BONILHA, Giovana. Conjoined constraints and phonological acquisition. Journal of Portuguese Linguistics, v. 2, n.2, 2003. Conjoined Constraints and phonological acquisition

Since the start of Optimality Theory (Prince & Smolensky, 1993), research on phonological acquisition has explored the explanatory potential of constraint theories. This study, also based on Optimality Theory, attempts to analyze the acquisition of CVVC syllable structure by Brazilian Portuguese children and addresses the issue of Local Conjunction (Smolensky, 1995, 1997) in research that deals with problems of phonological acquisition.


Introduction
Since Local Conjunction was proposed by Smolensky (1995Smolensky ( , 1997)), it has been basically used to analyze phonological aspects of the adult grammar.According to McCarthy (2002), local conjunction is a powerful mechanism.We need restrictions on conjunction as well as a better understanding of its function.
In this paper I deal with the use of conjoined constraints in phonological acquisition, arguing that these are "potential constraints", activated by the child in the process of learning his language.A proposal is made for the constraint hierarchies that play a role in syllable acquisition by Brazilian Portuguese children, following Bonilha (2000).Subsequently, the mechanism of conjoined constraints will be discussed and it will be argued that the way in which Local Conjunction functions in child language contains a suggestion about the way it may be properly restricted.
(3) H O =Onset, *ComplexNuc, NoCoda>>MAX I/O, DEP I/O In the initial hierarchy, markedness constraints dominate faithfulness constraints, which ranking automatically leads to the production of CV syllables.According to Matzenauer (1999), Onset is the first constraint related to syllable structure that needs to be demoted in the acquisition of BP, in order to account for the production of V syllables next to CV syllables.The question that must be answered is how the child knows that, to produce a V type target form, it must demote the Onset constraint?According to Tesar & Smolensky (2000), by analyzing sub-optimal/optimal pairs created by GEN, the algorithm starts the demotion of the relevant constraints, until the hierarchy delivers the optimal candidate.The number of informative pairs used for the analysis can show the complexity of a structure.More complex structures will probably make it necessary for a larger number of pairs to be analyzed, as a larger number of constraint demotions will be necessary to account for the target forms, as is shown in Table 1: The analysis of sub-optimal/optimal pairs shows which constraints are violated by the loser and the winner candidates.If a constraint is violated by both elements in a pair, the process of mark cancellation is applied.This process is used to 'purify' the information by cancelling out marks from the table which have no informative value, so it will eliminate from the table any marks which are shared by the winner and the loser, because what is relevant is the difference among candidates in terms of constraint violations.The relevant constraints will be demoted only after markedness cancellation has taken place.In Table I, no constraint is violated by either member of the pair; therefore, no violation mark will be deleted.Hence, the process of constraint demotion can start.According to Tesar & Smolensky (2000), in this situation the constraint ranking must be adjusted, so that for each pair of candidates analyzed, all the constraints violated by the potentially optimal candidate be dominated by at least one constraint violated by the sub-optimal candidate.Crucially, only constraint violations are relevant for the demotion process, as constraint satisfaction on the potentially optimal candidate is unable to reflect its position in the hierarchy.
By analyzing the informative pair a<c, one observes that Onset must be dominated by DEP I/O in order for candidate c to be selected as optimal.The hierarchy in (4) shows the first acquisition stage in Brazilian Portuguese.

Acquisition stage II
As was suggested in (1), for the learner to reach the second stage of syllable structure acquisition it is necessary that a new analysis of informative pairs be carried out.It is shown in ( 4) that, at the first acquisition stage, only non-branching nuclei are produced.Therefore, when at this stage the child faces a lexical target such as papai [pa.'paj] 'daddy', the optimal candidate will be chosen according to the constraint ranking of this stage of acquisition.Observe the tableau in ( 5 According to (5), the second candidate is chosen as optimal, because it does not violate *ComplexNuc, which is ranked above the faithfulness constraints.Although there is no ranking among the faithfulness constraints at this stage of development, the acquisition data in BP point to the existence of a subhierarchy involving Dep I/O and Max I/O, since children systematically prefer deletion instead of insertion.
The ranking proposed in (5), however, only allows the production of non--branching nuclei.The learning algorithm will guide the learner when he is confronted with new informative pairs towards achieving the ranking proposed in (6).The application of the Algorithm will then lead to the demotion of *ComplexNuc below the faithfulness constraints.The hierarchy in (7) shows the second stage in the acquisition of syllable structure in BP, when the child starts producing falling oral diphthongs.Taking into account that in (7) the production of falling diphthongs occurs due to the demotion of *ComplexNuc, the child can produce CV, V and VG syllable structures only by Onset and *ComplexNuc demotion.

Acquisition stage III
The demotion of *ComplexNuc in stage II results in the fact that the unmarked VG structure is acquired before the relatively marked VC structure, if the NoCoda constraint is demoted in H 3, as shown in ( 8).The hierarchy in (8) predicts that, when the learner arrives at stage III, he is able to produce a CVVC syllable structure 2 , i.e. a complex nucleus followed by a coda consonant.Let us next observe the tableau 03 in ( 9 According to (9), candidates a and b are not chosen as optimal, because they violate MAX-I/O, which is ranked above the markedness constraints.
2 Considering different views in the literature dealing with the syllable position of the fricative /s/ in words such as mais 'more', dois 'two', perspectiva 'perspective' solstício 'solstice', we agree with Lee (1999) and Cristófaro Silva (1999) that the fricative /s/ is in a syllable coda position in BP.
The candidate seis ['sejs] 'six' is selected as the best candidate, because it only violates *Complex(Nuc) and NoCoda, which are ranked below the faithfulness constraints.
According to the data, however, a CVVC syllable structure is not usually acquired in the acquisition stage III, witness its low production percentage.
Observe Table II Considering the fact that H 3 allows the production of syllables containing simultaneously a complex nucleus and a coda, a production percentage of the CVVC pattern higher than 58.3% is expected. 3n OT account of the acquisition of syllable structure cannot be given with the constraints proposed in (2) alone.Bonilha (2000) proposes a Local Conjunction of the constraints *Complex(Nuc)&NoCoda4 , which requires that syllables should not have a complex nucleus and a coda at the same time.The mechanism of Local Conjunction proposed by Smolensky (1995Smolensky ( ,1997) ) permits the conjunction of two simple constraints into one, including the con-junction of a constraint with itself (or local self-conjunction), which may not be violated in a specific domain.
According to McCarthy (2002), the possibility of conjoining constraints somewhat loosens the effects of the strictness of strict domination.Nevertheless, he points out that the conjoined constraint is also ranked with regard to the simple constraints that it contains and the conjoined constraint is also defined to apply within a local domain.
Following Fukazawa (2001), we illustrate the mechanism of Local Conjunction with the abstract examples in (10-11): Because C is ranked higher in the hierarchy in (10) than A and B, it cannot be violated by the optimal candidate, which can violate A or B. In (11), on the other hand, by virtue of the locally conjoined constraint, the optimal candidate can violate C, because the conjoined constraint [A&B] D cannot be violated5 .When A and B are both violated by a candidate, the conjunction of these two violations may force the violation of constraint C.
In the hypothetical languages represented by the hierarchies in ( 10) and (11), the constraints A and B can be violated separately, satisfying constraint C, but both of them cannot be violated at the same time.
Many questions have been raised with regard to the usage of conjoined constraints since they were first proposed: (i) Do Local Conjunction constraints form part of UG? (ii) How is locality established?(iii) Is it possible to conjoin only constraints that belong to the same family?(iv) How must the word 'family' be understood?What regard to the question in (i), Smolensky (1997), Fukazawa and Miglio (1998), and Fukazawa (1999, 2001) suggest that conjoined constraints are language-specific.According to these authors, if these constraints belong to UG, the universal set of constraints would be tremendously enlarged.They propose therefore that UG only contains the "&" operator, which allows conjoining constraints.According to Fukazawa and Miglio (1998), this sug-gestion is corroborated by the rareness of each particular type of local conjunction throughout the world's languages.Smolenky (1995Smolenky ( , 1997) ) suggests in relation to (ii) that conjoined constraints can only be created when violated in the same domain.Equally, Nathan (2001) assumes that the issue of locality is at the heart of constraint conjunction.
As for the question raised in (iii), Fukazawa and Miglio (1998) observe that, if there are no limitations on the types of constraints that can be joined, even if UG only provides the "&" operator, the grammar of any given language could still be enlarged tremendously as a consequence of the conjunction of constraints.The authors propose that only constraints which belong to the same family may be conjoined.Fukazawa (1999:216) criticizes Itô and Mester (1996) for using the NoCoda & *Voice constraint, showing that devoicing in German can be explained by using single constraints.Moreover, the proposed conjunction does not obey the 'same-family' restriction.However, it is possible to interpret the conjoined constraints used by Itô and Mester (1996) as belonging to the same family, if a broader definition of the concept of 'family' is adopted, such that three large families are distinguished among the constraints that are part of UG: markedness, faithfulness and alignment constraints.In other words, depending on the definition of what a constraint family is, constraints such as Dep [+ ATR] and Dep [Hi] refer to the more restricted family of Dep constraints, or to the much larger family of faithfulness constraints.To conclude, although Fukazawa and Miglio (1998) argue for the existence of a restriction in the constitution of Local Conjunction suggesting that both constraints must be of the same family, the issue regarding the proper definition of the concept of 'constraint family' remains unsolved.
According to Suzuki (1998:41), although the analyses that use conjoined constraints point to the existence of some restriction on constraint conjunction that has to do with family membership, the uncertainty about exactly which constraints should be allowed to be conjoined still remains an open one.It is therefore fruitful to reflect upon the way in which conjoined constraints emerge in the grammar, since it may be hypothesized that by understanding this mechanism it may be possible to gather some insight into the question which constraints may be conjoined.
For Fukazawa (1999Fukazawa ( , 2001) ) and Fukazawa and Miglio (1998), a conjoined constraint can only be used as a last resort, that is, when an analysis using single constraints is not able to explain a certain phonological generalization.More specifically, constraint conjunction must obey the following three restrictions: last resort, locality, and family membership.
In the following, the conjoined constraint *Complex (nucleus) & NoCoda suggested by Bonilha (2000) will be discussed from the perspective of these three restrictions.
From the tableau in ( 9), which presents the ranking of the single constraints in H 3 : MAX I/O, DEP I/O>>NoComplex (nucleus), Onset, NoCoda, it is impossible to account for the production of CVV and CVC syllables and the non-occurrence of CVVC syllables.The use of the conjoined constraint, therefore, is in accordance with the requirement that no ranking of single constraints can account for the grammatical outputs.Observe next the tableaux in ( 12) and ( 13): (12) Tableau 04 The ranking shown in (12), Max I/O >> *ComplexNuc, NoCoda, although it is capable of accounting for the production of CVV and CVC syllables, selects as the optimal output for a CVVC input the candidate in c, which is not produced by the child at this stage of acquisition.On the other hand, the ranking proposed in (13) -*ComplexNuc, NoCoda >> Max I/Ocan account for the non-production of CVVC.At this stage of acquisition, the selected outputs CVV and CVC are really the forms produced by the children for a CVVC input.However, the problem with this ranking is that it favors the candidates that violate Max I/O constraints for CVV and CVC inputs, although the children already produce those structures in conformity with the target form.The use of *Complex (nucleus) & NoCoda is able to account for the production of words such as pai 'father' and paz 'peace' in accordance with the target form.It also predicts adequately the non-production of words such as mais 'more' and seis 'six' at the same stage of acquisition.
Another issue to be observed is the fact that the constraints that constitute *ComplexNuc&NoCoda belong to the same constraint family, obeying, therefore, the family membership requirement proposed by Fuzakawa & Miglio (1996).As a matter of fact, the constraint proposed by Bonilha (2000) is in agreement with both the large and the narrow definition of constraint family, the narrow definition referring to the family of syllable structure constraints and the broader one to the family of markedness constraints.
Regarding the locality restriction, the tableaux in (14-15) show that, if no local domain were defined, the conjoined constraint would make the wrong predictions in some cases: According to the tableau in ( 14), the child would not only be unable to produce a CVVC syllable but also the words that have a complex nucleus and a coda in different syllables.The attribution of locality to the *ComplexNuc&NoCoda constraint in tableau (7) restricts the non production of a complex nucleus and a coda to the same syllable, not to the prosodic word as a whole, as was evidenced throughout the data that were used.
The proposed ranking in (15) shows the third stage of BP language acquisition.Observe the tableau in ( 16 16), accounts for the correct production of words that have CV, V, CVV and CVC syllable structures and also for the non--production of a CVVC syllable structure to the extent that these occur in the target forms.For a CVVC input, this grammar would not select candidate c as the optimal one, because it violates *ComplexNuc&NoCoda () which is ranked higher than MAX I/O, violated by candidates a and b.
The ranking in (16) shows how local conjunction functions in the learner's grammar.*ComplexNuc and NoCoda are violated by the production of CVV and CVC outputs, respectively.In order to produce a CVVC output, these constraints would have to be violated simultaneously.Given the ranking in ( 16), an output form that violates Max I/O instead of the conjoined constraint is still better.The deletion of the glide or the coda consonant is therefore motivated by the high ranking of [*ComplexNuc & NoCoda] () in the H 3 hierarchy.At this stage of acquisition, the simultaneous violation of the syllable structure constraints *ComplexNuc and NoCoda is much worse than the violation of just one of them at the time, even if the violation of one of the simple constraints implies the violation of a faithfulness constraint.Observe that the hypothetical example suggested in ( 10) is here evidenced by acquisition data.It is worth mentioning that the grammar in ( 16) selects for a given CVVC input not one but two candidates as equally optimal: candidates a and b.This state of affairs suggests that there is variation in the learner's production.However, the data provided by Bonilha (2000), as summarized in Table III below, show that a CVV structure is systematically produced, not a CVC structure.The preference for the CVV syllable structure as shown in Table III can be explained by ranking NoCoda higher than *ComplexNuc.

Acquisition stage IV
In order to explain the acquisition of a CVVC syllable structure type it is necessary to return briefly to the learning algorithm proposed by Tesar & Smolensky (1996).First, we will consider, in table IV, for each element of an informative pair created by GEN which constraints are violated: The markedness cancellation process eliminates the constraints violated by both pairs ses<sejs and sej<sejs, as is illustrated in table V: Considering that, after markedness cancellation has applied, no other constraint is violated by both members of the pairs, the process of constraint demotion can apply.Observe that [*ComplexNuc&NoCoda] () does not constitute a new stratum in the hierarchy, because the informative pair analyzed does not deter-mine that this constraint must be ranked below Onset, *ComplexNuc and NoCoda.[*ComplexNuc& NoCoda] () .Therefore, it is ranked in the highest possible stratum of the hierarchy.
The analysis of the pair b < c will not cause any alterations to the H 4 hierarchy.It is a non-informative pair.The analysis of b < c shows that NoCoda and [*ComplexNuc&NoCoda] (), which are violated by the optimal candidate, must be demoted below MAX I/O, which is violated by the sub-optimal candidate.As can be seen, the current learner's hierarchy H 4 already demonstrates this ranking.
It is essential that, even by changing the order of the proposed pairs in Table IV, the H 4 hierarchy will be obtained by taking into consideration the analysis of only one of the pairs, since all of the analyses imply the demotion of the constraint [*ComplexNuc&NoCoda] () below the MAX I/O constraint.If the b < c pair is analyzed first, the child finds out that NoCoda and [*ComplexNuc&NoCoda] () must be demoted below MAX I/O, which also leads to the ranking in H 4 .The only difference is that the pair b < c would become an informative pair.Consequently, it is seen here that the classification of a pair as informative or non-informative may depend on the order in which the relevant pairs are analyzed.Sometimes, however, the change in the order in which the pairs are analyzed does involve changes in the sequence of the provisional hierarchies that the learner goes through.According to Kager (1999), the difference in the order of analysis of the pairs will lead to an increase or decrease in the time span used by the learner to achieve a certain structure; however, this difference will not change the final hierarchy. 6n line with Bonilha (2000), the different hierarchies that account for the stages of acquisition in which the CV, V, CVV, VC, and CVVC structures gradually emerge, are shown in Table VII.

Alternative analyses
According to Fukazawa & Miglio (1998), the use of conjoined constraints should be restricted to those cases where no other possible ranking can account for the occurrence of a given phonological phenomenon.The tableaux in ( 12) and ( 13) show the incapability of the syllable structures constraints to deal with the presence of CVV and CVC syllables and the simultaneous absence of CVVC syllables at a given stage of acquisition.One question that must therefore be asked is whether it is possible to account for the same pattern by the interaction of a set of constraints that does not contain the conjunction [*ComplexNuc&NoCoda] () .One alternative analysis would give a prominent role to the constraints Binary (,) and Singly Linked () 7 .The former constraint requires that a syllable has two moras.The family of Singly Linked constraints requires that each element be linked to an element on a higher tier.As such, the second constraint prohibits the sharing of a mora by two syllable elements.Consider the hierarchy in ( 18): (18) H 3 {Binary (,), Singly Linked () }>>{Dep I/O, Max I/O}>>{Onset, *ComplexNuc, NoCoda} Given the hierarchy in ( 18) with the high ranking of Binary (,) and Singly Linked (), for a CVVC input, such as sejs 'six', the outputs defined as optimal would be CVV or CVC structures.For the forms that contain a CVVC syllable, the form [sejs], whether presenting three moras, each of them linked separately to one of the three segments in the syllable rhyme, or just two, one of them linked to both the vowel and the glide, would be eliminated for violating either Binary (,) or Singly Linked().In order to produce a CVVC syllable, it is necessary that at least8 Singly Linked() be demoted below the faithfulness constraints in the hierarchy in ( 19), which would then account for the fourth acquisition stage of syllable structure in BP.What is the evidence against the grammars represented in ( 18) and ( 19) as a way of accounting for the acquisition of syllable structure in BP? First, it is not obvious that one should use constraints referring to mora structure in stages of acquisition in which reference to syllable structure seems enough to account for the data.According to Levelt, Schiller & Levelt (2000), other syllable types that are found to be acquired late, such as VCC, CCVC and CCVCC clearly do not all involve complex rhymes.Research carried out by these authors based on 12 Dutch children showed that the order CV, CVC, V and VC reflects the order of acquisition for all analyzed subjects.From stage IV, the subjects divide into two groups with regard to the acquisition of more complex syllable structures9 : (20) Group A CVCC, VCC, CCV, CCVC and CCVCC; Group B CCV, CCVC, CVCC, VCC and CCVCC.Levelt, Schiller & Levelt (2000) use the mechanism of local conjunction, proposing the following conjoined constraints: [*ComplexOnset&*ComplexCoda], and [*ComplexCoda&Onset].We conclude that there are independent reasons to appeal to constraint conjunction in order to account for the acquisition of complex syllable structures.If there is a need for conjoined constraints to account for the acquisition of CVCC, VC, VCC and CCVCC structures, why should this constraint type be prohibited to account for CVVC syllable structure?It seems also relevant to mention that the constraint [Onset&NoCoda] used by the authors to explain the late VC syllable acquisition is ranked high in the grammar of the language known as Central Sentani, which admits CVC and V syllables but prohibits VC syllables.A similar typological observation backs up the constraint [*ComplexNuc&NoCoda] () proposed by Bonilha (2000).According to Blevins (1995), the languages Yokuts, Afar and Hausa admit CVC and CVV syllable structures, but lack CVVC syllables.

The use of Local Conjunction: implications for phonological acquisition
Conjoined constraints seem to be necessary as a "last resort" language--specific option.UG can make use of the "&" operator when the interaction between the constraints that are part of UG are unable to provide the grammatical output.In the language where constraint conjunction is active, it cannot be violated and must always be placed higher in the ranking than the simple constraints of which it is made up.If the conjunction [A&B] were ranked below the single constraints A and B, it would lose its function, which is to avoid the simultaneous violation of A and B. I therefore wish to propose that its relatively high positionrelative to the constraints that it conjoinsas well as its inviolability act as restrictions on their functioning.However, the restrictions suggested here seem to create a problem for later stages of acquisition, when the complex structures are eventually acquired and the conjoined constraints are consequently violated and demoted below the simple constraints.This state of affairs may be observed in the analysis proposed by Bonilha (2000).It is also relevant to the analysis by Levelt, Schiller & Levelt (1999), since the [*ComplexNuc&NoCoda] () constraint must be demoted below Dep I/O and Max I/O, in order to free the way for the emergence of the CVVC syllable structure.The conjoined constraint would be positioned in the same constraint stratum as the constraints of which it is made up, losing its main role in the grammar.In Levelt, Schiller & Levelt (1999), the ranking of constraints is postulated without the authors taking a stand with regard to the learning algorithm applied.Nevertheless, the proposed construction of hierarchies shows that Faithfulness constraints are being promoted throughout the phonological acquisition process.For example, in order to explain the third stage of acquisition when the child starts producing V and CV syllables, the faithfulness constraints are ranked higher than Onset.In fact, the faithfulness constraints start to dominate the markedness constraints one by one, accounting for the different stages of acquisition.In the analysis of Levelt, Schiller & Levelt, the promotion of faithfulness constraints ends up guaranteeing that the conjoined constraint remains ranked above the constraints that constitute it.Observe the example in (21), which shows the acquisition of CVCC at a stage before the structure VCC is acquired.Tesar & Smolensky (2000), on which the proposal used by Bonilha ( 2000) is based, the ranking proposed by Levelt, Schiller & Levelt does not position the conjoined constraint in the same stratum where the constraints are located that constitute it.Nevertheless, it is still the case that the conjoined constraint loses its main role, which is to prohibit "the worst of the worst", since there is no violable constraint between the conjoined constraints and its constituting parts.Observe, to see this, the rankings in ( 22) and ( 23 As can be seen in the tableaux in (24-26), the conjoined constraint [*ComplexNuc& NoCoda] () has no role to play in the emergence of CVV, CVC and CVVC syllables in BP, independently of its ordering.For a CVVC input, the optimal output emerges because it does not violate the Max I/O constraint, which is ranked higher than *ComplexNuc and NoCoda.This output would not be selected with the current constraints, if the conjoined constraint were ranked higher than the faithfulness constraint and, therefore, also above the constraints that constitute it.Then, whether [*ComplexNuc&NoCoda] () is positioned above the constraints that compose it, as in Levelt, Schiller & Levelt (1999 ) or is in the same stratum, as in Bonilha (2000), it no longer plays a role in the BP grammar that allows for the production of CVVC syllables.When ranked below the faithfulness constraint, the conjoined constraint becomes a mere repetition of the constraints that constitute it.Accepting the proposal by Bonilha (2000) would imply that the construction of conjoined constraints are primarily motivated by the learner's "difficulties" in language acquisition.Constraints such as [*ComplexNuc&NoCoda] () would be constructed because the learner experiences some difficulty, not because the language provides evidence for their necessity.If that were the case, how many conjoined constraints could be created to solve such "difficulties"?In other words, how many "inactive" constraints would end up composing the adult grammar?It would be more coherent to assume that the "&" operator may only be activated for the creation of a conjoined constraint that would really perform a function in the grammar.The acquisition data in BP show that conjoined constraints are constructed by children even when the adult data do not provide any motivation for their construction.This fact strongly suggests that constraint conjunction is directly related to the use of the learning algorithm, when the child is building up its input and demoting the constraints in order to reach the hierarchy of the target language.It might therefore be interesting to think about the "&" operator as a function of the learning algorithm, instead of a function of UG.It is important to observe that conjoined constraints always refer to highly marked forms.For instance, there is no conjoined constraint that prohibits a prosodic word to consist of two CV syllables.Conjoined constraints are used in a language to avoid "the worst of the worst".Under this approach, it is possible to consider that conjoined constraints are created by the learner as a learning strategy that enables him to deal with more complex structures at a point of acquisition when the demotion of the constraints that constitute it has already occurred, thus allowing the emergence of a structure that is grammatical in terms of the adult grammar, which does not contain the conjoined constraint, but which, because of a local accumulation of markedness, he is not yet able to produce.From this perspective, the "&" operator is activated exactly by the fact that a conjoined constraint performs a function: it prohibits the production of a marked structure that the low ranking of simple constraint already allows for.
How could one avoid the unnecessary proliferation of conjoined constrained in the learner's grammar, specifically in view of the fact that, after their demotion, their role in the grammar is canceled?Within the view of the "&" operator belonging to the learning algorithm, it makes sense to suggest that, after a conjoined constraint is demoted to the same stratum of the constraints that constitute itas in the hierarchy in (17)these constraints are split.Together with the existence of the "&" conjoiner operator, there would also be a splitter operator.On the other hand, the conjoined constraints that are not demoted below the constraints that constitute them, therefore not suffering violation, would be kept and would remain in the grammar.Besides having the task of discovering for their language the correct ordering of con-straints provided by UG, a learner also needs to activate conjoined constraints that really have a function in his grammar.It is important to highlight that the construction of conjoined constraints that do not militate in the adult's grammar would not be one more mechanism to reach the target grammar, but a resource made available by the learning algorithm to cope with difficulties in the production of complex structures.As can be seen in ( 27), the H 4 hierarchy (in 27b) that supports the production of a CVVC syllable structure is compared to the H 3 hierarchy in (27a) in the way their relation is understood here.Observe that the functioning of the conjoined constraint is different from the constraints that are part of UG, i. e. the ones that are not conjoined, because it must always be ranked higher in the hierarchy.This can actually point to the construction of those constraints by means of the "&" operator.
The existence of a splitting operator seems to be logical, if we consider the existence of the "&" operator.If it weren't for the "&" operator, conjoined constraints would be part of UG.According to McCarthy (2002), learners would have a relatively short list of innate constraints, with part of the learnability dedicated to finding out which universal constraints are conjoined in their language.Considering that OT is inherently typological and that the relation between the individual grammar and the universal grammar is a matter of constraint ranking, McCarthy's view is completely in line with OT logic.However, even if there is a principled way of restricting the types of constraints that can be conjoined and for providing the definition of possible domains, why should UG be overloaded with an even more excessive number of constraints?
It is suggested that conjoined constraints function in individual grammars: (i) because of the existence in CON of simple constraints, which can be classified in groups over which restrictions on conjunction may be defined; (ii) by the activation of the "&" operator in the learning algorithm.The universality of CON is retained and the same list of conjoined constraints can be potentially created in all languages.
However, this paper agrees with McCarthy (2002) that the different ranking of the constraints that are part of Universal Grammar explain the differences among grammars of different languages.Therefore, before conjoined constraints are added to a given grammar, it should first be clear that no possible ranking of simple constraints can account for the data.

Conclusion
According to the present research, we need to posit the grammar [*ComplexNuc & NoCoda] () >> Max I/O, Dep I/O >>*ComplexNuc, Onset, NoCoda to account for the third stage in the acquisition stage of BP syllable structure, when CVV and CVC (heavy) syllables are part of the child's grammar, but CVVC (superheavy) syllables are not.This grammar crucially contains a conjoined constraint.We have suggested that the way in which conjoined constraints emerge in language acquisition may shed some light on the function and the status of this constraints in a theory of language.In the present study, I have reflected on the function of conjoined constraints in the phonological acquisition process.The "&" operator is a function of the learning algorithm.The universality of CON is retained and the same set of conjoined constraints could still be created in the world's languages.c) The "&" operator is not activated by the existence of specific structures in the adult grammar only, but may also be triggered by the learner whenever he faces specific difficulties in the process of acquisition.This was argued specifically on the basis of the data presented in Bonilha ( 2000) and Levelt, Schiller & Levelt (2000); d) As conjoined constraints always refer to marked forms in a given language, it seems possible to argue that they are created by the learner as a way to cope with highly marked structures, when the demotion of the constraints of which it is composed has already occurred.e) After demotion of a conjoined constraint to the same constraint stratum that contains the corresponding 'simple' constraints, the former are split because they have lost their role in the grammar; f) The conjoined constraints that are not demoted, and which, consequently cannot be violated, will be maintained and will remain in the grammar according to the particular needs of each language.
According to this proposal, the phonological acquisition process could be seen as a reordering of constraints and the building of conjoined constraints that will actually perform a role in the target language.

Table I -
Constraint violations in zaza<a.zaand za<a.za :According to TableII, only 28 out of the 48 possible CVVC patterns really occurred, representing 58.3% of the input items with that structure.Of the remaining CVVC structures, 30.3% were produced as CVV or CVC syllables.

Table III -
CVC and CVV syllable structures production for CVVC targets

Table IV -
Constraints violated by the pairs ses<sejs and sej<sejs

Table V -
Elimination of constraints shared by the pairs ses < sejs and sej < sejsAfter markedness cancellation has applied, the constraint demotion process becomes active, based on the violated constraints given in table VI:

Table VI -
ses < sejs and sej < sejs candidate pairs are ready to activate demotion

Table VII -
Acquisition of syllable structure in Brazilian Portuguese