What has been presented in this study is a proposed algorithm and its application in second language (L2) phonology. In terms of theoretical contribution, the extended EDCD algorithm has provided an account for the interactions among Universal Grammar(UG), L1 and L2 grammars in L2acquisition with a dynamic view.
According to the bidirectional model, learners at the first stage (as those of the low-achieving group) are prone to resort to UG to process L2 stimuli. At this stage, interlanguage markedness constraints are promoted to the undominated position in order to obtain unmarked forms. At this stage, the influences of L2 grammars are not as obvious as later stages. The ranking of the constraints and their violability is illustrated in Fig. 4, where interlanguage markedness constraints are undominated.
Together with L1 constraints, interlanguage constraints are powerful enough to determine learners’ output forms. Following the convention of OT tableau indication, the lower ranked L2 constraints are shaded, meaning the violations of these constraints are, if not completely, often ignored. At this stage, any resemblance between the learners’ forms and the L1/L2 forms are due to the lack of constraint conflict between Interlanguage and the L1/L2 grammars. With the promoted interlanguage markedness constraints, the interlanguage grammar at this stage is different from L1 and L2 grammars.
Interlanguage Markedness
Fig. 4 Constraint domination in stage one
At early stage two, as shown in Fig. 5, the effect of interlanguage markedness constrains starts to fade with error-driven constraint demotion. At this point, both interlanguage markedness and L1 constraints contribute to the shape of learners’
forms. On the one hand, if L1 and L2 grammars share more similarities, like English and Chinese compound noun stress assignment, constraint reranking will be less complicated, for the conformation to L1 constraints equals to the observance to partial
•‧
國
立立 政 治 大
㈻㊫學
•‧
N a tio na
l C h engchi U ni ve rs it y
L2 constraints. However, fewer steps in constraint reranking do not promise better proficiency or a faster familiarity process. On the other hand, if L1 and L2 grammars are less similar, the constraint reranking will be more complicated, and more varieties of learners’ forms will surface in the interlanguage.
Interlanguage Markedness Constraints
L1 / L2 Constraints
inviolable violable
Fig. 5 Early stage two: error-driven constraint demotion
Next, as shown in Fig. 6, in a later stage, most interlanguage markedness constraints are demoted to the bottom rank of interlanguage grammar. At this point, most target forms can be successfully selected and grammar construction has come to a temporary end. The suspension of grammar construction allows some L1 constraints to coexist with L2 constraints on the higher rank and it is the main reason why interlanguage grammar is never identical to L2 grammar.
Interlanguage Markedness / L2 >> L1 Constraints
inviolable violable
Fig. 6 Mid stage two: error-driven constraint demotion
Fig. 7 shows the distribution of constraints at late stage two. At this stage, most interlanguage and L1 constraints are demoted to the right edge of the tableau. With the dominance of L2 constraints, attested outputs are ensured to surface at this stage.
However, the analysis in the previous sections has shown that not every prediction requires so many steps. In fact, most target forms can be successfully selected as optimal choices at mid stage two.
•‧
Fig. 7 Late stage two: error-driven constraint demotion
Based on the extended EDCD, the present study analyzes Chinese speakers’
stress assignment in English words and has established different sets of interlanguage subgrammars under the framework of Cophonology Theory. As far as simplex nouns are concerned, L2 learners’ preference for stress on penult and heavy ultima is motivated by universal unmarkedness. Learner’s forms can be predicted by promoting ALLFTR to the top position; target forms can be eventually predicted by error-driven constraint demotion. Next, the analysis on complex words shows that the prediction of learners’ forms can be obtained by promoting the interlanguage markedness constraints, ALLFTR and NON-FIN(σ). Target forms can be eventually predicted by error-driven constraint demotion. Unlike simplex nouns, which is solely determined by phonological factors, the stress assignment in complex words involves the morphophonological knowledge of different suffix types. At the first stage, learners’
forms are controlled by interlanguage markedness constraints. Low achievers possess only one cophonology for the three different types of suffixation. The three cophonologies for three suffixations are not established until a later stage when L2 learners’ level is advanced. Finally, L2 learners’ stress assignment in compound nouns is surprising in that the likeliness between Chinese and English compound stress grammars does not facilitate target forms to be acquired successfully at the very beginning. Rather, to override L1 and L2 stress grammar, and ALIGN-RIGHT(WD, FT) are promoted to the undominated position. The promotion of interlanguage markedness constraints often lead to unmarked forms, and learners’ forms are refined to target forms at a later stage. The cophonologies of stress assignment in English simplex nouns, complex words, and compound nouns are summarized in (1) and (2).
•‧
(1) Grammar lattice of L2 learners’ English word stress grammar at stage one Master Ranking
•‧
國
立立 政 治 大
㈻㊫學
•‧
N a tio na
l C h engchi U ni ve rs it y
•‧
國
立立 政 治 大
㈻㊫學
•‧
N a tio na
l C h engchi U ni ve rs it y
•‧
linguistic research. First, interlanguage grammar is found to be dynamic, imperfect, and independent. Although interlanguage grammar does follow certain route models, encounters with different types of data can trigger different constraint rerankings. For example, a multisyllabic complex word like interviewee may trigger the error-driven constraint demotion of ALLFTR, while a disyllabic counterpart like payee does not, since it does not violate ALLFTR. Second, interlanguage grammar is never perfect. The main reason is that the involvement of UG and L1 grammars complicates L2acquisition to some extent.However, the analysis shows that interlanguage grammar does not have to be identical to L2 grammar to predict target forms accurately. Since interlanguage grammar is never perfect, it is open to reconstruction anytime whenever there is a difference between the winner forms and the target ones. Third, interlanguage grammar is independent of L1 and L2 grammars.
The differences between interlanguage grammar and L1 and L2 grammars can be easily detected by their respective rankings presented at each stage.
The operation of the extended EDCD also shed light on the attributes of constraints activated in the interlanguage grammar. These constraints are found to be activated and promoted to override L1 and L2 constraints to generalize interlanguage forms. Besides, interlanguage markedness constraints are more mobile than those constraints in L1 and L2 grammars. They can be promoted to the undominated position, even if they are originally low-ranked or inactivated in L1 or L2 grammars. Despite the promotion of interlanguage markedness constraints, the extended EDCD claims that error-driven constraint demotion should be sequential and minimal. The present study finds the pivotal constraints in EDCD dispensable and argues that the sequence of demotion is in fact determined by the frequency of L2 stimuli, which can be detected by L2 constraint ranking. The reference to L2 constraint ranking excludes an arbitrary order of demotion as proposed in EDCD. Finally, the extended EDCD requires constraint demotion to be rank by rank so that constraint demotion can be minimal. Once the target form is successfully selected, constraint demotion will be suspended.
The extended EDCD is an L2 learning algorithm. The analyses in the previous sections have demonstrated how interlanguage grammar is formed. A broader picture of the whole operation of the algorithm can be illustrated in the model in Fig. 8. Following the arrows, L2 tokens enter the model in the form of overt structures. After being perceives, these L2 tokens are parsed to detect such structures as underlying forms, constraints that they conform to or violate. Next, by means of error-driven constraint demotion, the constraint ranking for
•‧
國
立立 政 治 大
㈻㊫學
•‧
N a tio na
l C h engchi U ni ve rs it y
certain types of L2 tokens is determined and reserved as L2 grammar. That is the reason why L2 constraints can join at stage one with a proper ranking. Following stage one, stage two is divided into three steps, including early-, mid-, and late stages. Once the constraint ranking is proved to be workable by L2 tokens, the interlanguage grammar surfaces as Interlanguage Grammar of a certain type. L2 learners apply the interlanguage grammar to process L2 tokens they consider belong to the same kind. If Grammar One fails, Grammar Two will be employed for processing the data. If all the Grammars fail to predict the target form, the whole model will restart from the very start, i.e. Full Structural Descriptions.
EDCD
Interlanguage Markedness Constraint Promotion L2 Constraint Incorporation
EDCD
Full Structural Descriptions
Grammar Stage One Early Stage Two Mid Stage Two
Late Stage Two Grammar Two
………..
Grammar One