Cilj ovoga rada bio je ispitati neke aspekte fonološkog kodiranja i fonetske izvedbe na temelju govornih pogrešaka izazvanih eksperimentalnom metodom – ponavljanjem tzv. brzalica. Ispitivanje se sastojalo od triju eksperimenata. U eksperimentu 1 podraţaj su ĉinile ĉetiri jednosloţne rijeĉi KVK strukture. Svaka je brzalica sadrţavala ciljani par suglasnika. Provjerile su se ĉetiri glavne hipoteze: (1) Ĉešće će se meĊusobno zamjenjivati glasnici ako su na istom poloţaju u rijeĉi nego ako su na razliĉitim poloţajima; (2) Ĉešće će se meĊusobno zamjenjivati glasnici ako se nalaze na poĉetnom poloţaju u rijeĉi nego na ostalim poloţajima; (3) Ĉešće su zamjene fonološki sliĉnih nego fonološki razliĉitih glasnika; (4) Anticipacije su ĉešće od perseveracija i razmjena. PotvrĊene su prva i treća hipoteza, ĉetvrta je djelomiĉno potvrĊena (ovisno o kriteriju po kojemu su definirane anticipacije, perseveracije i razmjene), a druga je pretpostavka odbaĉena. U eksperimentu 2 brzalice su ĉinile ĉetiri trosloţne rijeĉi, a svaka je brzalica takoĊer sadrţavala par ciljanih suglasnika. Rezultati nisu potvrdili glavnu hipotezu (hipotezu 5) u eksperimentu 2, koja kaţe da će se ĉešće zamjenjivati glasnici u naglašenim nego u nenaglašenim slogovima. U eksperimentu 3 podraţaj su ĉinile ĉetiri jednosloţne brzalice u kojima su ciljani suglasnici bili /s/ i /ʃ/. Akustiĉka analiza ciljanih suglasnika sugerira postojanje tzv. gradualnih pogrešaka, ĉime je potvrĊena hipoteza 6, prema kojoj se, osim kategorijalnih, u govornoj izvedbi ostvaruju i stupnjevite (gradualne) pogreške. Na temelju takvih zakljuĉaka preispitana su dva modela fonološkog kodiranja: model pretraţivanja i preslikavanja i model šireće aktivacije. Pored ovih šest glavnih pretpostavki, ispitani su još neki aspekti fonološkog kodiranja, kao i uspješnost metode brzalica u izazivanju pogrešaka. Zakljuĉeno je sljedeće: brzalicama se uspješno mogu izazvati govorne pogreške; sliĉnosti dvaju fonema više doprinose mjesto izgovora i zvuĉnost nego izgovorni naĉin; zamjene većinom pokazuju simetriĉan odnos (A → B = B → A); zamjene glasnika ĉešće su od izostavljanja; izostavljanja su ĉešća nego dodavanja; više je sintagmatskih nego paradigmatskih zamjena; fonološka sliĉnost ima snaţniji uĉinak na zamjene glasnika od njihova poloţaja u rijeĉi; poloţaj glasnika u rijeĉi ima snaţniji uĉinak na zamjene glasnika od metriĉkog statusa (jesu li u naglašenom ili nenaglašenom slogu); rijeĉ je, a ne slog, suprasegmentalna struktura koja upravlja fonološkim strukturiranjem izriĉaja.
The aim of this thesis was to investigate some aspects of phonological encoding and phonetic realization on the basis of laboratory induced speech errors – using so-called tongue twister paradigm. The research consisted of three experiments. In Experiment 1 twenty four female participants produced tongue twisters which included four monosyllabic CVC words of Croatian language. Each tongue twister consisted of target pairs of consonants. In the first group of the tongue twisters target consonants were in the word initial position, for example /h/ and /k/ in hod kas hip kum. In the second group of tongue twisters target consonants were in the word final position, for example /h/ and /k/ in the tongue twister dah sok puh mek. In the third group of tongue twisters one target consonant was in word initial position, and the other target consonant was in the word final position, for example /h/ and /k/ in the tongue twister had sok hip mek. The tongue twisters also differ from one another by the degree of similarity between the target consonant pair. The similarity is defined on the basis of the number of features shared by two phonemes. The system of three basic articulatory features is used: manner of articulation, place of articulation and voicing. The target consonant pair appeared in ABAB (for example /r/ and /l/ in Rim las red loš) or in BAAB (for example /r/ and /l/ in lug rep rok lan) order in the tongue twister. The task consisted of two parts: reading tongue twister aloud and then repeating it from memory. Each item started with the tongue twister presented in written form on the computer screen. After four seconds a little green circle appeared in the left corner of the screen, which was the signal to the participant to start repeating the tongue twister aloud. Participants were instructed to repeat tongue twister as long as the item and the little green circle were presented on the screen, but they could catch their breath whenever they needed to do so. After nine second, the item and the little green circle were removed and replaced by a blank white screen. After that, the same item showed up again, but just for two seconds in order to remind the participant of the stimulus words. After that the bigger green circle appeared in the center of the computer screen, indicating the start of the second part of the experimental task, in which the participants had to repeat the same tongue twister from memory. After nine seconds the green circle in the center of the screen was removed and replaced by a blank white screen. After a rest period of two seconds, the next item was presented in the same way. If the participant forgot the correct word during the recall portion of task performance, the experimenter showed the written stimulus once again, and then the participant repeated the same tongue twister from memory one more time. The rate of reciting the stimulus was 180 bits per second and it was controlled by the metronome. Items were presented in one of two randomized orders. Four main premises were tested by Experiment 1: (1) Substitutions between two segments are more frequent if they share position within the word than if they don‟t share position within the word; (2) Substitutions between two segments are more frequent if they share the word initial position than if they share the word final position; (3) Substitutions between two segments are more frequent if they are phonologically similar than if they are not phonologically similar; (4) Anticipations are more frequent than perseverations and exchanges. The results confirmed premises one and three. Premise four was only partially confirmed, depending on which of two different criteria was used to define anticipation, perseveration and exchange. Premise two was rejected. In Experiment 2 twenty female participants produced tongue twisters which included four three-syllable words of Croatian language. Each tongue twister consisted of two consonants (A and B) which were phonetically similar and constituted the target pair. There were four different stress conditions. In the first stress condition, both of the two target consonants were in the word initial position and in the stressed syllable, for example /ʒ/ and /z/ in the tongue twister ţalostan zubarski ţetveni zidati. In the second stress condition, A target consonant was in the word initial position, in stressed syllable, and B target consonant was in the word initial position, in unstressed syllable, for example /ʒ/ and /z/ in the tongue twister ţalostan zubari ţetveni zidari. In the third stress condition, both of the two target consonants were in the word initial position and in the unstressed syllable, for example /ʒ/ and /z/ in the tongue twister ţalostiv zubari ţetoni zidari. In the fourth stress condition, A target consonant was in third position inside the word and in stressed syllable, and B target consonant was in the word initial position and in the stressed syllable, for example /ʒ/ and /z/ u loţači zubarski teţina zidati. The adjacent consonants, on the third place in the word in condition 1–3, i.e. on the word initial position in the condition 4, were the same for all four stimulus words across all stress conditions, so as to exclude possible contextual influence on error rate. The target consonant pair appeared in ABAB or in BAAB order in the tongue twister. The Experiment 2 was self-paced and the speech rate of the participants was not controlled. The only instruction for participants was to speak a little bit faster than in the normal speech, but still be able to correctly and sufficiently produce each articulatory movement. All the other procedural parts were the same as in Experiment 1. The main premise (5) tested by this experiment was: Substitutions between two segments are more frequent if they are both in stressed syllable than if they are both in unstressed syllable. This premise was not confirmed by the results. In Experiment 3, stimulus was made of four tongue twisters which were composed in the same way as in Experiment 1. The target consonant pair consisted of /s/ and /ʃ/ in the word initial position. The procedure was the same as in Experiment 1. Acoustic analysis of these two consonants was used to detect potential gradient speech errors. Gradient errors are manifested as a muscular activity of the two different sounds simultaneously produced. These kinds of errors can be attributed to the partial activations of the phonological representations pertaining to competing segments during phonological encoding at the same time, as it is explained by the cascading activation models. The thesis (6), which says: Beside categorical errors, gradient errors are also possible in the speech production process; is confirmed. Apart from these six main theses, some other aspects of phonological encoding were examined, and a tongue twister paradigm as a tool for inducing speech errors as well. The conclusions are: tongue twisters are a very effective research tool in eliciting speech errors; Manner of articulation is more involved in phoneme substitutions than place of articulation and voicing; substitutions of two phonemes usually show symmetrical pattern (A → B = B → A); segment substitutions are more frequent than deletions; deletions are more frequent than additions; syntagmatic errors are more frequent than paradigmatic errors; phonological similarity is more involved in segment substitutions than a segment word position; word position has a stronger effect on segment substitutions than their metrical status (i.e. stressed or unstressed syllable position), suprasegmental structure responsible for governing phonological structure-arranging process should be attributed to the word, not to the syllable. On the basis of the results in these three experiments, two models of phonological encoding are reviewed: a spreading activation model and a scan-copier model.