Latvian children’s realisations of word-initial consonant clusters

1. Latvian children’s realisations of word-initial consonant clusters Sonority Sequencing, representational limits, and constraint reranking Martin Krämer & Olga Urek Lava Lunch, 10 March 2017

2. Introduction • Which errors do children make in the L1 acquisition of complex word-initial onsets in Latvian? • Which role does sonority play in the acquisition of complex onsets? • Do established constraints and principles on complex onsets account for the data? • Does constraint reranking suffice to account for the attested variation? • Does the factorial typology of OT constraints provide a 100% match of attested patterns? • If it doesn’t, how do we fix this? • Which other factors play a role? Perceptual biases, articulatory challenges? • Data from the LAMBA phoneme testPicture naming task with children

3. Participants • 20 monolingual Latvian-speaking children (mean age = 47 months, SD = 9.3); • Recruited in kindergartens in Riga, with Latvian as the only language of instruction; • Part of a larger study: • norming study for Latvian Phoneme Test (LAMBA); • 500 monolingual Latvian-speaking children total (36-72 months); • 20 children obtaining the lowest scores were selected;

4. Procedure • Latvian Phoneme Test: • 91 pictures depicting objects and actions; • items chosen to contain all consonants and selected clusters in initial, medial and final position • Children tested individually in a quiet room by two evaluators; • Picture-naming procedure: • Ev. 1 showed the pictures to the child; • Ev. 2 marked the responses on the scoring sheet; • Each session was audio-recorded and later transcribed; • Spontaneous production was elicited where possible; • If the child could not name the picture spontaneously, delayed imitation was used;

5. Data set: scope • All attempts at word-initial two-member clusters were extracted; • Six cluster types were considered: • stop + lateral (e.g. [bl-]) – 3 items • stop + rhotic (e.g. [tr-]) – 6 items • stop + nasal (e.g. [kn-]) – 1 item • s + lateral (e.g. [sl-]) – 2 items • s + nasal (e.g. [sn-]) – 2 items • s + stop (e.g. [st-]) – 5 items • NB: s + rhotic clusters are illicit in Latvian • Note: number of items per cluster type varies due to the structure of the Phoneme Test.

6. Data set: coding • Child responses were coded to reflect the production of the attempted cluster: • fully correct: /sluota/  [sluota] ‘broom’ • fully omitted: /gredzens/  [edents] ‘ring’ • retained with segmental changes: /kru:ze/  [klu:ze] ‘cup’ • reduced to the less sonorous element: /blu:ze/[bu:de] ‘blouse’ • reduced to the more sonorous element: /snieks/  [nieks] ‘snow’ • A profile was generated for each child showing their production patterns for each cluster type:

7. Data set: dealing with intra-speaker variation • Young children often demonstrate variable productions: i.e. not all inputs of the same type are consistently treated in the same way (e.g. Smith 1973, 2010) • Such variability is to be expected, since acquisition occurs gradually (Becker & Tessier 2010). In addition, both ‘precocious’ and ‘fossilized’ forms – i.e. individual ‘memorized’ words that do not reflect the current state of a child´s phonology – can contribute (Tessier 2009)

8. Data set: dealing with intra-speaker variation • OT literature: intra-speaker variation – in children and adults – is the result of partial constraint ranking (Becker & Tessier 2010, Anttila 1997), i.e. the situation where dominance relation between two or more constraints is not fixed. • In such cases, unranked constraints can assume any mutual ranking each time a speaker attempts some production, which leads to variable results: Attempt A: Cluster reduction Attempt B: No reduction

9. Data set: dealing with intra-speaker variation • In children, partial ranking is eventually resolved based on the positive evidence from the ambient language. • Crucially, the process is gradual: the proportion of non-target-like productions decreases slowly over time, and variability ‘dies out’ (see Becker & Tessier 2010 for technical implementation in OT). THEREFORE: • For children showing variation in their treatment of a given input: • if one of the patterns is preferred, we assumed it best reflects the state of child´s grammar, and disregarded the other one; • if no pattern is preferred, we assumed the existence of two grammar-states, and both were included into the analysis. • children who varied on more than one cluster type (N=2) were excluded.

10. Data set • Thus, 18 children gave us the total of 27 observations or 12 patterns. Pattern APattern B • Of these, 9 followed Pattern A, 6 followed Pattern B, and the remaining 12 were unique. • Bad news: acquisition paths vary a lot more than theoretical literature would have you believe. • Good news: more than half of the data is accounted for by just 2 patterns. Definitely not random!

11. The full typology

12. Principles of syllable phonotactics • The Sonority Hierarchy (Sievers 1881, Jespersen 1904, Kiparsky 1979, Selkirk 1984, Clements 1990...) High V – G – L – N – F – T Low(V = vowels; G = glides; L = liquids; N = nasals; F = fricatives; T = stops) • Sonority Sequencing (Selkirk 1984, Clements 1990): ‘Within a syllable, sonority rises towards the peak and then falls.’ • Sonority Distance in onsets (Steriade 1982, Selkirk 1984, Levin 1985, Zec 2007) :‘Within an onset, a bigger rise in sonority is better.’TG > TL > TN > TT or {*TT, *TT/TN, *TT/TN/TL, *ComplexOnset} • Nucleus Hierarchy: ‘Higher sonority is better in the nucleus.’V > L > N > F > T or {*T/nuc, *T,F/nuc, *T,F,N/nuc, *C/nuc} • Margin hierarchy: ‘Low sonority is better in the onset.’T > F > N > L or {Onset, *G/ons, *G,L/ons, *G,L,N/ons, *G,L,N,F/ons}

13. The appendix • Appendix is usually filled by only one segment, /s/ or /ʃ/ • Assumed to account for the sonority sequencing defying behaviour of sC onsets in many languages. • Some languages don’t allow “appendicitis”, e.g., Spanish (español). • Other languages allow appendix but no complex onsets • Spanish: *Appendix, Max >> Dep >> *ComplexOnsetOther languages:*ComplexOnset >> Max, Dep >> *Appendix

14. Constraints • Defining the constraints that shape our typology according to perceived wisdom: • *Appendix: Assign a * for every onset with an appendix. • *ComplexOnset (*Cplx): Assign a * for every complex onset. • Sonority Distance (SD): Assign a * for every complex onset with a sonority distance ≤ 2 (*{ST,SN,TN} • OnsetHarmony (OH): *L, *L/N, *L/N/S • Max: Assign a * for every input segment that is not present in the output. • Max-Initial (Max-1): Assign a * for every input segment at the left edge that is not present in the output.

15. Building the typology • To do a typology with OT Workplace, we abstract away from segmental changes.Sometimes L and R fluctuate.Sometimes only TL is present but not TR.Sometimes /pl/ is realized as [t], apparently coalescence rather than deletion. We count this nevertheless as reduction to least sonorant aka deletion of L. • We consider only deletion as an option. (No kid epenthesises vowels to break up consonant clusters anyway.) • Inputs: /SL, SN, ST, TL, TN/ • Input-output mappings/SL/ SL, S, L, S-L (hyphen = S in appendix)/SN/ SN, S, N, S-N/ST/ ST, S, T, S-T/TL/ TL, T, L/TN/ TN, T, N

16. The typology

17. The typology Phonetic target-like Target? GK GB GG Target? GF

18. Problem zones in the attested typology

19. Fixing problems – or what do the Latvian kids teach us about syllable phonotactics? • 8 Grammars undergenerated and 8 grammars overgenerated. • Overgeneration is not such a big deal: We had only 20 children. • Before we give up the Standard Theory: • What if syllabification is stratal? • Syllabification at Level I: Target-like • Add Faithfulness constraint on syllable structure to CON at level II.

20. Stratal OT in a nutshell Lexicon /input/ Stem level Evaluation with Ranking 1 GEN provides candidates Winner of Level I = input to level II GEN provides candidates Word level Evaluation with Ranking 2 Winner of Level II = input to level III Phrasal level Evaluation with Ranking 3 GEN provides candidates [output]

21. The other typology

22. The other typology Colour legend: Green = targetlike Light green = attested lg. Yellow: Attested in our data That other colour: Phonetically the same as one of our patterns

23. Fewer problem zones in the attested typology

24. Fixing remaining problems – or what else do the Latvian kids teach us about syllable phonotactics? • 4 Grammars undergenerated and 17 grammars overgenerated. • Overgeneration is not such a big deal: We had only 20 children. • Resilience of N to deletion in TN /kn/ 1: Perception interferes with grammar.Nasality cues mask stop release and stop’s place cues, children don’t perceive k. • Resilience of N to deletion in TN /kn/ 2: Articulation interferes with grammar.Timing of stop gestures and velum lowering difficult to coordinate. • Resilience of N to deletion in TN /kn/ 3: Possibly coalescence, as in /pl/ -> [t],N is stop anyway. • Resilience of N to deletion in TN /kn/ 4: N is stop anyway. Misanalysed as “low sonority nasal” (Krämer & Zec 2016) • GI: (L, N, T, TL, N) = 2nd C wins. We could add Contiguity to Con.Overgenerates *(L, N, S, N, L) • GL: (T, SN, T, T, T) We still wonder what that child is doing here.

25. Conclusions • The 20 weakest Latvian children produce patterns that • ...are not all attested in phonotactic typology • ...are not all generated by standard syllable theory. • Assuming stratal organisation of grammar and syllabification in the input to the last level matches results better, but overgenerates even more. • Though, we haven’t looked at complex onset tpyology in detail yet. • The remaining three child patterns can be explained by various grammatical and extragrammatical assumptions • Word of caution: We are looking at the worst out of 500. We’re potentially dealing with phonological pathologies, e.g., SLI.

26. Thank you!