Background Background Core question Core question Core question Romance conjugations Romance conjugations Generalisation Generalisation Elicited production (pt) Elicited production (pt) Additional studies Additional studies Beyond affix-stripping: Priming Priming Cross-modal priming Cross-modal priming Generalisation and processing of Masked priming Masked priming Conclusion Conclusion ‘pure morphology’ General research question João Veríssimo • What explains productivity? • What kind of linguistic knowledge enables it? Potsdam Research Institute for Multilingualism MoProc 24 June 2017 1/34 2/34 Background Background Two broad approaches Two domains for testing Core question Core question Romance conjugations Romance conjugations Generalisation Generalisation Elicited production (pt) Elicited production (pt) Additional studies Additional studies Rule-based; dual-mechanism (e.g., Pinker, 1999) Priming Priming Cross-modal priming Cross-modal priming Masked priming Masked priming Generalisation • Some linguistic knowledge/processing involves: Conclusion Conclusion • Rules (e.g., X → X ed +past ) • Context-free vs. context-sensitive operations • Structured representations (e.g., [[walk][ed]]) • Generalisation to nonce words (elicited production) Similarity-based approaches (Gonnerman et al., 2007) Decomposition • Morphological knowledge ‘emerges’ . . . • Structured vs. ‘undecomposed’ representations • . . . from regularities between form and meaning • Morphological priming (cross-modal, masked) • No rules • No structured representations 3/34 4/34
Background Background Core question ‘Pure’ morphology Core question ‘Pure’ morphology Romance conjugations Romance conjugations Generalisation Romance verb conjugations Generalisation Romance verb conjugations Elicited production (pt) Elicited production (pt) Additional studies Additional studies Priming Priming Cross-modal priming Cross-modal priming Masked priming Conjugation classes Masked priming Conclusion Conclusion [[[ cant] a] stem va] past imp. [[[ sorr] i] stem a] past imp. • Theme vowels define three arbitrary classes Conjugations as ‘pure’ morphology • Morphological, but not ‘meaning-bearing’ • Theme vowels select verb endings • Determine mappings between form and meaning • “Irreducible morphological categories” (Aronoff, 1994) 5/34 6/34 Background Background ‘Pure’ morphology General hypothesis Core question Core question Romance conjugations Romance conjugations Generalisation Generalisation Romance verb conjugations Elicited production (pt) Elicited production (pt) Additional studies Additional studies Priming Priming 1 st conj. stems Cross-modal priming Cross-modal priming Masked priming Masked priming Striking discrepancy in productivity Conclusion Conclusion • Context-free rule: X root → X a stem • In Portuguese, Italian, etc. . . . • Generalised irrespective of phonological properties • 1 st conj. welcomes novel words, borrowings, etc. • Constitute structured representations • 2 nd and 3 rd conjs. are seldom generalised 2 nd and 3 rd conj. stems Example • ‘Exceptions’ to default stem-formation rule • to blog ‘blogar’ (Port.), ‘bloggare’ (Ital.) • Generalisation is sensitive to phonology of root • [[[ blog] root a] stem r] • ‘Whole-stem’ representations 7/34 8/34
Background Background Core question ‘Pure’ morphology Core question Our studies Romance conjugations Romance conjugations Generalisation Romance verb conjugations Generalisation Elicited production (pt) Elicited production (pt) Additional studies Additional studies Priming Priming Generalisation Cross-modal priming Cross-modal priming Masked priming Conjugation classes Masked priming • Elicited production (Port.) Conclusion Conclusion • Computational simulations (Port.) • Elicited production (Ital.) • Theme vowels define three • Reanalysis of Albright (2002) (Ital.) arbitrary classes • Morphological, but not Priming ‘meaning-bearing’ • Cross-modal priming (Port.) • Masked priming (Port.) 9/34 10/34 Background Background Generalisation of conjugations Elicited production (pt) Core question Core question Romance conjugations Romance conjugations Generalisation Generalisation Veríssimo & Clahsen (2014), JML Elicited production (pt) Elicited production (pt) Additional studies Additional studies Method Priming Priming Cross-modal priming Cross-modal priming • 54 native speakers of European Portuguese Masked priming Masked priming Computational simulation Conclusion Conclusion • 78 novel verbs in the 1sg pres. ind. • Minimal Generalisation Learner (Albright, 2002) ( which does not display a theme vowel ) • Input: Pairs of 1sg and Infinitive forms • Participants had to fill a gap with an infinitive form • 1sg has no theme vowel ( which requires a theme vowel ) • Output: A set of phonological environments . . . Example • . . . and corresponding reliability values for each class Quase sempre tureço sozinho. • (e.g., in English past tense, ing is predictive of i → a ) Mas amanhã vou acompanhado. “I almost always tureço alone. But tomorrow I will someone.” • Possible answers: tureç a r , turec e r , turec i r 11/34 12/34
Background Background Core question Elicited production (pt) Core question Elicited production (pt) Romance conjugations Romance conjugations Generalisation Materials Generalisation Results Elicited production (pt) Elicited production (pt) Additional studies Additional studies Priming Priming Cross-modal priming Cross-modal priming • Three (weighted) regressions, predicting 1 st , 2 nd and Masked priming Masked priming MGL simulation 3 rd conj. response log-odds Conclusion Conclusion • Input: 3,117 Portuguese verbs • Each w/ similarity to the 3 conjugations as predictors • 1sg to Infinitive Responses (Log-odds) Construction of novel verbs 1 st (-ar) 2 nd (-er) 3 rd (-ir) Predictors • 78 novel verbs created from MGL rules Reliab. 1 st conj. . 03 − . 04 . 01 • Spanning a wide range of reliability values Reliab. 2 nd conj. . 67 ∗ − . 65 ∗ − . 10 • Each novel verb is associated with 3 reliability values Reliab. 3 rd conj. − . 41 ∗ − . 16 . 58 ∗ 13/34 14/34 Background Background Elicited production (pt) Additional studies I Core question Core question Romance conjugations Romance conjugations Generalisation Generalisation Results Elicited production (pt) Elicited production (pt) Additional studies Additional studies Priming Priming Cross-modal priming Cross-modal priming Masked priming Masked priming Model comparison (Veríssimo & Clahsen, 2014) Conclusion Conclusion (a) 1st Conjugation (b) 2nd Conjugation (c) 3rd Conjugation • Comparison of predicted proportions of responses in 2.5 0.0 0.0 MGL vs. ‘dual-mechanism’ implementation −0.5 −0.5 2.0 Log−odds of response Log−odds of response Log−odds of response • Default Generalisation Learner (DGL) −1.0 −1.0 1.5 • MGL underestimated 1 st conj responses and −1.5 −1.5 1.0 overestimated 2 nd and 3 rd conj. responses −2.0 −2.0 0.5 −2.5 −2.5 0.0 • DGL predictions for each of the three conjs. were 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 Reliabilities Reliabilities Reliabilities statistically indistinghuishable from human responses 15/34 16/34
Background Background Core question Additional studies II Core question Additional studies III Romance conjugations Romance conjugations Generalisation Generalisation Elicited production (pt) Elicited production (pt) Additional studies Additional studies Priming Priming Cross-modal priming Cross-modal priming Masked priming Masked priming Elicited production (Italian) (Veríssimo, in prep.) Conclusion Conclusion Reanalysis of Albright (2002) (Veríssimo, in prep.) • 35 native speakers • Acceptability judgements experiment (Italian) • 40 novel verbs (from Albright, 2002) • Ratings of 2 nd and 3 rd conj. forms were predicted by • 2 nd conj. responses predicted by MGL reliabilities MGL reliability metric • 1 st conj. responses predicted by trade-off effects • Ratings of 1 st conj. forms were predicted by root • No significant effects for 3 rd conj. responses well-formedness and trade-off effects 17/34 18/34 Background Background Generalisation in Romance Cross-modal priming Core question Core question Romance conjugations Romance conjugations Generalisation Generalisation Veríssimo & Clahsen (2009), Cognition Elicited production (pt) Elicited production (pt) Additional studies Additional studies Priming Priming Discussion Cross-modal priming Cross-modal priming Masked priming Masked priming • Generalisation of 1 st conjugation in Romance Conclusion Conclusion languages is not sensitive to the phonological properties of novel roots (cf. Albright, 2002) • 1 st conj. generalised more widely than what would be predicted by the reliability metric • Generalisation of 2 nd and 3 rd conjs. is based on phonological similarity Results support a model that makes use of both context-free and similarity-based generalisations 19/34 20/34
Recommend
More recommend
