In this part bilingual mental lexicon and its implications have been fully discussed.Models of lexical access include Serial Search Model and Parallel Access Model.The former means that the process takes place by scanning one lexical entry at a time.The best known Serial Search Model is Forster's Autonomous Search Model.The latter proposes that perceptual input about a word can activate a lexical item directly，and that multiple lexical entries are activated in parallel.That is，a number of potential candidates are activated simultaneously.Parallel Access Models include the Morton's Logogen Model，Conncetionist Model and Cohort Model.According to the Morton's Logogen Model，it proposed that words are not accessed by determing their location in the lexicon but by being activated to a certain threshold.Thus，a space analogy of lexical access such as one saw in Forster（1976）is replaced with a more electrical analogy—a word will“light up”when the electrical current is sufficient strong；Connectionism advocates use the analogy of the brain and neurons to develop models of cognition.Their computer models of cognitive processes（such as lexical access）are instituted in“neural nets”composed of nodes and connections between these nodes.Nodes are three types：input nodes，which process the auditory or visual stimuli；output nodes，which determine responses，and hidden nodes，which perform the internal processing between when we hear and see a word and when we respond to it.The hidden nodes do the lion's share of lexical processing.Like the Morton's Logogen Model and the Connectionist Model，the Cohort Model of lexical access poists that multiple candidates are activated in parallel，like its cousins，the Cohort Model states that the list of word candidates is narrowed as the auditory input proceeds serially.It explains frequency and nonword effects in much the same way as the Morton's Logogen theory.

One of the most influential psychologuistic models for speech production，developed by Levelt，views it as a linear progression of four successive stages：（a）conceptualization（b）formulation（c）articulation and（d）self-monitoring and these four stages have been fully discussed.

Models of semantic representation include Hierarchical Network Model and Feature Comparison Model.One important aspect of Hierarchical Network Model emphasizes on cognitive economy，another aspect of the model is the category size effect，that is，the larger the category，the longer hierarchical network view have been revealed.Feature Comparison Model includes（a）defining features and（b）characteristic features.Features Comparison in semantic decision tasks is assumed to be a two-stage process.In the first stage，all the feature-defining and characteristic of the two concepts are compared in a global comparison.Spreading Activation Network，like the Hierarchical Model，is an associated network.An important aspect of this model is the principle of spreading activation.When a single concept is activated，the“electricity”spreads to connected concepts，decreasing in strength as it emanates outward.

In the section of mental representation of the bilingual lexicon，different aspects as input representation and their lexicalization，from the conceptual system to the verbalization system，the verbalization function，from semantic form to lemma are carefully studied.

According to the balanced theory，the picture is of weighing scales，with a second language increasing at the expense of the first language.Cummins'Separate Underlying Proficiency Model of Bilingualism conceives of the two languages operating separately without transfer and with a restricted amount of‘room’for languages.But its fallacies have been revealed.According to the Iceberg Analogy，it assumes that the two languages are visibly different in outward conversation underneath the surface，the two icebergs are fused such that the two languages do not function separately.Both languages operate through the same central processing system，the trend is toward acceptance of the Common Underlying Proficiency of Bilingualism.

Concept-mediation vs word-association have been carefully studied.The word association model proposes that second language words are associated to first language words，and that only through first language mediation do second language words gain access to concepts.In contrast，the concept-mediation model proposes that second language words directly access concepts.Empirical evidence has shown that early in second language learning，individual mediate understanding of second language words through the first language.That is，they follow the word association model，at later stages of second language learning，individuals may be able to directly understand the meaning of second language words without first language mediation.For the less fluent subjects，translation times were faster than picture naming times，as the word association model predicts.For the more fluent subjects，translation and picture naming were not statistically different，as the concept mediation model predicts.

It has been found that the speed and accuracy of translation depends on the direction of translation.Performance is faster and more accurate to translate from L2 to L1 than from L1 to L2；lexical links from L2 to L1 are hypothesized to be stronger than lexical links from L1 to L2，but conceptual links for L1 are stronger than those for L2.

Evidence has shown that the two directions of translation differ because translation from L2 to L1 is more likely to be accomplished lexically than translating from L1 to L2，which is more likely to engage conceptual processing.From the discussions，it can be seen that the G-T Method is also very effective in lexical learning.