Linguistic typology and cross-linguistic psycholinguistics

1. Tsung-ying Chen Dept. of Foreign Languages & LiteratureNational Tsinghua UniversityTaiwan James Myers Graduate Institute of LinguisticsNational Chung Cheng UniversityTaiwan Linguistic typology andcross-linguistic psycholinguistics 50th Annual Meeting of the SocietasLinguisticaEuropaea, Zürich12 Sept 2017

2. Welcome to our workshop!

3. Why do typological psycholinguistics? • To see how language experience affects learning and processing, and thus... • ... to seek out and explain linguistic universals • Two methodological approaches • Empiricist: Induce universals from large samples • Rationalist: Deduce universals from learnability • Expanding the typology Overview

4. Typologists are doing more experiments • E.g. Bickel et al. (2015) • Psycholinguists are doing more typology • E.g., Norcliffe et al. (2015) • Both could learn more from each other • Experiments to test processing-based explanations for language universals • Experiments on a wider variety of languages to test hypothesized processing universals • These goals cannot be achieved in isolation! Typology and psycholinguistics

5. Universal processing models? O'Seaghdha et al. (2010), modifying Levelt et al. (1999)

6. Or universal learning models? Chinese experience Universal learning algorithm English experience

7. Rationalist/deductive approach • Chomsky (1965): Learners need universals • So assume all is universal unless shown otherwise • Empiricist/inductive approach • Greenberg (1963): Languages vary in limited ways • So study the typology of cross-linguistic variation • What about psycholinguists? • Empiricists when studying individual languages • But rationalists when making universal claims Testing universals

8. A typical psycholinguistic study

9. A typical cross-linguistic study

10. Two-language studies imply a factorial design • But cross-linguistic variables are confounded • English vs. Chinese orthography (ABC vs. 字) • English vs. Chinese homophones (few vs. many) • Some variables are also gradient • Number of homophones: few, more, many, ... • Additional variables keep getting discovered • English vs. Chinese relative clauses: N-RC vs. RC-N • ... but also many language-specific factors (Lin, 2017) Two languages don’t make typology

11. Lexical researchers know these problems well • Participants come into lab knowing their lexicon • Words cannot be experimentally manipulated • Lexical variables are also gradient & confounded • Gradience: Frequency, length, ... • Confounding: High-frequency words are shorter • Designing factorial lexical experiments is thus a “confounded nuisance” (Cutler, 1981) • A recently popular solution: megastudies • Regression analyses on large, diverse data sets The role of empiricism/induction

12. The English Lexicon Project (Balota et al. 2007) • 1000s of visual lexical decisions and naming responses for 1000s of written English words and non-words: http://elexicon.wustl.edu/ • Similar projects in Dutch, French, Chinese, Malay (Balota et al. 2012; Keuleers & Balota 2015) • Also auditory lexical decision, priming, picture naming, ... http://crr.ugent.be/emlar2015/list%20of%20%20lexical%20databases.html The megastudy movement

13. Myers (2016) • Generalizing megastudy logic cross-linguistically So how about Meta-megastudies...? Behavior in task (a black box) Response Language experience Experimental task Item-level variables Language-level variables

14. (e.g., Cysouw, 2005) Sampling languages for grammar Universal principle “Natural” Type A Type B Borrowing Lang Lang Lang Lang Learning fromspeech communityhides natural patterns Lang Lang Descent

15. Sampling languages for processes Language experience “Natural” Borrowing & descentcan only bias up here Universallearningalgorithm Process A Process B Lang Lang Lang Lang Processing ignores (?) community conventions

16. For psycholinguistics, language samples only need to deconfound processing-related variables • Borrowing and descent unlikely to affect processing independently of the typological features themselves • “Small” samples OK (e.g., Vittinghoff & McCulloch, 2007) • Only need around 10 x number of variables • Partial confounds OK (e.g., O’Brien, 2007) • Even with r2 = .9 or higher diversity > representativeness

17. Dressler et al. Chen & Myers White et al. Namboodiripad Culbertson & Braquet Dimitriadis et al. Sauppe et al. Blumenthal-Dramé & Kortmann Language samples in today’s talks Bulgarian Cantonese Danish Dutch English Estonian Finnish French German Greek Hindi Hungarian Italian Korean Lithuanian Malayalam Mandarin Russian Saami Spanish Turkish

18. If learners need universals, then why not study universals by looking at learners? • Natural language learning • Children are biased (e.g., Lidz & Gagliardi, 2015) • Adults treat “natural” and “unnatural” patterns within their own language differently (e.g., Hayes et al., 2009) • Artificial grammar learning • Simulates second language learning, allowing for controlled tests of learning biases (Culbertson, 2012) The role of rationalism/deduction

19. Natural first language acquisition by children • Dressler et al. • Artificial language learning by adults • Culbertson & Braquet • Dimitriadis et al. • White et al. • Presumed effects of first language on adults • Blumenthal-Dramé & Kortmann • Chen & Myers • Namboodiripad • Sauppe et al. Learning in today’s talks

20. Artificial grammar learning Learningsucceeds Language Aexperience Universal learning algorithm Language Bexperience Learningfails

21. Artificial learning in context English-modulatedlearning algorithm A>B Englishexperience Universallearning algorithm Lang. A Lang. B Chineseexperience A>B Chinese-modulatedlearning algorithm

22. Make lots of friends... • Bates et al. (2003): 7 languages, 22 authors • ... and/or emulate the typological grammarians • Thousands of independent scholars collect data on specific languages for their own purposes... • ... and then a relatively small number of typologists compile databases (e.g., Haspelmath et al., 2005) • Could psycholinguistics be done this way too...? How to expand the typology?

23. The rest of Today’s schedule

24. Adelman et al. (2014). A behavioral database for masked form priming. Behavior Research Methods, 46 (4). Balota et al. (2007). The English Lexicon Project. Beh. Res. Meth., 39. Balota et al. (2012). Megastudies: What do millions (or so) of trials tell us about lexical processing? In Adelman (Ed). Visual word recognition, Vol. 1. Psychology Press. Bates et al. (2003). Timed picture naming in seven languages. Psychonomic Bulletin & Review, 10 (2). Bickel et al. (2015). The neurophysiology of language processing shapes the evolution of grammar: Evidence from case marking. PLoS ONE, 10(8). Chomsky (1965). Aspects of the theory of syntax. MIT Press. Culbertson (2012). Typological universals as reflections of biased learning: Evidence from artificial language learning. Language & Linguistics Compass, 6. References (1/4)

25. Cutler (1981). Making up materials is a confounded nuisance, or: Will we be able to run any psycholinguistic experiments at all in 1990? Cognition, 10. Cysouw (2005). Quantitative methods in typology. In Kohler et al. (Eds.) Quantitative linguistics: An international handbook. Walter de Gruyter. Feng et al. (2001). Rowed to recovery: The use of phonological and orthographic information in reading Chinese and English. J. of Experimental Psychology: Learning, Memory and Cognition, 27 (4). Greenberg (1963). Some universals of grammar with particular reference to the order of meaningful elements. In Greenberg (Ed.) Universals of language. MIT Press. Haspelmath et al. (Eds.) (2005). The world atlas of language structure. Oxford University Press. References (2/4)

26. Hayes et al.. (2009). Natural and unnatural constraints in Hungarian vowel harmony. Language, 85(4). Keuleers & Balota (2015). Megastudies, crowdsourcing, and large datasets in psycholinguistics: An overview of recent developments. Quarterly Journal of Experimental Psychology, 68 (8). Levelt et al. (1999). A theory of lexical access in speech production. Behavioral and Brain Sciences, 22 (1). Lidz & Gagliardi (2015). How nature meets nurture: Universal grammar and statistical learning. Annual Review of Linguistics, 1(1). Lin (2017). Sentence processing: Relative clauses. In R. Sybesma et al. (Eds.), Encyclopedia of Chinese language and linguistics, vol. 3. Brill. Myers (2016). Meta-megastudies. The Mental Lexicon, 11(3). References (3/4)

27. Norcliffe et al. (2015). Cross-linguistic psycholinguistics and its critical role in theory development: Early beginnings and recent advances. Language, Cognition and Neuroscience, 30(9). O’Brien. (2007). A caution regarding rules of thumb for Variance Inflation Factors. Quality & Quantity, 41. O’Seaghdha et al. (2010). Proximate units in word production: Phonological encoding begins with syllables in Mandarin Chinese but with segments in English. Cognition, 115. Vittinghoff & McCulloch (2007). Relaxing the rule of ten events per variable in logistic and Cox regression. American Journal of Epidemiology, 165 (6). References (4/4)