1. Introduction
   1. Campbell summarizes research methods of mental processes in translation, namely; Think-aloud protocols, word-based experimental techniques (rooted in cognitive psychology), and neuroimaging. He refers to an 'overlooked' process which can shed light on the translator behavior; the target text itself.
2. The basic architecture of mental processing
   1. Campbell builds his model on a competition model in interpreting research which was initially developed to study language acquisition. It also part of connectionist models which assume that "the brain relies on a type of computation that emphasizes patterns of connectivity and activation" (Mcwhinney 1997:222),rather than on serial processing.
   2. Campbell first attempted to model lexical decisions-making based on the notion that translation involves sequential steps. Later, he concluded that the process of decision making involves the simultaneous sequencing of information of various types; lexical, semantic, syntactic, pragmatic, etc.
3. Principles of choice network analysis
   1. Target texts can be used as tangible source of evidence of mental processing in translation, and an alternative to experimental data and think-aloud protocols.
   2. The products of a sample of subjects translating the same text into the same language will reveal a range of differences and similarities in the behaviours of the subjects.
   3. As the sample becomes larger, the complete range of behaviours of translators of that text between those languages is approached.
   4. A model of the mental processing underlying the translation of that text in that language combination can be inferred through a comparison and classification of the behaviours of the sample of subjects;
   5. General principles about mental processing can be extracted from analyses of specific texts and language combinations, and used as hypotheses for examining other texts and language combinations'
4. Examples of choice network analysis
   1. Examples of networks are presented covering translation between Arabic and English and Spanish and English in such areas as lexis, complex noun phrases, ellipsis, and passives. The study attempt intended to identify the mental processing underlying the translation of texts by multiple subjects. The products were compared, and a number of strategies identified; these strategies deal with, among other things, the treatment of metaphors and the extent to which subjects use information beyond the level of the word to inform their choices.
5. Range of applications of choice network analysis
   1. CNA can be used to reveal contrasts between the translatability of specific language pairs, making this an empirically based rather than a theory-driven exercise.
   2. CNA is also useful for estimating the relative difficulty of parts of source texts.
   3. CNA can also be used to compare the strategies of novice and expert translators, and those of translators into the first and second language.
   4. CNA is a valuable tool for generating hypotheses about translation that can be tested by other research methods.
6. Choice network analysis and theory
   1. CNA is a theory-free approach that generates are models, i'e. representations of complex systems.
   2. The researcher is free to make these models as simple or as complex as the hypothesis under investigation demands.
   3. The researcher is also free to incorporate theory into a network;
   4. The power of CNA lies precisely in the fact that it provides a test bench for theory.
7. Principles for building networks
   1. It must account for every piece of data in the sample that is relevant to the domain of the theoretical framework of the investigator.
   2. It must be linguistically plausible
   3. It must be optimally parsimonious, i.e. it should contain the minimum number of nodes and branches that will account for all the data, while remaining plausible
8. Linearity
   1. The sequence of processing in a network is a function of two factors:
      1. One is the general principle that, despite subsequent passes and post-editing, text processing is basically linear. We tend to process from the beginning of a string.
      2. The second is that choices at any point in the string may constrain subsequent choices; these constraints are the unavoidable product of grammar.
9. Conclusion
   1. CNA is an empirical research procedure that takes the approach of comparing multiple subjects' translations of the same text.
   2. It complements and, in some respects, improves on current research paradigms in translation research such as think-aloud protocols and psychological word-based experiments.
   3. The procedure has a range of applications in contrastive translation research, in the study of difficulty, and in the investigation of translation competence. It is a sound procedure for generating research hypotheses, and for testing psycholinguistic theories about translation.