1 Previous Research

In previous studies, it has been observed that 3- to 5-year-old English-speaking children have difficulty comprehending and producing 2-POSS sentences (Limbach and Adone Reference Limbach and Adone2010; Pérez-Leroux et al. Reference Pérez-Leroux, Castilla-Earles, Béjar and Massam2012), and that Wapichana-Portuguese bilingual children’s performance in the comprehension of 2-POSS Wapichana sentences is not perfectly adult-like, even at the age of 7 and 8 (Amaral and Leandro Reference Amaral and Leandro2013). It has also been reported that Japanese-speaking children tend to start giving adult-like interpretations to 2- to 4-POSS sentences all at once at around 4 years of age (Fujimori Reference Fujimori2010). The results of these studies suggest that although the age at which children come to understand or produce recursive possessives in an adult-like manner varies, they undergo the following two developmental stages:

This poses a theoretical challenge: what property must be triggered in order for children to recognize the unlimited productivity of recursive possessives?Footnote ³

One explanation for this question is the DP substitution account (Roeper Reference Roeper2011; Hollebrandse and Roeper Reference Hollebrandse, Roeper, Roeper and Speas2014). In this account, recursive possessives are possible when POSSPs are projected and DPs are substituted for NPs. Until POSSPs and DPs emerge, possessive phrases with genitive markers are considered to be modifiers of the head N as if they were lexical possessives such as his in his father. More specifically, the noun phrase with a possessive phrase has a non-recursive structure as in (7) at first.

(7)

In the next stage, POSSPs are projected, and NP nodes at the top of noun phrases and inside of POSSPs are replaced by DPs, as shown in (8).

(8)

Since the structure in which DPs contain another DP through POSSPs is available, recursive possessives are possible at this stage.

2 Experiment 1

A crucial observation that supports the developmental path in (6) above is found in the results of Fujimori’s (Reference Fujimori2010) experiment. In her experiment, Japanese-speaking children who gave adult-like interpretations to 2-POSS sentences tended to give adult-like interpretations to 3-POSS and 4-POSS sentences. These results are interesting, but the number of participants of her experiment was relatively small.Footnote ⁴ To examine whether the children’s responses reported in Fujimori (Reference Fujimori2010) truly represented a general tendency, we conducted an experiment on a larger scale using similar test sentences.

2.2 Materials

The experiment was carried out using a question-answering task methodology. Twenty-seven questions were asked while three pictures were being shown along with verbal descriptions of the characters in the pictures. Out of the twenty-seven questions, sixteen were target 1- to 4-POSS sentences and eleven were filler sentences.Footnote ⁵ Each picture was presented with nine questions: five target sentences and four filler sentences for two of the three pictures, and six target sentences and three filler sentences for the other picture. Shown in Figure 10.1 is one set of the materials (a sample picture (Figure 10.1), its verbal description, and the target sentences assigned).

Figure 10.1 Sample picture in Experiment 1

• Verbal description (English translation):

(9)

This is Orenji. This is Orenji’s father, Shiro. This is Orenji’s friend, Murasaki. Orenji and Murasaki have the same flower on their hats as a sign of their friendship. This is Shiro’s dog. This is Orenji’s dog. This is Murasaki’s dog. Shiro’s dog and Orenji’s dog are friends. They have the same flower on their hats as a sign of their friendship.

• Target sentences:

(10)	Shiro-san-no	booshi-wa	nani-iro	kana?		(1-POSS)
	Shiro-san-gen	hat-top	what-color	q
	‘What color is Shiro’s hat?’
(11)	Orenji-chan-no	inu-no	fuusen-wa	nani-iro	kana?	(2-POSS)
	Orenji-chan-gen	dog-gen	balloon-top	what-color	q
	‘What color is Orenji’s dog’s balloon?’
(12)	Orenji-chan-no	inu-no	tomodachi-no	booshi-wa		(3-POSS)
	Orenji-chan-gen	dog-gen	friend-gen	hat-top
	nani-iro	kana?
	what-color	q
	‘What color is Orenji’s dog’s friend’s hat?’
(13)	Murasaki-chan-no	tomodachi-no	inu-no	booshi-wa		(3-POSS)
	Murasaki-chan-gen	friend-gen	dog-gen	hat-top
	nani-iro	kana?
	what-color	q
	‘What color is Murasaki’s friend’s dog’s hat?’
(14)	Shiro-san-no	kodomo-no	tomodachi-no	inu-no	fuusen-wa	(4-POSS)
	Shiro-san-gen	child-gen	friend-gen	dog-gen	balloon-top
	nani-iro	kana?
	what-color	q
	‘What color is Shiro’s child’s friend’s dog’s balloon?’

Figure 10.1 shows not only a picture but also notes about the characters’ names and their relationships. In the experiment, however, the participants were shown only a picture. A description of the characters was given verbally as in (9) above, accompanied by pointing.

The human characters in the picture were named after the colors of their clothes so that children could easily memorize their names. For example, one girl in Figure 10.1 was called Orenji, which means ‘orange’ in Japanese, because she wore an orange shirt.Footnote ⁶

2.4 Results

As reviewed in Section 1, it has been observed that children go through the two developmental stages in (6). In order to examine whether this is a general tendency, we divided our child participants into four groups according to their responses to 1-POSS and 2-POSS sentences: children who were not fully adult-like in their responses even to 1-POSS sentences (Group 1); children who were adult-like with respect to 1-POSS sentences and gave adult-like responses to 2-POSS sentences 40 percent of the time (Group 2) or 60 percent of the time (Group 3); and children who were adult-like with respect to both 1-POSS and 2-POSS sentences (Group 4). The overall results for the four child groups and the adult group are shown in Figure 10.2.Footnote ⁷

Figure 10.2 Results of Experiment 1 (the ratio of adult-like responses)

What is most important to our purpose is the performance by Child Groups 2, 3, and 4. Child Groups 2 and 3, who gave adult-like responses to the 1-POSS sentences but not to the 2-POSS sentences, were not adult-like with respect to the 3-POSS and 4-POSS sentences. Even Child Group 4, who made adult-like responses to both the 1-POSS sentences and the 2-POSS sentences, was not fully adult-like with respect to the 3-POSS and 4-POSS sentences.Footnote ⁸ Moreover, in each of the child groups, there was no difference between the responses to the 3-POSS and 4-POSS sentences. These results suggest that Japanese-speaking children start to give adult-like interpretations to 1-POSS and 2-POSS sentences consecutively, and then come to interpret 3-POSS and 4-POSS sentences in an adult-like manner almost at the same time.

3 Experiment 2

In Experiment 1, the children’s performance for the 3-POSS and 4-POSS sentences was not as good as that for the 2-POSS sentences. One possible reason for this is that the phrases containing tomodachi ‘friend,’ namely N-no tomodachi ‘N’s friend’ and tomodachi-no N ‘friend’s N,’ were included in the 3-POSS and 4-POSS sentences but not in the 2-POSS sentences. In adult Japanese, these phrases can have two interpretations depending on prosody when the N is animate. The children’s poor performance for the 3-POSS and 4-POSS sentences might be because of their insensitivity to the effect of prosody on interpretation. Let us take a look at an example of a target sentence containing N-no tomodachi ‘N’s friend.’ Consider (12), which is repeated here with the relevant part in bold type:

With normal prosody, where no particular stress falls on any morpheme and no prosodic break is made inside of the phrase, the bold part has the interpretation in (15).

On the other hand, with stress on the first genitive marker and a prosodic break after that, the bold part has the interpretation in (16):Footnote ⁹

In Experiment 1, all the sentences were read with normal prosody, and the adult participants assigned the interpretation in (15) to the bold part of (12). However, children who are insensitive to prosody may assign the interpretation in (16) to the bold part. In such cases, the relevant hat in (12) is understood to be not Orenji’s dog’s friend’s hat but Orenji’s dog’s hat. If children interpret the bold part of (12) as in (16), it is reasonable for their responses to differ from adults’.

In a similar way, children’s responses could also be different from adults’ in the target sentences containing the phrase tomodachi-no N ‘friend’s N.’ Consider (13), which is repeated here with the relevant part in bold:

The bold part has the interpretation in (17) with normal prosody, while it has the interpretation in (18) if there is a stress on the second genitive marker and a prosodic break after that.Footnote ¹⁰

When the bold part has the interpretation in (17), the relevant hat in (13) is Murasaki’s friend’s dog’s hat, which was the adult interpretation in Experiment 1. When the bold part has the interpretation in (18), the relevant hat is Murasaki’s dog’s hat. If children are insensitive to prosody, their responses to (13) may be different from adults’.Footnote ¹¹

Two out of the four 3-POSS sentences and three out of the four 4-POSS sentences used in Experiment 1 contained either the phrase N-no tomodachi ‘N’s friend’ or the phrase tomodachi-no N ‘friend’s N,’ whose interpretation could change depending on prosody. The child participants gave a response that could be regarded as evidence for their insensitivity to prosody 23.1 percent of the time (12/52 trials) to these two 3-POSS sentences and 32.1 percent of the time (25/78 trials) to these three 4-POSS sentences. However, such responses could also be made when the children simply drop the POSSP tomodachi-no‘friend’s’ in those sentences.Footnote ¹² It is not clear from the results of Experiment 1 whether the children’s poor performance for the 3-POSS and 4-POSS sentences was a result of their insensitivity to prosody or a result of their inability to generate structures with three or four possessive phrases. Therefore, we conducted a further experiment in which the phrases in question were not used.

3.2 Materials

The method was the same as for Experiment 1: the question-answering task methodology. Twenty-one questions were asked while three pictures were shown along with verbal descriptions. For each picture, seven questions were presented. Four were target 1- to 4-POSS sentences, and the other three were filler sentences. The target sentences were different from those in Experiment 1 in that they did not contain the phrases N-no tomodachi ‘N’s friend’ or tomodachi-no N ‘friend’s N.’ Figure 10.3 shows an example of a set of the materials:

Figure 10.3 Sample picture in Experiment 2

• Verbal description (English translation):

(19)	This is Midori, and this is Midori’s father. This is Orenji, and this is Orenji’s father. This is Midori’s dog, and this is his father’s dog. This is Orenji’s dog, and this is her father’s dog. They are all wearing a hat, and the hats have a flower on them.

• Target sentences:

(20)	Midori-kun-no           booshi-wa      nani-iro             kana?                  (1-POSS)
	Midori-kun-gen      hat-top               what-color   q
‘What color is Midori’s hat?’

(21)	Orenji-chan-no          booshi-no        hana-wa            nani-iro              kana?                (2-POSS)
	Orenji-chan-gen        hat-gen           flower-top     what-color      q
‘What color is Orenji’s hat’s flower?’

(22)	Midori-kun-no           otoosan-no     inu-no       booshi-wa     nani-iro             kana?   (3-POSS)
	Midori-kun-gen         father-gen     dog-GEN     hat-top          what-color       q
‘What color is Midori’s father’s dog’s hat?’

(23)	Orenji-chan-no        otoosan-no           inu-no          booshi-no      hana-wa      (4-POSS)
	Orenji-chan-gen        father-gen           dog-gen           hat-gen            flower-top
	nani-iro      kana?
	what-color      q
‘What color is Orenji’s father’s dog’s hat’s flower?’

The other two sets of materials have a similar design.

3.4 Results

Figure 10.4 shows the overall results of Experiment 2.Footnote ¹³

Figure 10.4 Results of Experiment 2 (the ratio of adult-like responses)

Experiment 2 was not conducted on adults. However, it is expected that adult responses would be over 90 percent for each type of target sentence in Experiment 2, considering the results of Experiment 1 above. Thus, the results shown in Figure 10.4 could be interpreted as follows: The Junior Group was not (fully) adult-like in their responses to all the four types of sentences; the Middle Group was adult-like with respect to the 1-POSS and 2-POSS sentences but not fully adult-like with respect to the 3-POSS and 4-POSS sentences; and the Senior Group was almost adult-like with respect to all the four types of sentences.

The results of Experiment 2 as well as those of Experiment 1 show that there is a stage where children’s performance for 3-POSS and 4-POSS sentences is not good compared to 2-POSS sentences. Since the 3-POSS and 4-POSS sentences used in Experiment 2 did not contain the phrases N-no tomodachi ‘N’s friend’ or tomodachi-no N ‘friend’s N,’ it could be said that the ambiguity of these phrases was not a main reason for the children’s poor performance for the 3-POSS and 4-POSS sentences.

4 Discussion

Main findings in our experiments can be summarized as follows:

Our data suggest that there are three developmental stages in the acquisition of recursive possessives, as shown in (27).

The developmental path in (27) cannot be fully attributed to incremental parsing difficulties. Under the incremental parsing hypothesis, children’s performance for sentences with recursive possessives is expected to become worse as the number of possessive phrases increases. As our data indicate, however, the children showed no greater difficulty with the 4-POSS sentences than with the 3-POSS sentences.

The results of our experiments cannot be explained directly by the DP substitution account, either. Under the DP substitution account, it is predicted that any number of possessives can be generated once POSSPs are projected and DPs are substituted for NPs, and so the difference between the second and the third stages in (27) remains to be explained. However, if we take into consideration three types of possessives available in UG and modify the DP substitution account, two analyses are possible. After illustrating the three types of possessives in Section 4.1, we provide two analyses in Sections 4.2 and 4.3. We also briefly discuss the transition between the developmental stages in Section 4.4.

4.1 Three Types of Possessives

It has been observed that there are three types of possessives: lexical possessives, DP-possessives and NP-possessives (Munn Reference Munn, Samian and Schaeffer1995; van Hout et al. Reference van Hout, Kamiya and Roeper2013; Hollebrandse and Roeper Reference Hollebrandse, Roeper, Roeper and Speas2014). We assume that these three types of possessives are available in UG.Footnote ¹⁴ Lexical possessives such as his in his car are considered to be modifiers of the head N, as (28) shows.

(28)

DP-possessives and NP-possessives are POSSPs with a DP projection and an NP projection, respectively, in their Spec, as shown in (29) and (30).

(29)

(30)

The evidence of these two types of non-lexical possessives comes from the ambiguity of the phrase the man’s hat. This phrase has two interpretations: ‘the hat owned by the man’ and ‘the hat for men.’ The first interpretation derives from the structure in (31a), where the DP the man is combined with a genitive marker to make a DP-possessive. The second derives from the structure in (31b), where the NP man makes an NP-possessive together with a genitive marker.

(31)   a.        b.

4.2 Analysis 1

Given the three types of possessives, the DP substitution account can be modified in two ways to explain the developmental path in (27) above. One analysis is obtainable if we assume that the DP substitution inside of POSSPs is delayed compared to the DP substitution at the top of noun phrases. In this analysis, a structure with one non-lexical possessive phrase develops, as in (32):

(32)   

In the first stage, the structure for lexical possessives is the only option available for possessive phrases, because DPs and POSSPs have not yet been projected. In this stage, possessive phrases with genitive markers are taken to be noun modifiers. In the next stage, POSSPs are projected, and the DP substitution takes place only at the top node of noun phrases. In the third stage, NP nodes within POSSPs are also replaced by DPs. Differing from the original DP substitution account, this analysis expects children to go through a stage where the structure in (32b) is generated.Footnote ¹⁵

In this analysis, only a single possessive phrase is possible in the form of a noun modifier in the first stage. In the second stage, children can generate not only 1-POSS structures such as (32b) above but also non-recursive 2-POSS structures such as (33), which contain an NP-possessive and a possessive phrase as a noun modifier.

(33)

In the final stage, the DP substitution within POSSPs triggers recursive structures in which a DP contains another DP through a DP-possessive, such as (32c) above. Once this structure is acquired, children can generate adult-like recursive 2-POSS structures, such as (34), by using multiple DP-possessives.

(34)

Children also have no difficulty generating adult-like structures with more than two possessive phrases at this stage.Footnote ¹⁶

4.3 Analysis 2

An alternative analysis for the developmental path in (27) above is obtainable even if we maintain the original assumption of the DP substitution account that NPs are replaced by DPs all at once. In this analysis, it is assumed that multiple occurrences of the same type of POSSP need to be licensed by a certain mechanism, which is acquired late.Footnote ¹⁷ Let us assume here that a mechanism similar to the one which licenses multiple Negative Polarity Items or WH-operators is at work in adult grammar: The POSS head is an operator-like element, and needs to be licensed by the topmost POSS in the structure.

In this analysis, children’s initial grammar is the same as the one assumed in the original DP substitution account and Analysis 1: possessive phrases with a genitive marker are considered to be a noun modifier at the first stage of acquisition. In this stage, only 1-POSS structures, such as (35), can be generated in child grammar.

(35)

In the second stage, 1-POSS structures containing a DP-possessive, such as (36), can be generated because POSSPs are projected and the DP substitution takes place both at the top node of noun phrases and inside of POSSPs.

(36)

Children at this stage can also generate non-recursive 2-POSS structures, such as (37), by using a DP-possessive and a possessive phrase as a noun modifier.

(37)

However, they cannot generate recursive 2-POSS structures, such as (38), which contain two DP-possessives, because the mechanism that licenses multiple DP-possessives has not yet been acquired.

(38)

Children at this stage are also unable to generate structures with more than two possessive phrases.

In the final stage, the licensing mechanism is acquired which allows several POSSPs of the same type to occur in one noun phrase, and adult-like 2-POSS structures, such as (38), can be generated. Recursive structures with more than two possessive phrases are also available.Footnote ¹⁸

5 Concluding Remarks

In this chapter, we have reported the results of two experiments on Japanese-speaking children, and demonstrated that there are three developmental stages in the acquisition of recursive possessives. We have also shown that two analyses are possible for our findings if we modify the DP substitution account.

The two analyses make different predictions on the acquisition of recursive possessives in relation to other constructions. In Analysis 2, a correlation is expected between the acquisition of recursive possessives and that of multiple Negative Polarity Items or WH-operators. These constructions all involve some licensing mechanism. It would be no surprise that they are acquired almost at the same developmental stage. For Analysis 1, in contrast, such a correlation is not expected.

The two analyses also make different predictions about the children’s response to particular noun phrases that contain two possessive phrases.Footnote ²⁰ From Analysis 1, it is plausible to speculate that children in the second developmental stage have more difficulty comprehending phrases such as the boy’s father’s car than phrases such as John’s father’s car, because the 2-POSS structure in (33) above, which is the only structure obtainable for noun phrases with two possessive phrases in the second stage for Analysis 1, does not have a position for the definite determiner in phrases like the boy’s father’s car. In Analysis 2, in contrast, the 2-POSS structure in (37) above is available in the second stage. In such a structure, the definite determiner could occur in the position for the lower D head. Therefore, it is not necessarily the case that children in the second stage have more difficulty with phrases such as the boy’s father’s car than with phrases such as John’s father’s car.

Further acquisition research is necessary to conclude which of our two analyses is tenable. We would like to examine this issue in future research.

1 The Kawaiwete Language

The Kawaiwete language is part of the Tupi-Guarani family, within the Tupi stock (Rodrigues Reference Rodrigues1986). The Tupi-Guarani family is divided into eight subgroups. The Kawaiwete language belongs to subgroup 5; the languages Asuriní do Xingu and Araweté are also part of this subgroup.

The Kawaiwete consist of approximately 2242 people.Footnote ² In 2014, most of the Kawaiwete people lived in the Xingu Indigenous Territory, divided among approximately thirty-four villages. A smaller part of the population lives in other indigenous territories outside Xingu (Mato Grosso).

As described by Souza (Reference Souza2004) and as confirmed by a sociolinguistic evaluation of the Kawaiwete villages located in the Xingu Indigenous Territory (Lima and Santos Reference Lima and Santos2008), the Kawaiwete people are mostly bilingual, speaking Kawaiwete and Brazilian Portuguese. Due to marriages across language groups, some Kawaiwete also speak a third language, usually another indigenous language spoken in the Xingu territory. The Kawaiwete themselves argue that the only territory where most of the population speaks the Kawaiwete language is the Xingu Indigenous Territory, where this research took place. The first material that described the Kawaiwete language was a word list collected by Schmidt (Reference Schmidt1942). The Villas Boas brothers also elicited a list of words (Villas Boas and Villas Boas Reference Villas Boas and Villas Boas1989). Missionaries have described the phonology and morphosyntax of this language (Dobson Reference Dobson1980, Reference Dobson1988, Reference Dobson1997, Reference Dobson2005) and produced a dictionary (Weiss Reference Weiss1998), as well as compilations of mythological narratives (Dobson Reference Dobson1990). Academic linguists have investigated the phonology, the pronominal system (Souza Reference Souza2004), free word order, and second position clitics (Faria Reference Faria2004; Gomes Reference Gomes2002, Reference Gomes2007).

The contributions of this chapter are threefold: first, it furthers our understanding of recursive structures in Tupi-Guarani languages, alongside the contributions of Duarte; Vieira; and Thomas in this volume. Second, it contributes to a better understanding of the acquisition path of recursive structures (see also Pérez-Leroux et al. in this volume). Finally, it dialogues with the studies of recursive possessive and postpositional phrases of Terunuma and Nakato, Sandalo et al. and Maia et al. in this volume.

2 Complex Embedding 1: Possessives

The first type of embedding we will discuss in this chapter is the type we found in possessive phrases. Possessive relations in Kawaiwete can be expressed either by possessive prefixes or by juxtaposition of NPs. The possessive prefixes are derived from pronominal forms:

In the second case, no possessive prefix is used and the possessive relation is usually expressed by pronouncing the possessor (e.g., ‘João’ in (1)) after the possessed NP (e.g. kasuru ‘dog’ in (1)). The fact that possessive prefixes are not attested with full NP possessors suggests that they ought to be analyzed as pronominal clitics rather than as agreement morphemes:

In this chapter, we are interested in possessive phrases with full NP possessors, as illustrated in (1). We will discuss the properties of possessive phrases with multiple degrees of possession. Such constructions involve stacking of NPs in the order possessee < possessor. For example, the possessee in examples (1–3) is kasuru ‘dog’ and the possessor can be simple (1) or complex (2–3):

The construction of complex possessive phrases clearly involves semantic embedding, as illustrated in (4):

(4)   

In view of this syntactic and semantic complexity, we conducted a comprehension study to assess when Kawaiwete children start to understand complex possessive constructions with embedded relations between possessors and possessees.

3 Study: Comprehension Task

3.2 Materials and Methods

This study is based on the comprehension task designed by Roeper (personal communication) and reproduced by other researchers such as Fujimori (Reference Fujimori2010) for Japanese in which children answered questions based on visual stimuli. The participants consisted of ten children: four children not yet in school (4 and 5 years old) and six children in school (6 to 10 years of age). Pikuruk Kaiabi, who is a local teacher well known by the children and their parents, was present at all of the data collection sessions. The study took place in the local school in the Diauarum village. The instructions and the study itself were fully conducted in Kawaiwete. The study was divided into a pre-test phase and the actual test phase.

3.2.1 Warm-Up and Pre-Test Phase

First, all children were introduced to six characters in two different scenarios:

‘Bowl’ Scenario

“This is Maria; she has a bowl. This is Carla, a friend of Maria. She also has a bowl. This is João, he is Carla’s brother. He also has a bowl.”

‘Basket’ Scenario

“This is Pedro. He has a basket. This is João, a friend of Pedro. He also has a basket. This is Paulo, he is João’s brother. He also has a basket.”

In the pre-test phase, we checked whether the children knew the names of the colors used in the materials by asking the color of the bowls and baskets presented in both scenarios. When the children did not know the name for a color, the experimenter introduced the name of the color in Kawaiwete. The results of the pre-test phase are presented in Table 11.2.

Table 11.2 Results of pre-test phase

Target questions	Expected answer	Age
		4	4	4	5	6	7	7	9	10	10
What is the color of the bowl?	Black	–	✔	–	✔	✔	✔	✔	–	✔	✔
	Green	✔	✔	–	✔	✔	✔	✔	–	✔	✖
	Red	✔	✔	✔	✖	✔	✔	✔ (BP)	–	✔	✔
What is the color of the basket?	Green	✔	✔	–	✖	✔	✔	✔	✔	✔	✔ (BP)
	Blue	–	✔	–	✖	✔	✖	✔	✔	✔	✔ (BP)
	Yellow	–	✔	–	✖	✔ (BP)	✔	✔	✔	✔ (BP)	✔ (BP)

In this phase, most of the 6-to-10-year-old children knew the words for colors (✔). Some answered in Brazilian Portuguese (BP). In the 4-to-5-year-old group, children were less familiar with the names for the colors. Two of them expressed hesitation (–) and one answered the questions incorrectly (✖). In those cases, Pikuruk introduced the name for the color in the Kawaiwete language.

This test phase was critical because it is reported in the literature that the acquisition of color words is slower in comparison to other adjectives such as big and little (see Carey Reference Carey, Gleitman and Wanner1982; Landau and Gleitman Reference Landau and Gleitman1985; Backscheider and Shatz Reference Backscheider, Shatz and Beals1993). Sandhofer and Smith (Reference Sandhofer and Smith1999) reported a longitudinal study in which they observed the following pattern: by 27 months, children are able to group color terms such as red, yellow, and blue; that is, at this age they are able to identify that these words belong to the same “lexical class,” but do not know the specific meaning of these words. That is, in tasks where children had to answer questions about the color of an object, children often chose the wrong color (but did not choose a word from a different lexical class, showing that they distinguish color terms from other categories). A few months later (between 28 and 29 months), children start demonstrating a better performance in comprehension tasks where they have to identify objects based on their colors. Approximately at 30.5 months, children start to use color to match objects in nonlinguistic tasks. For comparison, by 27 months children are successful in comprehension tasks with size adjectives.

The slow process of language acquisition of color words is attested in other lexical domains, such as number words (see Mix, Sandhofer, and Baroody Reference Mix, Sandhofer and Baroody2005:337). The literature reports that despite the fact that young children are able to recite number words at around two years of age (see Wynn Reference Wynn1990), their ability to match number words to the amounts that they correspond to emerges quite late (Le Corre and Carey Reference Le Corre and Carey2007; Sarnecka et al. Reference Sarnecka, Kamenskaya, Yamana, Ogura and Yudovina2007; Condry and Spelke Reference Condry and Spelke2008; Huang, Spelke, and Snedeker Reference Huang, Spelke and Snedeker2010). Before four years of age, children know some small number words but do not master the interpretation of larger numbers. For example, 2.5 year olds are ‘two-knowers’: they correctly associate the numeral “two” to quantities of two objects, but fail with higher numerals; a few months later it is reported that they become ‘three knowers’ (Huang, Spelke, and Snedeker Reference Huang, Spelke and Snedeker2013).

3.2.2 Test Phase (Critical Items)

After the pre-test phase, children participated in comprehension tasks in which we asked the name of colors in possessive constructions with increasing degrees of semantic embedding. Note that the pictures were still visible after the experimenter’s question. We tested three levels of semantic embedding:

• ‘Bowl’ Scenario

  Level 1
  (11)	Man     ipit     y’a      Maria   ma’e
  	What   color   bowl      Maria        thing
  	‘What color is Maria’s bowl?’

  Level 2
  (12)	Man   ipit     y’a   ijekoty’aawa      Maria   ma’e
  	What color     bowl friend                 Maria        thing
  	‘What color is Maria’s friend’s bowl?’

  Level 3
  (13)	Man         ipit       y’a       eki’yt     ijekoty’aawa       Maria     ma’eaFootnote 4
  	What   color   bowl   brother friend                 Maria   thing
  	‘What color is Maria’s friend’s brother’s bowl?’

• ‘Basket’ Scenario

  Level 1
  (14)	Man        ipit          tamakari        Pedro      ma’ea
  	What      color     basket            Pedro       thing
  	‘What color is Pedro’s basket?’

  Level 2
  (15)	Man        ipit            tamakari      eki’yt        Pedro      ma’e
  	What    color      basket           brother      Pedro          thing
  	‘What color is Pedro’s brother’s basket?’

  Level 3
  (16)	Man        ipit           tamakari eki’yt ijekoty’aawa Pedro ma’ea
  	What    color        basket          brother friend         Pedro thing
  	‘What color is Pedro’s friend’s brother’s basket?’

3.2.3 Results

Pre-School Group

The results for pre-school children (four children) are presented in Table 11.3. Each child answered two questions per level of embedding (one associated with the ‘bowl’ scenario and another associated with the ‘basket’ scenario).

Table 11.3 Percentage of correct answers per level of complexity (pre-school children)

Level	Result
Level 1	25%
Level 2	37%
Level 3	62%

The results suggest that the level of semantic embedding of possessive constructions did not impact children’s answers negatively. That is, children’s performance in this task seems to have actually improved as the experimenter moved to a more syntactically and semantically complex sentence. Their low performance in levels 1 and 2 could be an effect of being unsure about the name of the color for a particular object (as observed in the pre-test phase). This could also be an effect of the order of presentation of the critical items: as children got more comfortable with the task, their performance at each level improved.

Crucially, these results provide evidence that 4-year-old children can already manage complex possessive sentences, as observed in the results for level 3 where the possessor was composed of three nouns (‘What color is Pedro’s friend’s brother’s basket?’).

4 Word Order Inside Possessive Phrases

The literature on the Kawaiwete language has shown that its word order is free (see Dobson Reference Dobson1980, Reference Dobson1988; Gomes Reference Gomes2002, Reference Gomes2007; Faria Reference Faria2004). It seems that the most basic order is an SOV order, while the orders SVO and VSO are also productive:

In possessive constructions, the order of the possessor and possessee is also apparently free: the possessor may precede or follow the possessee:

Furthermore, not only does the order between the possessee and possessor vary (compare 20 and 21a), but the possessee and possessor can be discontinuous (21b, 22b, and 23b). When discontinuous, the possessee precedes the possessor:

We propose that possessive phrases are headed by a functional head (Poss) that is phonologically null. Poss selects a possessee complement, and the possessor is located in the specifier position of the phrase. Assuming a default head-complement order, this analysis derives the basic word order that was observed in preceding examples (possessee < possessor). However, the specifier can be extraposed to the right periphery of the NP, which derives another attested word order in possessive phrases:

(24)

5 Complex Embedding 2: Locative Phrases in Kawaiwete

As presented in the introduction, locative phrases can also be used to study recursion in Kawaiwete. Complex semantic embedding is also attested with locative phrases in this language, which are postpositional phrases:

In an elicitation session with two adult Kawaiwete speakers, the consultants saw a drawing that described one of two scenarios. In the first kind of scenario, several portions of a substance were distributed in several containers, which we will refer to as the conjunctive/distributive reading (see Roeper and Snyder Reference Roeper, Snyder, Di Sciullo and Delmonte2005; Roeper Reference Roeper, França and Maia2010). In the second kind of scenario, a single portion of a substance was located in a single container, which was itself recursively located in a series of other containers, which we will refer to as the recursive/‘matryoshka doll’ reading:

(26a)

Distributive/conjunctive reading

(26b)

Recursive ‘matryoshka doll’

The consultants were presented with a total of six items (three items per scenario). The scenarios were not accompanied by any verbal or written description. The participants saw only pictures (similar to the pictures in 26a and 26b) and had to provide their best description of what they were seeing. The sentences produced by the speakers are provided in the following, preceded by a verbal description of these scenarios in English:

Scenario 1 (conjunctive/distributive reading): someone put a flower inside a cup, and inside a bowl, and inside a pan and these three different containers are above a chair:

Scenario 2 (‘matryoshka doll’ – recursive): someone put a flower inside a cup, and this cup is inside a bowl, and this bowl is inside a pan and this pan is above a chair:

Scenario 3 (conjunctive/distributive reading): someone put honey inside a bottle, which was put inside a bowl, which was put inside a box.

Scenario 4 (‘matryoshka doll’ – recursive): someone put honey inside a bottle, and this bottle is inside a bowl, and this bowl is inside a pan and this pan is inside a box:

In the examples above, we present two pairs of scenarios: in the first pair, we manipulated an object (ywotyra ‘flower’) and in the second pair we manipulated a substance (eira ‘honey’). One relevant question was whether speakers would provide different descriptions of the locative relation in distributive and recursive scenarios. Another relevant question was whether the nature of the content of the container (object versus substance) would affect their answers.

Examples (27a) and (27b) show that speakers differentiate distributive and non-distributive scenarios by using different verb forms. The verb ‘put’ in Kawaiwete has two forms: one that is associated with multiple events omongy ‘put (pl)’ (27a) and one – amyĩ ‘put’ – that does not specify whether the event was distributive or not. As such, the latter form is compatible with non-distributive/recursive scenarios as in (27b).

In (27a), three containers were lined up and each one contained a flower (distributive interpretation). A successful description of this scenario would consist of mentioning the three containers each containing a flower, but it would not require mentioning the containers in a particular order. In (27b), however, the order of the containers matters, as we are talking about containers contained in one another in a particular fashion (recursively). The contrast illustrated in (27a) and (27b) shows that the consultants were sensitive to the distinction between the recursive and the distributive scenarios as they provided different descriptions for each of them.

In scenarios 3 and 4, a substance was manipulated – eira ‘honey.’ Here, the consultants provided the same description for the recursive and distributive scenarios (28). Crucially, the order of presentation of the containers in the visual scenario was respected (which is a requirement for the description of a scenario such as 4 and optional for the description of a scenario such as 3). Note that the verb utilized in the scenarios that involved a substance ujan ‘put’ was different from the verbs utilized in the scenarios that involved an object ywotyra ‘flower’. For objects, the verb form depended on whether we were manipulating a plurality of objects or not (e.g., omongy ‘put’ is used only when a plurality of objects is manipulated and amyĩ ‘put’ is used when a single object is manipulated).

The sentences produced by the Kawaiwete speakers show that semantic recursion of locatives is attested in Kawaiwete as observed for complex possessive phrases. In locative phrases, the located entity (ywotyra ‘flower’ or eira ‘honey’) precedes the location and the location can be simple (if an object or substance is inside a single container, as exemplified in (25a) or complex (as in (27) and (28)):

(29)

Assuming that adpositions are head-final and that PPs are right-adjoined to the phrases they modify, the expected word order for the sentence in (29) would be:

On the other hand, assuming left adjunction of PPs, the expected word order would be:

None of these word orders correspond to the sentences that were spontaneously produced by the Kawaiwete consultants. However, if PPs are right- adjoined, recursive embedding of locative PPs results in center-embedding. A ban on center-embedding may then trigger the successive extraposition of PPs, which generates the attested word order:

(32)

6 Conclusions

In this chapter, we have shown that semantic recursion is attested both with possessive phrases and with locative phrases in Kawaiwete. In the first part of this chapter we saw that preliminary data suggest that children as young as four interpret recursive embedding of possessive phrases correctly and that they do not try to interpret multi-level possessive phrases as conjunctive/distributive. In future studies, we intend to manipulate conjunctive/distributive interpretations of complex embedding possessive and locative phrases in comprehension and production tasks. Production tasks will be pursued in the future in order to know whether children’s ability to comprehend this kind of structure (as discussed in Section 4) precedes their ability to produce them.

In addition, we have seen that the relation between word order and hierarchical structure is not transparent with locative phrases. However, we can maintain a recursive analysis of locatives assuming right adjunction and multiple extraposition.

1 Introduction

In the last decade the debate about the role of recursion in human languages (e.g., Hauser et al. Reference Hauser, Chomsky and Fitch2002; Everett Reference Everett2005) also inspired research on the acquisition of recursive constructions (Roeper and Snyder Reference Roeper and Snyder2004; Hollebrandse et al. Reference Hollebrandse, Hobbs, de Villiers, Roeper, Gavarró and Freitas2008; Limbach and Adone Reference Limbach and Adone2010; Pérez-Leroux et al. Reference Pérez-Leroux, Castilla-Earles, Béjar and Massam2012). Initially, prepositional phrases and genitive constructions received more immediate attention from researchers, since they seemed like good candidates to study the acquisition and processing of self-embedding (see Chapters 6, 7, 11, 13, and 14 in this volume). This chapter presents data on a different type of construction: self-embedded relative clauses in Wapichana, an Arawak language spoken in Roraima (Brazil) and in the southwest region of Guyana. In this volume, relative clauses are discussed both here and in Storto et al. (Chapter 13), who present data on Katiriana, a Tupi language. With two different language families represented, it is already possible to see that relative embedding varies considerably among Brazilian languages. These are just the first steps in understanding the multiple ways in which recursion can manifest itself in these less studied languages.

The Wapichana group in Brazil is estimated to consist of around 6500 people with roughly 40 percent of the population still speaking the language. The younger generations have direct contact with Portuguese, which is in most cases the dominant language. Some of the Wapichana speakers in Brazil also speak English and have strong ties to Guyana. Most communities in the Serra da Lua region, where the data collection took place, use both Wapichana and Portuguese in their daily lives. Similarly to the situation of other native American languages, there are very few monolingual speakers of Wapichana, who are mostly elderly members of the community. In the younger generation studied in this project, we could not find a single monolingual speaker.

This study looks at the interpretation of multiple-embedded relative constructions in Wapichana by speakers from different age groups, ranging from 8 to 64 years of age. Although all participants in our study showed some degree of bilingualism, the adults had more extensive contact with monolingual speakers and lived in communities that until 10 to 15 years ago did not have electricity or extensive exposure to TV and radio in the national language. Because the language contact situation is evolving very rapidly in Wapichana communities, studies that contrast adult grammars with the grammar of younger speakers can help shed some light on how the indigenous language could be changing given current contact situations.

As we show in Section 2, relative clauses in Wapichana differ from their counterparts in English and Portuguese because they lack a relative pronoun and show a different internal structure of constituents (see Section 2 for details). Wapichana presents an interesting case in which relative constructions can be headed by at least three different parts of speech (verbs, postpositions, and adjectives), which allows us to test attachment preferences based on the lexical properties of such heads. As far as we know, there are no studies that have looked into the interpretation of sequences of relative clauses, especially regarding the attachment preferences of the second clause.

This chapter has two distinct goals. The first one is to test the interpretation of multiple embedded relative clauses in a language that allows for different parts of speech to serve as phrasal heads for relative constructions. The second goal is to test a potential variability of interpretation patterns by speakers of different age groups to see if there are any potential changes in progress regarding such patterns, given the language contact situation described above. After this brief introduction, Section 2 presents the description of relative clauses in Wapichana with a proposed treatment using head-driven phrase structure grammar (HPSG) (Pollard and Sag Reference Pollard and Sag1994). In Section 3, we look at some recent studies on the acquisition of recursion and highlight some related studies done by other authors in this volume. Section 4 presents the study and its findings.

2 Relative Constructions in Wapichana

2.1 ‘–uraz’ as a Morpheme

Wapichana uses a morphosyntactic element (‘uraz’) to license relative constructions like in (2). By comparing matrix clauses, such as in (1), with relativized ones, we can see that ‘–uraz’ appears right after the head of the construction and changes its syntactic behavior in two distinct ways. The verb can no longer be the head of a finite matrix clause, and becomes the head of a relative clause that modifies its preceding noun, as shown in example (3).

(1)	Zyn kaiwada-pa-n   kuwam.
	Girl wear-prog-in       hat.
	‘The girl is wearing a hat.’

(2)	Daunaiur   tyka-pa-n      zyn kaiwada-pa-uraz   kuwam.
	Guy         see-prog-in   girl wear-prog-rel          hat.
	‘The guy is seeing the girl that is wearing a hat.’

(3)	Zyn kaiwada-pa-uraz   kuwam.
	Girl wear-prog-rel        hat.
	‘The girl that is wearing a hat.’

Because Wapichana does not have copula verbs, other parts of speech can function as the heads of matrix clauses, such as adjectives (4) and postpositions (5). These lexical items can also be followed by ‘–uraz’ licensing embedded relative constructions that are not headed by verbs, such as in (6) and (7). Notice that ‘–uraz’ can appear right after the head of VPs, PPs, or AdjPs to form relative constructions without affecting the word order of the rest of the constituents.

(4)	Karich        barakau.
	Book      white.
	‘The book is white.’

(5)	Karichinhau miisa pa’auwa.
	Books            table on.
	‘The books are on the table.’

(6)	Un=tykap-nii         karich barakau-uraz.
	1SG=see-NPRES    book white-rel.
	‘I saw the book that was white.’

(7)	Un=tykap-nii      karichinhau miisa pa’auwa-uraz.
	1SG=see-NPRES books            table on-rel.
	‘I saw the books that were on the table.’

We propose an HPSG (Pollard and Sag Reference Pollard and Sag1994) analysis that treats ‘uraz’ as a morpheme introduced by a post-inflectional lexical rule (Sag et al. Reference Sag, Wasow and Bender2003).

This rule changes the subcategorization frame of lexical items that could function as heads to matrix clauses, forcing them to behave as nominal modifiers.Footnote ¹ One advantage of such a morphological analysis is that it avoids problems with potential discontinuous constituents that could emerge in a syntactic analysis of ‘uraz’ as a relative pronoun.

(8)  Post-inflectional lexical rule for ‘–uraz’.

$\left[\begin{array}{l}uraz-lexical-rule\hfill \\ \hfill \\ \text{INPUT}\left[\begin{array}{c}\text{PHON}\\ \text{SS|CAT}\end{array}\begin{array}{l}\begin{array}{|c|}\hline 1\hfill \\ \hline\end{array}\hfill \\ \left[\text{VAL}\left[\text{SUBJ}\begin{array}{|c|}\hline 2\hfill \\ \hline\end{array}\left[\text{HEAD}noun\right]\right]\right]\hfill \end{array}\right]\hfill \\ \hfill \\ \text{OUTPUT}\left[\begin{array}{ll}\text{PHON}\hfill & f_{uraz}(\begin{array}{|c|}\hline 1\hfill \\ \hline\end{array})=\begin{array}{|c|}\hline 1\hfill \\ \hline\end{array}-uraz\hfill \\ \text{SS|CAT|VAL}\hfill & \left[\begin{array}{c}\text{SUBJ}〈〉\\ \text{MOD}\begin{array}{|c|}\hline 2\hfill \\ \hline\end{array}\end{array}\right]\hfill \end{array}\right]\hfill \\ \hfill \end{array}\right]$

There are some important characteristics of this rule that must be highlighted. First, it only applies to parts of speech that have a non-empty value for SUBJ, which means that in Wapichana they are potential heads for matrix clauses. Even in the case of postpositions, where they can vary in terms of the subcategorization patterns of SUBJ, only those that require some nominal element as SUBJ’s value can serve as the input to this rule. Second, by changing the value of VAL, mapping the value of SUBJ into the value of MOD, we disallow the lexical item to function as the head of a matrix clause. It now becomes a nominal modifier, which means that it has to be embedded in a construction with a noun.

There could be a potential problem with this rule if the noun that was being modified were to be the object of a verb in the relative clause instead of the subject, such as in example (9) for English. In this case, our lexical rule would not allow for such constructions to be licensed. However, in Wapichana ‘–uraz’ is restricted to cases where in English the relative pronoun would function as the subject of the relative clause. In object cases, Wapichana uses a combination of topicalization with demonstratives to express the same meaning, as we can see in (10), and does not allow for the use of ‘–uraz’ in the embedded construction, as in (11).

(9)	The girl likes the guy that I saw.

(10)	Zyn naydap wyryy      daunaiur   un=tykap-nii   .
	Girl like      thisDEIC      boy        1ps=see-NPRES
	‘The girl likes the guy that I saw.’

(11)	*Zyn   naydap (wyryy)          daunaiur un=tykap-nii-uraz   .
	Girl     like      (thisDEIC)        boy       1ps=see-NPRES-rel
	‘The girl likes the guy that I saw.’

3 Some Studies in Acquisition

To the best of our knowledge, there are no studies on either processing or acquisition that have looked into the interpretation patterns for the attachment of the lower clause when two relative clauses are presented in a row. This prevents us from directly comparing our results with previous findings. The acquisition of relative clauses by children has been studied before in different languages (e.g., Corrêa Reference Corrêa1995; Hamburger and Crain Reference Hamburger, Crain and Kuczaj1982; Adani Reference Adani2011; Sevcenco and Avram Reference Sevcenco and Avram2012, among others). However, as far as we know, there are no studies that looked into the interpretation of multiple embedded relative clauses, such as the ones in this chapter. In this section we review some previous studies that have looked into the acquisition of other constructions that show recursive properties, and we highlight some studies presented by other authors of this volume.

Limbach and Adone (Reference Limbach and Adone2010) studied the acquisition of recursive possessive phrases in English and showed that children as early as 3 years old already show instances of adult-like interpretations, though not with the same consistency in performance as adult speakers. In their experiment they used stories where the characters represented by little dolls had different objects. Their scenarios allowed for five different types of answers, as illustrated in Figure 12.1 taken from Limbach and Adone (Reference Limbach and Adone2010, p.5).

Figure 12.1 Limbach and Adone scenario

Their results also showed that 3 year olds have a tendency to drop one of the DPs, while 4 and 5 year olds provide a high number of conjunctive interpretations. Another interesting finding is that non-native speakers’ recursive interpretations are significantly below native ones, with only 63 percent of recursive responses.

Terunuma and Nakato (this volume) also looked into the acquisition of recursive genitives by Japanese children. They designed two experiments that consisted of participants answering questions based on pictures with multiple characters. Their target questions were about the color of the items using sentences that varied from one to four genitives, such as in Terunuma and Nakato (this volume).Footnote ⁴

Their results suggest the existence of three developmental stages in the acquisition of genitive constructions in Japanese. In stage 1 only single possessives would be allowed. In stage 2 two possessives are interpreted correctly, and finally in stage 3 more than two possessive phrases are accepted.

Leandro and Amaral (Reference Leandro and Amaral2014) describe the structure of multiple embedded genitives in Wapichana, and provide the results of two experiments with children and adults, both in Brazil and in Guyana. Their second experiment was done with monolingual (English) and bilingual (English and Wapichana) speakers in Guyana and used a similar methodology as the one presented above for Terunuma and Nakato (this volume). Participants had to look at a picture with multiple characters and their objects, while hearing a brief story describing the characters. They then had to answer questions that contained up to four genitives about the objects’ colors, as in (17).

Their results showed that adult speakers of Wapichana interpret multiple embedded genitives the same way as adult English speakers do, and that bilingual children (Wapichana – English) outperformed monolingual English-speaking children in the interpretation of recursive genitives, as bilinguals more easily understand constructions with more than two embedded genitives.

Pérez-Leroux et al. (this volume) studied the acquisition of PP modification in DPs by looking at two different types of constructions: the recursive PP modification, as in (18), where the alligator is in the water, and the double modification of a single noun, as in (19), where the plate is under the table. According to them, by comparing double and simple embedding, it can be determined: “whether children’s structural representations of complex DPs are continuous to adults’ and whether children have a bias towards less embedded representations” (p.297).

Their results show that both children and adults produce target descriptions to their test items twice as often to non-recursive modified NPs when compared to the recursive ones. They argue that recursion directly contributes to complexity “beyond the referential demands of double modification.”

4 Current Study

As we mentioned in Section 2, all sentences with double relative embedding can be ambiguous depending on the attachment of the lower clause. Given the four possible combinations presented in examples (12), (13), (14) and (15), our research questions are:

1. 1. Will different types of embedded heads yield different interpretation patterns by adult speakers? In other words, is there a specific sequence of relative clauses (based on their lexical heads) that will favor multiple embedding? Our initial hypothesis is that there will be no difference in patterns of interpretation for ambiguous relative clause attachment.

2. 2. In case there is a pattern in the adult grammar that favors the recursive reading, will this pattern be found in the grammar of younger speakers who have been exposed to Portuguese earlier on in life?

In order to test the preferred interpretation of the four different kinds of clauses presented in Section 2, we used a picture matching test. Participants saw three pictures while hearing a sentence that described one of the scenes in them. They then had to choose the picture that was best described by the sentence. There were twelve situations (test items) in total with three items per condition, i.e., per type of sequence of embedded clauses as presented above. The experiment was programmed in PsychoPy (Peirce Reference Peirce2007), and run with sixty-six speakers between the ages of 8 and 62. The twelve items and the fillers were randomly presented to participants. Figure 12.2 and the sentence in (20) show an example of one of the test items for the V+V condition.

All pictures had three possible scenarios. The first one reflected the interpretation consistent with the embedded reading (low attachment), as illustrated by picture 2 in Figure 12.2, where the girl is wearing the hat. The second one depicted the interpretation consistent with the conjunctive reading, such as in picture 1 (Figure 12.2), where the guy is wearing a hat. The third one depicted a situation where the scenario described by one of the relative clauses was not true. In the following example, the situation described by the lower relative clause would be true (the girl was wearing a hat), but the first part of the utterance was not true (the guy was not seeing the girl).

Figure 12.2 Example of test item (V+V)

Figure 12.3 shows an example of a test item for another condition (P+P), where both relative clauses are headed by postpositions. Again, we see three possible scenarios. Picture 1 requires the embedded reading, picture 2 illustrates the conjunctive reading, and picture 3 is a distractor. The sentence that was used as the stimulus is presented in example (21).

Figure 12.3 Example of test item (P+P)

4.1 Results

As Figure 12.4 shows, there was a clear preference for the embedded reading in cases where both relative clauses were headed by verbs. For this condition, 87 percent of the participants chose the interpretation in which the lower clause modified the immediately preceding noun. Our results also show that the other two conditions where the verb was the head of the first relative clause favored the embedded interpretation. The only condition where there was no clear preference for the embedded interpretation was the one with verbless relative constructions. Participants older than 14 years old chose the embedded reading 73 percent of the time for conditions V+Adj and V+P, while condition P+P only accounted for 45 percent of embedded interpretations.

Figure 12.4 Percentage of embedded readings for 14 year olds and over

We ran a Kruskal-Wallis one-way analysis of variance with the scores for the four conditions, and the overall results showed that the differences are statistically significant (H=20.733, 3 d.f., p<0.001). When we check the paired-group comparisons to look at the normal ranges of the rank differences between conditions, we see that V+V and P+P differ significantly from all other conditions, and the only difference that is not significant is between V+Adj and V+P.

We now turn to the results by younger speakers. We grouped younger participants within three groups: (i) 8 and 9 year olds (N=17), (ii) 10 and 11 year olds (N=11), (iii) 12 and 13 year olds (N=15). The decision to include all participants at the age of 14 and above in the “adult group” was based on the analysis of the data that showed that after the age of 14 there were no statistically significant differences in participants’ responses.

Our results indicate that out of the four conditions we included in the experiment, only the one where both relative clauses were headed by verbs showed any significant difference in the interpretation patterns among groups.

Figure 12.5 Percentage of embedded readings per condition for all groups

We ran a Kruskal-Wallis one-way analysis of variance on the results by individual group within each condition. The differences in interpretation among the four groups for condition P+P was not statistically significant (H=3.361, 3 d.f., p=0.3392), with the 8–9-year-old group giving the embedded readings 34 percent of the time, the 10–11 year olds 43 percent and the 12–13 year olds 47 percent. The differences for the V+P readings were even smaller and also not statistically significant (H=1.637, 3 d.f., p=0.65101), with adults giving the embedded interpretation 73 percent of the time, 8–9 year olds 72 percent, 10–11 year olds 64 percent and 12–13 year olds 80 percent. Similar results are found for the V+Adj condition with adults providing an embedded reading in 73 percent of their responses, 8–9 year olds in 65 percent, 10–11 year olds in 52 percent and 12–13 year olds in 74 percent, where again the differences among the results are not statistically significant (H=3.536, 3 d.f., p=0.31608).

The V+V condition is the only one where we can find a statistically significant difference in the interpretation patterns among the four groups (H=20.868, 3 d.f., p<0.001). When paired-groups comparisons are analyzed, we find that the only group whose results vary from all others is that of the 8–9 year olds. It is an interesting fact that what seems to be the construction that triggers the largest numbers of non-ambiguous readings (favoring embedding) in the adult grammar does not have the same status in the grammar of 8 and 9 year olds. Although we do not have enough data from younger speakers to make such claims, these results seem to agree with the hypothesis presented by Roeper (Reference Roeper2011) and Terunuma and Nakato (this volume) about a possible two-step (or three-step) development in the acquisition of recursive constructions from more conjunctive interpretations to more recursive ones.

5 Conclusion

In this chapter we presented the structure of relative clauses in Wapichana that are formed by the morpheme ‘–uraz’ in multiple embedded relative constructions. Relative clauses in Wapichana can be headed by different lexical categories (verbs, adjectives and postpositions), and when more than one relative clause is used in the same sentence, there could be potential ambiguities regarding the noun that is being modified by the second clause, as shown in Section 2.2.

We also provided the results of a comprehension experiment we used to test attachment preferences in sentences with more than one relative clause. The experiment used required participants to select a picture that depicted situations in which the second relative clause could be modifying one of the two available preceding nouns. Our results showed that whenever the first relative clause is headed by a verb, there is an overall preference on the part of speakers to interpret the second relative clause as modifying the embedded noun adjacent to it (what we called the “embedded reading”). In cases where there are two verbless relative constructions (P+P), participants did not provide any consistent preference for attachment patterns.

As far as we know, this is the first time that the interpretation of sentences with two relative clauses has been tested. Therefore, we could not compare the results for Wapichana with those of other languages. For future work, it would be interesting to run similar experiments in Indo-European languages with relative pronouns to see if any attachment preferences exist in those languages.

1 Properties of Relative Clauses in Karitiana

Karitiana has been described as an ergative language (Landin Reference Landin and Dooley1984). Ergativity is unmarked in arguments, but is evident in verb morphology − the intransitive subject and the direct object agree with the verb in main clauses – and in wh-extraction strategies, which are different from ergative and absolutive arguments (Storto Reference Storto1999). Storto (Reference Storto1999) described Karitiana as a language in which transitive main clauses are verb-initial (VOS and VSO) or verb-second (SVO or OVS) and the verb is inflected with agreement, tense, or mood affixes, whereas embedded clauses are verb-final (SOV and OSV) and have bare verbs, without person agreement, tense, or mood morphology.Footnote ² There are no overt nominalizers in embedded clauses in general, except in clauses that serve as complements of the copula in copular and cleft sentences.Footnote ³ There are no relative pronouns or complementizers in Karitiana relatives.

Since aspect is the only functional head that may project above the verb phrase in embedded clauses, Storto (Reference Storto1999) considers them to be aspectual phrases (AspPs). The structure of a subject relative clause is given in Figure 13.1.

Figure 13.1 Subject relative clause

Karitiana relative clauses have typological characteristics of both externally headed and internally headed clauses (Culy Reference Culy1990; de Vries Reference Vries2002). The former is true because the head of a relative clause in natural production in the language is always fronted (Storto Reference Storto1999), which is not common in languages with head-internal relatives. When a transitive relative clause has a subject as its head, the order is SOV (as in (3) and in Figure 13.1), and when the head is the object, the order is OSV (as in (4) and in Figure 13.2). In the latter, the verb is prefixed by the object focus (OFC) morpheme {ti–}, obligatory in object relatives as well as in focus and wh-extraction of objects (Storto Reference Storto1999).

Figure 13.2 Object relative clause

According to Storto (Reference Storto1999), the behavior of internally headed relatives can be seen when case-marking on the head is related to the embedded verb instead of the main verb. Indeed, we know from the literature that in some languages, case-marking on the head is a main piece of evidence to distinguish between head-internal and head-external relatives. This can be illustrated by the following data from Ancash Quechua. This language has both internally and externally headed relatives and, when the head is external, the case-marking on the head is related to the main verb. In internally headed relatives, as in (2), the head is marked with the case morphology demanded by the embedded verb:

If sentences (3) or (4) were head-external relatives, they could have the external head taso marked with the oblique case {–ty} required by the main verb so’oot ‘to see’ when it is used with an object,Footnote ⁴ but this is not what we find. Instead, the oblique suffix applies to the whole clause:

Figure 13.3 Simplified tree (without tense, agreement, and mood) of (3)Footnote ⁵

We are aware that this argument is not conclusive, since languages do not necessarily mark the head of an external relative with case. Also, oblique marking is not structurally case assigned by the main verb. Nonetheless, the verb has an oblique argument and this argument is the whole clause, which is compatible with a head-internal analysis.

Fronting of the head in relatives is used by speakers in their production, possibly to avoid ambiguity between subject and object relatives, but we have recently discovered, through experimental means, that it is possible for speakers to produce non-fronted heads as well (Vivanco Reference Vivanco2014). We conclude that, since externally headed relatives cannot have non-fronted heads, Karitiana relatives should be analyzed as internally headed.

In Section 1.1 we show in greater detail the results of the experiments that corroborate the head-internal analysis of relative clauses.

1.1 Vivanco (Reference Vivanco2014)

In a task inspired by Labelle’s (Reference Labelle1990) experiment, Vivanco’s (Reference Vivanco2014) fourteen Karitiana subjects had to choose one of two pictures presented to them, and then they had to tell the researcher, using a Karitiana sentence, which of the pictures they had chosen. These pictures depicted two identical characters or objects that could only be differentiated through the action of another element in the scene:

(5)	Researcher: Here we have two T-shirts. Ana sewed this one over here [point to the one on the left] and Luciana sewed this one over there [point to the one on the right]. Pick one of the T-shirts and tell me which one you chose.
	Subject: [intended production] ‘I chose the T-shirt that Ana/Luciana sewed’.

Figure 13.4 Context for an object relative

Altogether, there were twenty contexts suited for the production of relative clauses (ten for subject and ten for object relatives).

The results show that, although there is a strong preference for having the head of the relative at the beginning of the clause, it is also possible to have non-fronted heads. In subject relatives, the majority of the data obtained consist of SOV relatives, but OSV subject relatives were also produced. Examples of each structure are presented in (6) and (7):

(6)	Subject relative with SOV word orderFootnote 6

Yn	Ø-na-aka-t	i-pyting-Ø	[taso	him	by-hip]-i-ty
I	3-DECL-cop-nfut	nmlz-want-abs.agr	[man	meat	caus-cook]-ep-obl
‘I want the man who cooked the meat.’

(7)	Subject relative with OSV word order

Yn	Ø-na-aka-t	i-pyting-Ø	[opi	jõ nso	by-’it]-i-ty.
I	3-DECL-cop-nfut	nmlz-want-abs.agr	[earring	woman	caus-do]-ep-obl
‘I want the woman who made the earring.’

Figure 13.5 Word order distribution in subject relatives (n = 115)

The word order variation observed in the experiment as a whole was even greater in object relatives. The formerly reported OSti-V word order was the preferred structure, but other possibilities were also produced: a non-initial head (SOti-V), a verb without the OFC morpheme (OSV), and the latter two conjugated (SOV). In the following are examples of these four sentential types produced by speakers:

(8)	OSti-V object relative

Yn	Ø-na-aka-t	i-pyting-Ø	[gijo	Luciana	ti-tak]-a-ty
I	3-DECL-cop-nfut	nmlz-want-abs.agr	[corn	Luciana	ofc-grind]-ep-obl
‘I want the corn that Luciana ground.’

(9)	SOti-V object relative

Yn	Ø-na-aka-t	i-pyting-Ø	[Ana	pykyp	ti-pipãrama]-ty
I	3-DECL-cop-nfut	nmlz-want-abs.agr	[Ana	clothes	ofc-sew]- obl
‘I want the clothes that Ana sewed.’

(10)	SOV object relative

Yn	Ø-na-aka-t	i-pyting-Ø	[Ana	gok	amangã]-ty
I	3-DECL-cop-nfut	nmlz-want-abs.agr	[Ana	manioc	plant]-obl
‘I want the manioc that Ana planted.’

(11)	OSV object relative

Yn	Ø-na-aka-t	i-pyting-Ø	[ambi	taso	by-’a]-ty
I	3-decL-cop-nfut	nmlz- want-abs.agr	[house	man	caus-do]-obl
‘I want the house that the man built.’

Figure 13.6 Word order distribution in object relatives (n = 103)

Vivanco concludes that, since externally headed relative clauses cross-linguistically cannot have their heads in other positions besides the left or right periphery of the clause, it makes sense to classify Karitiana relative clauses as internally headed. Head frontalization in Karitiana seems to be similar to that described in Yuman languages (Gorbet Reference Gorbet and Gorbet1976; Basilico Reference Basilico1996) in that it is not an essential operation for relative clause formation, but is employed in relative clauses that are ambiguous, in order to make them more explicit.

Note that in this chapter we work solely with the clauses that speakers use in natural production, i.e., SOV for subject relatives and OSti-V for object relatives. We choose to do so because we are not certain of the reasons why other word orders arose in Vivanco’s experiments. One of the sources of concern is that all of Vivanco’s subjects were animate and her objects were inanimate, which unambiguously identifies each argument in her relatives. We would like to know, for instance, whether the variation in word order would also occur if both subjects and objects were proper names of humans, a condition in which ambiguity arises. Until the conditions for the variation found by Vivanco are better understood, we prefer to work with the orders that are naturally produced.

2 Constituent Order and Clause Structure in Kotiria

2.2 Oblique Heads of Oblique Relatives

Note that sentences with oblique relative clauses such as (33) may alternatively be realized with an oblique suffix on the head of the relative co-occurring with the oblique suffix at the end of the clause:

(33)	Y-py-so’ oot-y-n	yn	[kinda ‘o]-ty	[Maria	ti-amangã]-ty
	1-assert-see-ep-nfut	I	[fruit]-obl	[Maria	ofc-plant]-obl
‘I saw the fruit that Maria planted’

One might think that this case is a juxtaposition or conjunction between a phrasal oblique argument kinda ‘o and a clausal oblique argument (Maria tiamangã) with a null object. This would be a plausible analysis because we know independently that null heads are allowed in free relatives with an indefinite reading (as in (34)):

(34)	Y-py-so’ oot-y-n	yn	[Maria	ti-amangã]-ty
	1-assert-see-ep-nfut	I	[Maria	ofc-plant]-obl
	‘I saw what Maria planted’

However, if the juxtaposition or conjunction analysis were correct, the order between the two oblique noun phrases could be exchanged, and this is not the case as a comparison between (33) and (35) clearly shows:

(35)	*Y-py-so’ oot-y-n      yn [Maria ti-amangã]-ty      [kinda ‘o] ty 1-
	assert-see-ep-nfut I     [Maria ofc-plant]- obl       [fruit]-obl
	‘I saw the fruit that Maria planted’

If the structure is not a juxtaposition or conjunction, it is reasonable to analyze it as a regular case of extraction of the head to the left edge of the relative clause. The fact that the object focus prefix is present in other cases of extraction to the left periphery of the clause (object focus and wh-movement) gives support to this analysis. In intransitive verbs, the analysis is confirmed by other examples:Footnote ⁸

(36)	Y-py-so’ oot-y-n	yn	[Francisco	ga-p	erery	hoop]-o-ty
	1-assert-see-ep-nfut	I	[Francisco	field-in	cotton	grow]-ep-obl
	‘I saw the cotton that grew in Francisco’s field.’

(37)	Y-py-so’ oot-y-n         yn [erery]-ty   [Francisco      ga-p   hoop]-o-ty
	1-assert-see-ep-nfut I     cotton-obl     Francisco      field-in grow-ep-obl
	‘I saw the cotton that grew in Franscisco’s field.’

The adjunct PP Francisco gap is interpreted inside the relative and the moved head occurs before this adjunct, at the left edge of the relative (in Spec, AspP).

At this point, there is an issue that needs further clarification. Given our discussion about the status of Karitiana relative clauses as internally headed, one could wonder whether the clause in (33) could be analyzed as an externally headed relative clause because of the oblique marking on the head. However, relative clauses with the corresponding template of (38) are ungrammatical, indicating that the oblique marking in the whole clause is a requirement of the main verb:

(38)	*Yn   Ø-na-aka-t              i-so’oo-t               ombaky-ty taso ti-oky
	I        3-decl-cop-nfut   nmlz-see-abs.agr.  jaguar-obl man ofc-kill

We conclude that cases like (33) are not externally headed relative clauses but internally headed relatives whose heads are redundantly marked as oblique.Footnote ⁹ Head-external relatives do not seem to exist in Karitiana at all, given Vivanco’s (Reference Vivanco2014) findings about the possibility that heads may occur internally to the relative clause, which does not occur in externally headed languages. Possibly, this redundant case-marking is a strategy to identify the head: since internally headed relative clauses usually have all their elements in situ, some languages employ mechanisms to make their heads more explicit. One common strategy reported in the literature is the frontalization of the head, but other operations such as intermediate movement (Gorbet Reference Gorbet and Gorbet1976; Basilico Reference Basilico1996) and head duplication (Kendall Reference Kendall1974; Moore Reference Moore2006) have also been identified.

To conclude this section, we present cases of multiple embedding in which the same extraction of the head of the relative occurs:

(39)	Y-py-so’ oot-y-n            yn erery-ty          Maria pytagngãm-ty   João sondyp-y-ty
	1-assert-see-ep-nfut I cotton- obl     Maria steal-obl           João know-ep-obl
	‘I saw the cotton that João knows that Maria stole.’

In all such cases of multiple embedding, the oblique head of the oblique relative clause must be at the left periphery, whereas each clause marked as oblique may occur in any order:

(40)	Y-py-so’ oot-y-n         yn [erery]-ty  [João sondyp]-y-ty [Maria pytagngãm]-ty
	1-assert-see-ep-nfut   I   cotton-obl João know-ep-obl    Maria steal-obl
	‘I saw the cotton that João knows that Maria stole.’

In the case of (39) and (40), this must follow from the fact that both the verb sondyp and pytagngãm are intransitive and thus require oblique clausal arguments, as in (41):

Figure 13.7 Simplified structure (without tense, agreement, and mood) of (41)

Furthermore, some oblique markers can be dropped as in (42) and (43):

(41)	Y-py-so’ oot-y-n              yn [erery]-ty   [Maria pytagngãm]-ty [João iip]-i-ty
	1-assert-see-ep-nfut I   cotton-obl     Maria steal-obl                    João say -ep-obl
	‘I saw the cotton that João said that Maria stole.’

(42)	Y-py-so’ oot-y-n       yn   erery-ty   Maria pytagngãm-ty        João iip
	1-assert-see-ep-nfut  I        cotton-obl Maria steal-obl           João say
	‘I saw the cotton that João said that Maria stole.’

(43)	Y-py-so’ oot-y-n         yn erery Maria pytagngãm-ty João iip
	1-assert-see-ep-nfut   I   cotton Maria steal-obl        João say
	‘I saw the cotton that João said that Maria stole.’

It is not clear whether there are subtle differences in meaning related to these variations of the same sentence. In order to certify that all the variants have the same meaning, an experimental task was carried out with three speakers.

3 Experimental Task with an Oblique Relative

In this task, a story illustrated by four drawings was presented to the speaker, one drawing after the other on a computer screen. By the end of the story, the speaker would see all four drawings at once. When a drawing was being displayed, the researcher described in Portuguese the situation depicted (the sentences used in each context are included below, translated into English). Next, the researcher would read a multiple-embedded sentence in Karitiana from the computer screen and ask the speaker if the sentence was appropriate to describe that particular context. If the sentence was not appropriate, the speaker could correct the sentence. The whole task was recorded either in audio or in video. In the following, we show one of the contexts used:

(44)	Researcher:	“In this story, Luciana cooked porridge.”
	“Thiago saw Luciana cooking the porridge.”
	“They put the porridge outside to cool it.”
	“Then a tapir ate the porridge”

After this step, the speaker was presented with a Karitiana sentence on a computer screen. The multiple-embedded sentence displayed on the screen for this particular context is transcribed in (46); it is ungrammatical:

(45)	Researcher:
[reads the sentence aloud] “Can you say that in this context?”

In this example, the main verb is transitive, so we do not expect an oblique argument at the end of the complex relative clause, but the verb ‘to see’ requires an oblique object. Therefore, we would expect speakers to correct this sentence in that respect. Three native speakers were tested and their judgments were collected and compared. Sentence (46) was corrected by speakers so as to include a type of infinitive suffix {–p} at the end of the clause, translated as ‘the porridge that Thiago saw’:

This is not what was expected by the authors, but it was consistent for the three speakers. To be able to compare these results with those of the other two experiments, researchers changed all multiple-embedded clauses presented so as to include that same suffix. Speakers only differed in the order in which they preferred each embedded clause to be pronounced. The head of the relative was always given at the left edge. No oblique suffixes were allowed.

In the second context, we had the following procedure:

(48)	Researcher: “In this story, Luciana cooked porridge.”
	“Thiago saw Luciana cooking the porridge.”
	“They let the porridge finish cooking.”
	“Then a child bought Luciana’s porridge.”

The sentence given to speakers in this case was (49), in which the main verb amy ‘to buy’ is intransitive and may have an oblique object.

The results showed that the oblique suffix may appear only once, at the end of the complex relative as in (49), or twice, at the end of each clause. The head of the relative was only produced at the left edge of the complex relative clause, with (two speakers) or without the oblique suffix (all speakers). If the oblique head was allowed at all at the end of the clause (by one speaker), it was clearly pronounced as an afterthought (after a pause). In this experiment, we have confirmed that the oblique suffix required by the main verb is being repeated at the end of each clause and after the relative’s head as an alternative to the same meaning.

In the third context, the procedure was the following:

(50)	Researcher:	“In this story, Luciana cooked porridge.”
		“Thiago saw Luciana cooking the porridge.”
		“They put the porridge outside to cool it.”
		“Then a child ate Luciana’s porridge and liked it.”

The sentence given in the third experiment was (51):

In this sentence, the main verb is intransitive and its oblique object occurs at the end of the sentence. Results were consistent with what was found in sentence (49), which has the same structure.

The experiment was able to confirm the facts observed in the data presented in this chapter: that when one oblique relative clause is the object of an intransitive verb, the head at the left edge of the relative may be marked as oblique, as well as any intermediate clauses without a change in meaning. Even though in this particular task speakers used a different type of relative clause with a poorly understood infinitival suffix on the first embedded verb, the behavior of such clauses with respect to the oblique suffix was exactly the same as that observed in the other examples. Because of this, we consider that such clauses have the same structure as the others, given in Figure 13.8.

Figure 13.8 Simplified tree (without tense, agreement, and mood) of (51)

4 Final Discussion

Our analysis of embedding in the cases under consideration is that the oblique suffix that appears at the end of the relative is required by the main verb when an object is added to its argument structure, and that the option of extracting the head to the front in such cases is always realized in natural production to avoid ambiguity between subject and object relatives. When the extracted head has an oblique suffix in addition to the oblique case suffixed to the relative, this is a strategy the language uses to keep track of the referent of the relative, marking it as oblique because it is the head of an oblique relative clause. When more layers of embedding are added, the same strategy is maintained. It is a result of successive-cyclic movement of the head to the left periphery of intermediate clauses, but it is not agreement in the complementizer position as reported in other languages of the world for two reasons: it is not obligatory and, in our analysis, Karitiana does not project complementizers in embedded clauses.

The addition of more embedded clauses makes the relative even more ambiguous, because they add more NPs that could be the head of the oblique relative clause. Some languages employ various strategies such as frontalization or duplication to make their head more explicit. As seen before, Karitiana already has optional movement of the head, so the language must be using an additional strategy – case-marking of the head – as a device for the same purpose. We do not know yet how to explain the syntactic implementation of the repetition of the oblique suffix in the head and in every intermediary clause in such cases, but it is possible to hypothesize that it is agreement of a non-obligatory type in the aspectual head of the embedded clause, because to raise to Spec, AspP, the head of the relative probably stops by every intermediary Spec, AspP projection (successive-cyclic movement).

Finally, it is worth saying that the Karitiana relative clauses discussed here shed light on the topic of clause recursion in Amerindian languages. We have seen that Karitiana not only provides evidence that it is capable of embedding one clause into another, but that it also displays a more complex case of recursion.