Design of artificial languages to study language perception and its evolutionary origins Matthew G. Collison
Initial aims My initial aims for the study were: • Develop an artificial language that behaviourally tests specific components of language perception. • Design an artificial language that integrates these components of language to indicate processing of perception. • Design an artificial language that spans the linguistic perception capability across the species to determine the evolutionary conservation of language perception. • Optimise the paradigm for future imaging studies.
Backgroun Literature • Phonetic identification • Petkov ’09 • Statistical segmentation • Saffran ’96 • Abla ’08 • Structural perception • Federici ‘06 • Comparative structural perception studies in humans and monkeys • Fitch and Hauser ’05 • Saffran and Hauser ‘08
Aims of the artificial language I designed an artificial language to integrate the three theories into a comparable function and develop an overall model of language perception summarising interaction between the 3 components. • I included a tonal and word based parameter which gave an indication as to influence of statistical and phonetic input into different levels of structure. • I included a multiple exemplar function to give two levels of structural complexity. • I included manipulated the discriminatory testing to allow us to behaviourally evaluate to what extent subject rely on statistics or rule breaks to discriminate structure at different levels.
The Languages 10 possible Sequences from grammatical structure ADCGFC ACGFCG ADCFCG ADCFC ACFCG ACGFC ADCGF ACFC ADCF ACGF Basic structure from Saffran and Hauser Sequence A phrase + B phrase + (C phrase) A phrase A + (D) B phrase C + F C phrase C + (G) ( ) letters in brackets are optional elements Rules present in this language: • The basic structure must include A –C – F – • If D is present it must follow A. • If G is present it must follow C
Procedural Paradigm There are two stages to the experimental procedure Training – For 5 minutes participants are exposed to training programme Testing participant are presented with completely novel stimuli and have to determine whether they think it is grammatical or ungrammatical.
Results Figure 2. A graph to show performance in discriminating local structure in tone based languages after 5 minutes exposure to the structure. Figure 1. A graph to show performance in discriminating local structure in non sense word based languages after 5 minutes exposure to the structure.
Results Figure 3. A graph to show performance in discriminating global structure in non sense word based languages after 8 minutes exposure to the structure. Figure 4. A graph to show performance in discriminating global structure in tone based languages after 8 minutes exposure to the structure.
Conclusions • We have designed an artificial language that has advantages for human and monkey study: • 1. Behaviour can be evaluated without giving feedback (monkeys will not be given feedback). • 2. Tone languages can be created. Nonsense words (speech) can be used if needed. • 3. We can evaluate if subjects (human or monkey) rely on statistics or rule breaks. • Our human data suggests that subjects can learn local and global structure in both word based and tonal artificial languages.
Next Steps: • We believe the paradigm which has now been behaviorally tested in humans is: • Ready for monkey behavior (using preferential looking paradigms) • Ready for fMRI study. • Long term objective: Through comparative behavior and fMRI, the paradigm that we have developed has the potential to reveal what would have been the precursory network in the nonhuman primate brain that evolved to support language in humans. Thank you for your attention.