member of the community dies, or when they remember a deceased relative. Rarumati can last hours; also, the family continue singing rarumati days after the departure.
Another important aspect of traditional kakataibo chants is that although there are recurring themes in each type of chant, each performance is unique and different between singers and even in the same singer. For example, a singer while chanting no bana 'iti never sings in the same way his military service. In other words, there is not a predetermined text and kakataibos improvise every performance. Although, there are formulas and other poetic forms that are used repeatedly.
With regard to the learning process of chants, men used to learn chants in the midst of an initiation rite for young adolescents, which was conducted by experienced adults who were in charge of teaching the Kakataibo way of life (Frank 1983; Montalvo Vidal 2010). Youths were isolated in a house built especially for the occasion, the mëkëkë xubu 'the house of the medicine', and the rite could last for weeks or several months. During these weeks, the initiates learned the different types of chants and their functions, as well as internalising the rhythm, content, melody, and the form of the chants. In addition, this initiation rite involved a series of rigorous food and physical practices, such as the prohibition of chili and cooked meat, and a harsh exercise regime during the mornings. In addition, any sexual relationship was prohibited, even seeing a pregnant woman or in her menstruation was forbidden (Frank 1983). Otherwise, the initiates could not learn the "magical" power attributed to chants, for example, stopping rain and thunder or curing. However, this "rite of initiation" was not the only moment of learning; on the contrary, learning itself was a process initiated in the early stages of adolescence and carried out throughout life.
At present, kakataibos no longer practice this rite of passage and many Kakataibo declare they no longer know how to sing "traditional chants"; nevertheless, a considerable group of adults, between 50 and 80 years old, actively sing and seek to revitalize this practice. According to my own fieldwork experience, chanting is usual in Yamino and Sinchi Roca communities, especially in the latter. In addition, Kakataibos have been in contact with missionaries of the Summer Institute of Linguistics and other evangelical groups for more than sixty years, due to this, nowadays some Kakataibos sing evangelical songs in their language and they are even composing new songs with religious topics (Brabec de Mori 2011).
Finally, traditional Kakataibo chants have a high metaphorical content and a clear lexical and morphosyntactic imprint of Shipibo-konibo, another Panoan language, due to the prolonged contact with this group (Prieto Mendoza & Zariquiey 2018).

Semantic parallelism: some initial questions for a typology
Semantic parallelism has been widely studied from the stylistic analysis of the poetic books of the Old Testament carried out by Robert Lowth (1753) Later, during the first half of the 20th century, a long series of studies proliferated in many different languages of the world: for example, Vietnamese, Chinese, Classic Maya, Quechua, Kuna, Russian, Turkish, etc. (Fox 2014;Jakobson 1966).
Following Roman Jakobson and later authors (Fabb 2017;Fox 1988Fox , 2014Frog 2014;Jakobson 1966), semantic parallelism contributes to the organization and text coherence of verbal art by the organized co-occurrence of elements in a text unifying these elements at a semantic level and a formal level through a relationship of repetition, similarity and/or contrast. Other names for semantic parallelism are canonical parallelism, syntactic parallelism and grammatical parallelism.
Several authors propose that semantic parallelism fulfills another diverse range of functions. For example, it highlights and gives importance to certain elements in the discourse (Fox 1971(Fox , 1977(Fox , 1988, it establishes interpersonal relationships between interlocutors (Rodríguez Cuevas 2013; Tannen 1987), it is intrinsically related to memory, both long-term and short-term (Fabb 2017), and it is related to the culture and society of the speakers and/or language in question (Mannheim 1986). In this article, I define semantic parallelism as the semirepetition (or variation) of a pairs of lines, the base line and the goal line, in which a number of semantically related linguistic elements alternate.
In order to clarify this definition, I propose that semantic parallelism has two basic parts: the cotext and the parallel units. I have defined cotext as the set of identical or quasi-identical repeated linguistic elements that accompany the parallel units, and I understand for parallel units the linguistic elements that alternate in a cotext. An example of semantic parallelism from Nahuatl is presented below in (1). (1) in ahuaque in a-hua-que the water-owner-PL 'The lords of the water' in tepehuaque in tepe-hua-que The mountain-owner-PL 'The lords of the mountain' (Bright 1990, 440) What we have referred to as a cotext, according to Bright (1990), is manifested in Nahuatl as a repeated morphosyntactic structure, the one highlighted in italics, and, in turn, what we have referred to as 'parallel units' would be the lexical elements a-'water' and tepe-'mountain', which, following the terminology of Bright (1990), can be grouped in Nahuatl 'Coloquios' through semantic relations of synonymy, antonym, and contrast -highlighted in bold in (1). However, since semantic parallelism is also a linguistic phenomenon intrinsically linked to culture, it should be emphasized that the semantic relationship between parallel units can not only be explained under strictly semantic criteria -synonymy, antonymy, semantic fields, contrast, part-everything, etc.-, but, also, it is necessary to understand the semantic relations between the parallel units within the same historical-cultural development of the discursive genre of the language in question. For example, see the following example of the Quechua oral tradition: hill-LOC owl-LIM cry-3S 'The owl sings at the hill of my departure' snowy.mountain-LOC magpie-LIM cry-3S 'The magpie sings in the snow-capped mountain where I am leaving' (Adelaar 2004, 64) Following Adelaar (2004), the tuku / waychaw (owl / magpie ) pair, specifically is associated, not because of strictly semantic criteria, but, on the contrary, for the reason that the song of both birds is interpreted in the frame of Quechua culture as premonitory of bad omen; in the same way, other pairs of cultural association are rumi / sacha (stone / tree), mayu / qaqa (river / stone), sunqu / ñawi (heart / eye), and others (Adelaar 2004, 64).

The structure of the cotext
What it concerns to the cotext, it was defined backwards as the set of identical or quasi-identical repeated linguistic elements that allow to identify the parallel units. For example, see below the following example of the Quechua oral tradition (3) unuy rirpu llullam kanki unu-y rirpu llulla-m ka-nki water-1POSS reflex lie-EVD be-2S '(As) reflection of water, you are an illusion' yakuy rirpu pallqum kanki yaku-y rirpu pallqu-m ka-nki fount-1POSS reflex farce-EVD be-2S '(As) reflection of the fount, you are a resemblance' (Husson 1993, 65-66).
In (3), both lines are composed of an identical (repeated) morphosyntactic structure composed of the following elements: the infinitive suffix -y, rirpu 'reflex, mirror', and, finally, the verb ka-nki 'to.be-2S'. This is schematized in (4): However, the cotext is not always an invariable repeated morphosyntactic structure; but, on the contrary, in certain traditions, such as that of the Mayan languages (Hull 2003 In (5a), lines 283 and 284 have a strict formal relationship in the cotext; nevertheless, a part of this, specifically, yak, has been eliminated. Considering this, (5a) exhibits a case of partial ellipse of the cotext.
There is a different situation to (5b), in this case, lines 82 and 83 are both composed under the same lexical and morphosyntactic elements; however, the oxlajuneb element has been added, this being a case of cotext augmentation. The same situation is on lines 318 and 319, where aw-o'tan and-o'tik are added, which, in terms of this classification, is also a case of cotext increasing.
A typological-comparative approach to cotext structuring should ask, what are the structural possibilities of the cotext in a language and in world languages? is it just a partial repetition of two pair of lines? is it possible to increase or to reduce the cotext in a language?

Lexical and grammatical type of parallel units
Another important topic for a typological approach of semantic parallelism is the possible lexical and grammatical categories of the parallel units. Although there is not a detailed study of this aspect, in the languages in Annex 1, the parallel units tend to be nouns, verbs and adjectives. For example, in the Lahu liturgical discourse (Tibeto-Burman), parallel units can be nouns, as in (6), qhɔú 'hill' / qhɔmɛ̄ 'hill skirt': (6) ŋà {qhɔú cɔ gä qo yù lɛ qhɔú tā lâ} 'when I circle round the hilltops put them on the hilltops for me' {qhɔmɛ̄ cɔ gä qo yù lɛ qhɔmɛ̄ tā lâ} 'when I circle round the foothills put them at the foothills for me' (Matissof 1991,19) In the same way, in the sung tale Tom yaya kange of Ku Waru (Trans New Guinea Nuclear), the parallel units can also be verbs; for example, in (7), the verbal roots pily-'listen, know, feel' / kan-'see': Pilyini kub nam ko nitim e 'whom do you want to hear? She said' Kanini kub nam ko nitim e 'whom do you want to see? She said' (Rumsey 1995, 112) However, not only nouns, verbs and adjectives can be used, but also other domains of grammar such as morphemes; see the example in (8) According to Rodríguez Cuevas (2013), in Chol semantic parallelism, grammatical features of a person can be used as well as nouns, verbs and adjectives. Also, she notices that this type of unit is especially prolific in situations in which the interlocutor repeats or constructs the cotext. Considering this, in (8), the parallel units are "the absolutive suffix of the first person plural exclusive, -oñloñ 'us', and the second person of the plural, -etyla 'you'" (Rodríguez Cuevas 2013, 132). Note that (8), similar to the examples analyzed so far, is a perfect candidate to be considered as semantic parallelism, since it also exhibits a repeated cotext in which a semantic relationship is established between two parallel units. However, these parallel units are person suffixes and not names, adjectives or verbs.
The same situation is found in the following examples in (9) from the Yucatec Maya and Tzeltal languages, respectively. In these languages, parallel units can also be aspectual grammatical features such as aspect (wa 'hypothetical' / k 'habitual' (9a), person (te'ex 'you' / ten 'me') and negation (wa ma '' yes no '/ mix ba'a' nothing '). (9) a. wa úuch ten lòobe' a 'if something wrong happens to me' kuyúuchu te'ex lòob 'something wrong happens to you' wa ma' úuch ten lòobe' 'if anything wrong happens to me' mix ba'a kuyúuchu te'ex 'nothing wrong happens to you' (Becquelin & Becquey, 2008:128) In summary, some initial questions are: what are the possible lexical and grammatical categories to employ as parallel units? Is there a category more common than others in a language and in world languages? And if so, what this can tell us about language structure in general? What are the possible combinations of lexical and grammatical categories as parallel units in a language?

Number of parallel units
An excluded aspect in the study of semantic parallelism is the number of parallel units. As it was previously defined, two lines composed under semantic parallelism show a minimum of two alternated elements semantically. Considering this, another important question is the number of parallel units within a pair of lines or whether there a production differences between those numbers. I will refer to each pair of parallel units as a slot. In other words, is there a general number of slots in a tradition? How many slots can be in a pair of lines? and in case there is a more general number of slots, what conditions allow it?

Associated phenomena
Finally, in addition to the structural characteristics of semantic parallelism, the use of it usually leads to associated behavioral phenomena. Some of these associated phenomena are fixed or recurrent pairing and diphrasism (difrasismo).

Fixed or recurrent pairs
In the first case, semantic parallelism can lead to the presence of fixed or recurrent pairing of parallel units.  (Emeneau 1937), and Rote (Fox 2016, 414-416), there is a group of units that requires the 'mandatory' or recurrent presence of a particular unit in the goal line. For example, although there is no systematic study of fixed pairs in Quechua oral tradition, there has been proposed the recurrent pairing of the following units: waqa-'to cry' / llaki-'to suffer', tayta-'father' / mama 'mother', ripu-'to leave' / pasa-'to leave (Spanish loan), urqu 'hill' / qasa 'x', inti 'sun' / killa 'moon', quri 'gold' / qulqi 'silver', punchaw 'day' / tuta 'night' (Lienhard 2005).
To illustrate, take the case of the pair apa-'to carry' / pusa-'to lead', which is verified from 17th century in the Nueva corónica y buen gobierno of Guaman Poma de Ayala (Husson 1993, Mannheim 2015 and in the 20 th century in huaynos, harawis and other genres (Adelaar 2004, Cavero 1985, Lienhard 2005, Mannheim 1986, 1998 In other words, the fixed or recurrent use of two lexical items as parallel units can become non-extensive in meaning. For example, the use of tzicuehua 'splinter' and tlapani 'break' means 'to be born' (Montes de Oca Vega 1997). See the following example: (14) 160 ay to ta amahtantesel 'your offer still remains' 161 ay to ta awoboltesel 'your contribution still remains' (Becquelin & Becquey 2008, 124) In this case, (14) shows two parallel units with a clear semantic relation; however, the frequent or fixed use of a pair, in this case amah 'offering' / obol 'contribution', generates a non-compositional meaning over time, in this case, 'suffering'.

A brief summary
As I have proposed in this section, the crosslinguistic and typological study of semantic parallelism should take into account the following questions. In relation to cotext, how is expressed in a given tradition? Does it allow ellipsis or increasing? How common are these two processes in comparison with regular cotext? In second place, considering parallel units, some areas of research are their lexical and grammatical type, their combination, or, a topic not worked in this paper, their morphological behavior. Other topics related to parallel units are the semantic relation between them or number of slots. Are there categories more plausible to function as parallel units than others? Is there a more general number of slots in a given tradition? Is one slot more common than three slots across world languages? Finally, I have commented on two associated phenomena of semantic parallelism: fixed or recurrent pairs, and diphrasism. About the former one, what conditions allow it and what can tell us one about culture? Is diphrasism found only across Mayan languages and neighbor ones?

Kakataibo cotext
The kakataibo cotext is always composed by the repetition of two lines, and the goal line does not exhibit any type of change. I have not found in my corpus any cases of ellipsis or increasing of the cotext as in Mayan languages. Following the example (15), its cotext is schematized in (16): (16) parall.un=n 'a-këxun-ma parall.un=GEN do-O>A(PE)-NEG

Type of parallel units
Regarding parallel units, I identified a total of 209 pairs of them in 840 lines. These 209 pairs are composed mostly of nouns, verbs, and adjectives, (see the following Figure n°1 showing the productivity of these three classes). Adverbs and closed classes are grouped into the category "other" due to their low productivity; I have only found two cases of adverbs in all my corpus.  The example (17) exhibits two slots. The first is composed by the verb roots tupu-'advise' and ñui-'say', which comply with the Kakataibo cotext. On the other hand, considering the morphemes -i 'IMPF' and -a-'PERF, these also appear in an identical cotext and they are also semantically related due to their condition of aspect morphemes. However, it should be noted that this example is unusual in my data. In this way, the compositional structure of the example in (17) would be the following (18).

Type of semantic relations in parallel units
To begin with, in this section I only analyze the semantic relations in nouns and adjectives in a total amount of 114 pairs of parallel units. I exclude verbs because I have not yet identified a stable criterion to explain them, this class needs further investigation.
Nouns as parallel units are related by the following four semantic fields: kinship terms, colors, type of animals, and locations; adjectives only by colors. For example, (19) shows the pairs xëxa 'ravine' / baka 'river' of the location semantic field. (19) 73. xëxa rëbun kamabi xëxa rëbun kamabi ravine peack all 'He was walking by the peak of the ravines' 74. baka rëbun kamabi baka rëbun kamabi river peack all 'He was walking by the peak of the rivers' (AP)-RA-rarumati_mariscal-2017.73-74 In the case of colors, animal nouns can also function as adjectives by modifying with a principal color attribute of the animal (Zariquiey & David 2013 Considering these four semantic fields, kinship terms is the most productive one (67.5%), followed by locations (14.9%), colors (8.8%) and type of animals (8.8%), this is schematized in the next Figure n°3.

Recurrent pairs in Kakataibo
As I commented in the previous section, in some traditions such as Rote and Quechua, there are a group of pairs that demands its obligatory copresence. A similar situation applies for traditional Kakataibo chants.
The following Table n°4 shows all the pairs in my corpus related by the semantic field of locations. Out of a total of seventeen pairs, twelve of them were the pair me 'land'/baka 'river'. An interesting detail is that that there is no established order in this pair, it can be me/baka or baka/me. However, the copresence of these units is not exclusive since both can be associated with other items: baka/xëxa and me/ñaká. I postulate that this particular pair are recurrent for the reason that it is usually employed in other Kakataibo dialects that I have worked with: San Alejandro and Sungaroyacu. The following example (21) comes from a kankiti chant in Sungaroyacu dialect. This chant was recorded in Mariscal Cáceres and it was sung by DG. She was visiting her sister who married a Kakataibo from this community. DG has lived all her life in Puerto Azul and her singing was considered old and different by the kakataibos of Mariscal Cáceres.
(21) 9. tita bakama tita baka=ma mother river=NEG 'my mother is not in this river' 10. tita mema tita me=ma mother land=NEG 'my mother is not in this land' (AP)-DG-kankiti_mariscal-2017.9-10 A similar situation is found in the 'kinship terms' semantic field. The pairs bakë 'son' / tua 'baby', and tita 'mama'/papa 'dad' usually appeared together in my corpus. In this way, in traditional kakataibo chants there are a group of pairs recurrently associated and not fixed pairs in the classical way.

Number of slots in Kakataibo
The number of slots in traditional Kakataibo chants can be one to three, as in the Quechua oral tradition. I show a case of three slots in (22 However, the numbers slots per pair of lines show different levels of productivity see the following Figure  n°4 showing the productivity of slots in Kakataibo semantic parallelism.

Kakataibo semantic parallelism as a compositional strategy
At last, these three main poetic forms exhibit high levels of productivity for line-composing. See the next Figure n°5 showing the productivity levels of these three poetic forms. Considering Figure 5, semantic parallelism is the most productive poetic form followed by enjambment and repetition⁵. 204 pair of lines (48.57%, 408 lines) in my corpus have been composed under it. This percentage is interesting for comparative purposes, considering that Phillips (1981, p.125) found a similar result for the sijobang narrative poem sung from East Sumatra: 37% of the oral recitation was composed based on semantic parallelism. In addition, following Frog (2009), a compositional strategy is a linguistic principle that allows the systematic, continuous, and spontaneous creation of lines in the same event of composition in performance. I interpret all three of these poetic forms as compositional strategies due to their high productivity for line composing.

Final remarks
In the present article, I have sought to fulfill two main goals. The first one was to outline a basic comparative framework for the typological study of semantic parallelism. The following aspects were studied: structure of the cotext, type of parallel units, number of slots, and associated phenomena: recurrent pairs and diphrasism. The second goal was to apply this initial proposal to analyze semantic parallelism in traditional kakataibo chants of Lower Aguaytía. This poetic form in Kakataibo shows the following characteristics: In first place, the cotext is an identical repetition between two pair of lines, without cases of increase or ellipsis. According to the lexical and grammatical category of parallel units, nouns are more productive than verbs and adjectives; aspect morphemes are also possible to function as parallel units. In addition, the combination of parallel units shows the following tendencies: nouns are the most productive category, followed by verbs, and adjectives; verbs are exclusively combined with verbs, nouns can be combined with other nouns or with adjectives, and adjectives are always combined with nouns. Adverbs and other categories are not presented in this paper due to their low productivity.
Furthermore, the semantic relation of nouns and adjectives are based on the following four semantic fields: kinship terms, locations, type of animals, and colors. Kinship terms is the most productive semantic field. As for number of slots, lines are composed, almost entirely, by only one slot, although cases of two and three slots are also possible. There is a small group of recurrent pairs, for example, baka 'river' and me 'land' tend to appear together.
On another hand, I showed how semantic parallelism is used repeatedly and/or in combination with two other poetic forms, enjambment and repetition, in order to create thematic passages of variable length.
Finally, I proposed that these three poetic forms are a compositional strategy due to their high productivity in enabling the Kakataibo singer a spontaneous and continuous creation of lines. Semantic parallelism is the most productive poetic form with 42% of lines being composed under it.
To conclude, as I remark in the introduction, the crosslinguistic and typological study of semantic parallelism has not been yet taken. I hope I have contributed to this goal and I invite to replicate and expand this initial proposal.
The study of semantic parallelism is an important area of research for the understanding of the formal, typological and areal characteristics of the speech and verbal art of the Amazonian area, an area of research that undoubtedly deserves more attention.