Abstract: An important issue in the study of language production concerns the time course of lexical access and possible interactions between lexical selection and phonological encoding. A number of different models have been proposed in the literature to account for how a word is processed for speaking. For example, the discrete two-stage model assumes that the stages of lexical selection and phonological encoding operate in a strict successive order, and no interaction between the stage of lexical selection and phonological encoding. Alternatively, on the basis of spreading activation, the interactive activation model postulates a bi-directional flow of information between different processing levels. Semantic, lexical, and phonological levels are interconnected in a network fashion. Lexical candidates activate their phonological forms before any single candidate has been selected. That is, not only the word forms of the target lemma, but also the semantic competitors become partially activated. The cascading model assumes that there exists interaction between lexical selection and phonological encoding, but no feedback from the stage of phonological encoding to lexical selection. The debates among three models are (1) whether there is the interaction between lemma selection and phonological encoding; (2) whether all activated semantic items or only a single selected semantic item gives rise to the phonological activation; (3) whether there is feedback from phonological encoding to lexical selection.
Due to confounding of orthography and phonology in alphabetic languages (i.e., English and Dutch), the results of the previous studies (Lupker, 1982; Starreveld & La Heij, 1995) need to be interpreted cautiously. Using the independence of orthography and phonology in Chinese, the present experiments investigated (1) the multiple phonological activations and the multiple orthographic activations of semantically related words of picture name; (2) whether there is the feedback from phonological encoding to lexical selection with picture-word interference paradigm.

Method
In three experiments, the picture-word interference paradigm was used. The participants were required to name a target picture and ignore a distractor word. We varied both the stimulus-onset asynchrony (SOA) and the picture-word relationship along different lexical dimensions. Fifty-five undergraduate students participate in three experiments (20 in experiment 1 and 2 respectively, 15 in experiment 3). ANOVA analyses were performed in the participant and the item analysis in three experiments.

Results
The data analyses showed that multiple phonological activation of semantically related to target was found in experiment 1, whereas no orthographic activation of semantically related to target was observed in experiment 2. After improving the conditions of experiment 1 and 2, experiment 3’s results suggested that: 1) no phonological or orthographic activation of semantically related to target was obtained, 2) no orthographic activation of phonologically related to target was found, and 3) phonological activation of orthographically related to target was obtained.

Conclusion
The results suggested that there exist the interaction of lexical selection and phonological encoding, but no activation feedback from phonological stratum to semantic stratum in picture naming. Therefore, the present study supported the cascading model, rather than the discrete two-stage model or the interactive activation model in language production

Key words: speech production, picture-word interference paradigm, lemma selection, phonological encoding, interaction