Childrens preference for modal strength Paloma Jereti c - - PowerPoint PPT Presentation

Acquiring scope interactions of root modals and negation

Children’s preference for modal strength

Paloma Jeretiˇ c

paloma@nyu.edu

Child Language Lab October 10, 2018

The learning problem

Consider utterances (1) and (2)

(1) (2)

(1) that baby is very small and Xn’t have pie.

Mar´ ıa hears this at 1;11

(2) actually it’s really hot outside he Yn’t wear a scarf.

Madeleine hears this at 1;09

The learning problem

Consider utterances (1) and (2)

(3) (4)

(1) that baby is very small and Xn’t have pie.

Mar´ ıa hears this at 1;11

(2) actually it’s really hot outside he Yn’t wear a scarf.

Madeleine hears this at 1;09

◮ can we guess what X and Y are?

The learning problem

Consider utterances (1) and (2)

(5) (6)

(1) that baby is very small and Xn’t have pie.

Mar´ ıa hears this at 1;11

(2) actually it’s really hot outside he Yn’t wear a scarf.

Madeleine hears this at 1;09

◮ can we guess what X and Y are? ◮ at best, we can guess that (1) is a prohibition, and (2) a

denial of necessity

The learning problem

Consider utterances (1) and (2)

(7) (8)

(1) that baby is very small and Xn’t have pie.

Mar´ ıa hears this at 1;11

(2) actually it’s really hot outside he Yn’t wear a scarf.

Madeleine hears this at 1;09

◮ can we guess what X and Y are? ◮ at best, we can guess that (1) is a prohibition, and (2) a

denial of necessity

◮ but even if we know that X and Y are modals, both force and

scope with respect to negation are indeterminate, making it impossible from this data to guess the semantics of X and Y

The learning problem

Consider utterances (1) and (2)

(9) (10)

(1) that baby is very small and Xn’t have pie.

Mar´ ıa hears this at 1;11

(2) actually it’s really hot outside he Yn’t wear a scarf.

Madeleine hears this at 1;09

◮ what do kids do when confronted with this data?

The learning problem

Consider utterances (1) and (2)

(11) (12)

(1) that baby is very small and Xn’t have pie.

Mar´ ıa hears this at 1;11

(2) actually it’s really hot outside he Yn’t wear a scarf.

Madeleine hears this at 1;09

◮ what do kids do when confronted with this data? ◮ assuming they are able to learn from contextual cues (cf.

Tomasello, 2003), there are several possibilities:

The learning problem

Consider utterances (1) and (2)

(13) (14)

(1) that baby is very small and Xn’t have pie.

Mar´ ıa hears this at 1;11

(2) actually it’s really hot outside he Yn’t wear a scarf.

Madeleine hears this at 1;09

◮ what do kids do when confronted with this data? ◮ assuming they are able to learn from contextual cues (cf.

Tomasello, 2003), there are several possibilities:

◮ they infer the meaning of the whole modal+negation complex ◮ don’t assume anything about the meaning of the form ◮ have a default assumption about the meaning of the form ◮ they first learn the non-negated forms, then can infer scope

Scope interactions of root modals and negation

force of modal form

possibility

existential quantification

necessity

universal quantification

can must, have to peut faut, besoin puede debe, tiene que

Scope interactions of root modals and negation

force of modal form

possibility

existential quantification

necessity

universal quantification

can must, have to

+

peut faut, besoin puede debe, tiene que can’t mustn’t, don’t have to

¬

peut pas faut pas, pas besoin no puede no debe, no tiene que

Scope interactions of root modals and negation

force of modal form

possibility

existential quantification

necessity

universal quantification

strength of modal expression weak

logically equiva- lent to existential quantification

can

+

peut

NA

puede don’t have to

¬

pas besoin, doit pas no necesita, no tiene que

strong

logically equiva- lent to universal quantification

must

+

NA

faut tiene que can’t mustn’t

¬

peut pas faut pas, doit pas no puede no debe, no tiene que

Scope interactions of root modals and negation

force of modal form

possibility

existential quantification

necessity

universal quantification

strength of modal expression weak

logically equiva- lent to existential quantification

can

+

peut

NA

puede don’t have to

¬

?

pas besoin, doit pas no necesita, no tiene que

strong

logically equiva- lent to universal quantification

must

+

NA

faut tiene que can’t mustn’t

¬

peut pas faut pas, doit pas no puede no debe, no tiene que

Scope interactions of root modals and negation

force of modal form

possibility

existential quantification

necessity

universal quantification

strength of modal expression weak

logically equiva- lent to existential quantification

can

+

peut

NA

puede don’t have to

¬

?

pas besoin, doit pas no necesita, no tiene que

strong

logically equiva- lent to universal quantification

must

+

NA

faut tiene que can’t mustn’t

¬

peut pas faut pas, doit pas no puede no debe, no tiene que

◮ Does this asymmetric typology arise from a learning bias

active in acquisition?

Acquisition of modal force

◮ kids are probably not adult-like from the beginning

Acquisition of modal force

◮ kids are probably not adult-like from the beginning ◮ previous studies:

◮ English-acquiring children are sensitive to truth conditions and

relative force by age 5 (Hirst & Weil, 1982; Noveck et al., 1996, a.o.), but they may not compute scalar implicatures until age 7 (Noveck, 2001)

Acquisition of modal force

◮ kids are probably not adult-like from the beginning ◮ previous studies:

◮ English-acquiring children are sensitive to truth conditions and

relative force by age 5 (Hirst & Weil, 1982; Noveck et al., 1996, a.o.), but they may not compute scalar implicatures until age 7 (Noveck, 2001)

◮ Dieuleveut et al. (in prep.): relative to their input,

English-learning children produce more possibility than necessity modals, and more negation with possibility than with necessity

Child corpus study

◮ selected corpora from CHILDES (MacWhinney, 2000) of

including spontaneous French and Spanish speech of 11 children and their input

Language Corpus Child Age range # utterances up to age 3 child adult French Paris (Morgenstern & Parisse, 2007) Antoine 1;0–6;4 6931 36609 Julie 0;11–5;3 1898 9323 Madeleine 1;0–7;0 7503 19447 Th´ eophile 0;7–4;11 4211 22528 Anae 1;04–5;02 5859 15747 Spanish LlineasOjea Irene 0;11–3;02 11159 15419 Yasmin 1;10–2;9 2713 5510 L´

• pez Ornat (1994)

Mar´ ıa 1;7–4;0 8112 10320 Vila (1990) Emilio 0;11–4;08 4857 8058 Aguirre (2000) Magn 1;7–2;11 10921 12599 Remedi Vicky 1;10–2;11 1866 2497

Table: Corpora information

Coding

Coded child and adult speech for:

◮ strength

◮ intended [strong, weak] ◮ target [strong, weak, ambiguous]

◮ target force [existential, universal] ◮ sentential negation [absent, present] ◮ flavor [root, epistemic] ◮ utterance type [declarative, question] ◮ repetition [previous utterance, answer to question, song, self]

Results

• ¬

¬ ♦ ¬♦

must mustn’t not have to can can’t

CHI 252 32 2 244 43 ADU 1876 265 78 1326 317

Table: Counts of French forms

• ¬

¬ ♦ ¬♦

must mustn’t not have to can can’t

CHI 119 7 1 39 98 ADU 717 28 12 264 400

Table: Counts of Spanish forms

Results: comparing proportions

comparing French Spanish p-values CHI residuals p-values CHI residuals

• ♦

p = 0.001 (-1.95, +2.27) p = 0.584 ¬ ¬♦ p = 0.013 (-1.99, +0.94) p = 0.748 ¬ ♦ p = 0.013 (-2.73, +0.62) p > 0.999 ¬

• p = 0.010

(-2.21, +0.43) p > 0.999 ¬ ¬ p = 0.025 (-1.76, +0.91) p > 0.999 ♦ ¬♦ p = 0.084 p = 0.013 (-1.71, +1.34)

• ¬♦

p > 0.999 p = 0.008 (-1.44, +1.87)

• ¬

p = 0.590 p = 0.332 ¬ ¬♦ p = 0.639 p > 0.999

Aggregate results for χ2 or Fisher exact tests comparing forms across children and adults

Results: comparing proportions

comparing French Spanish p-values CHI residuals p-values CHI residuals 1 –

• ♦

p = 0.001 (-1.95, +2.27) p = 0.584 ¬ ¬♦ p = 0.013 (-1.99, +0.94) p = 0.748 ¬ ♦ p = 0.013 (-2.73, +0.62) p > 0.999 ¬

• p = 0.010

(-2.21, +0.43) p > 0.999 ¬ ¬ p = 0.025 (-1.76, +0.91) p > 0.999 ♦ ¬♦ p = 0.084 p = 0.013 (-1.71, +1.34)

• ¬♦

p > 0.999 p = 0.008 (-1.44, +1.87)

• ¬

p = 0.590 p = 0.332 ¬ ¬♦ p = 0.639 p > 0.999

Aggregate results for χ2 or Fisher exact tests comparing forms across children and adults

• 1. non-negated existentials are preferred over non-negated universals

(French)

Results: comparing proportions

comparing French Spanish p-values CHI residuals p-values CHI residuals 1 –

• ♦

p = 0.001 (-1.95, +2.27) p = 0.584 2      ¬ ¬♦ p = 0.013 (-1.99, +0.94) p = 0.748 ¬ ♦ p = 0.013 (-2.73, +0.62) p > 0.999 ¬

• p = 0.010

(-2.21, +0.43) p > 0.999 ¬ ¬ p = 0.025 (-1.76, +0.91) p > 0.999 ♦ ¬♦ p = 0.084 p = 0.013 (-1.71, +1.34)

• ¬♦

p > 0.999 p = 0.008 (-1.44, +1.87)

• ¬

p = 0.590 p = 0.332 ¬ ¬♦ p = 0.639 p > 0.999

Aggregate results for χ2 or Fisher exact tests comparing forms across children and adults

• 1. non-negated existentials are preferred over non-negated universals

(French)

• 2. weak negated universals are dispreferred relative to all other forms

(French)

Results: comparing proportions

comparing French Spanish p-values CHI residuals p-values CHI residuals 1 –

• ♦

p = 0.001 (-1.95, +2.27) p = 0.584 2      ¬ ¬♦ p = 0.013 (-1.99, +0.94) p = 0.748 ¬ ♦ p = 0.013 (-2.73, +0.62) p > 0.999 ¬

• p = 0.010

(-2.21, +0.43) p > 0.999 ¬ ¬ p = 0.025 (-1.76, +0.91) p > 0.999 3

• ♦

¬♦ p = 0.084 p = 0.013 (-1.71, +1.34)

• ¬♦

p > 0.999 p = 0.008 (-1.44, +1.87)

• ¬

p = 0.590 p = 0.332 ¬ ¬♦ p = 0.639 p > 0.999

Aggregate results for χ2 or Fisher exact tests comparing forms across children and adults

• 1. non-negated existentials are preferred over non-negated universals

(French)

• 2. weak negated universals are dispreferred relative to all other forms

(French)

• 3. negated existentials are preferred to non-negated forms (Spanish)

Results: Binomial Tests for concurrent acquisition

◮ Strength: 2 out of 5 French children and 2 out of 5 Spanish

children acquired strong forms before weak forms; the other children showed no significant results

Results: Binomial Tests for concurrent acquisition

◮ Strength: 2 out of 5 French children and 2 out of 5 Spanish

children acquired strong forms before weak forms; the other children showed no significant results

◮ Force: 2 out of 5 French children and 4 out of 5 Spanish

children acquired existentials before universals; the other children showed no significant results

Discussion

• 1. Bias away from weakness

◮ Evidence:

◮ dispreference for weak negated universals (don’t have to) in

French, relative to input

◮ preference for negated over non-negated existentials in

Spanish, relative to input

◮ acquisition of strong expressions before weak expressions

Discussion

• 1. Bias away from weakness

◮ Evidence:

◮ dispreference for weak negated universals (don’t have to) in

French, relative to input

◮ preference for negated over non-negated existentials in

Spanish, relative to input

◮ acquisition of strong expressions before weak expressions

◮ Possible explanation: weak is more difficult to produce than

strong (in particular when combined with negation)

Discussion: Weak is harder

Theoretical and experimental reasons for weakness to be harder:

Discussion: Weak is harder

Theoretical and experimental reasons for weakness to be harder:

◮ children have trouble with indeterminacy, i.e. entertaining

multiple representations at once (Ackerman, 1981; Acredolo & Horobin, 1987; ¨ Ozt¨ urk & Papafragou, 2015)

Discussion: Weak is harder

Theoretical and experimental reasons for weakness to be harder:

◮ children have trouble with indeterminacy, i.e. entertaining

multiple representations at once (Ackerman, 1981; Acredolo & Horobin, 1987; ¨ Ozt¨ urk & Papafragou, 2015)

◮ existential quantification involves entertaining multiple

representations at once by generating alternatives (at least in the nominal domain: Kratzer & Shimoyama, 2002, a.o.)

◮ ”you may do X” has the alternative ”you may do not X” or

”you may do Y”, for any contextually relevant Y

Discussion: Weak is harder

Theoretical and experimental reasons for weakness to be harder:

◮ children have trouble with indeterminacy, i.e. entertaining

multiple representations at once (Ackerman, 1981; Acredolo & Horobin, 1987; ¨ Ozt¨ urk & Papafragou, 2015)

◮ existential quantification involves entertaining multiple

representations at once by generating alternatives (at least in the nominal domain: Kratzer & Shimoyama, 2002, a.o.)

◮ ”you may do X” has the alternative ”you may do not X” or

”you may do Y”, for any contextually relevant Y

◮ Children are notoriously bad at generating alternatives

themselves up until 5-6 years old, at least for deriving scalar implicatures (Barner & Bachrach, 2010; Chierchia et al., 2001; Huang & Snedeker, 2009; Noveck, 2001; Papafragou, 2006; Skordos & Papafragou, 2016, a.o.)

Discussion

• 2. Bias toward existentials

◮ can’t be due to semantics, since there is a bias away from

weakness

Discussion

• 2. Bias toward existentials

◮ can’t be due to semantics, since there is a bias away from

weakness

◮ Evidence:

◮ preference for non-negated existentials compared to universals,

relative to input

Discussion

• 2. Bias toward existentials

◮ can’t be due to semantics, since there is a bias away from

weakness

◮ Evidence:

◮ preference for non-negated existentials compared to universals,

relative to input

◮ French kids use this form to ask for permission and request

(more than adults, and Spanish kids): not so much of interest

Discussion

• 2. Bias toward existentials

◮ can’t be due to semantics, since there is a bias away from

weakness

◮ Evidence:

◮ preference for non-negated existentials compared to universals,

relative to input

◮ French kids use this form to ask for permission and request

(more than adults, and Spanish kids): not so much of interest

◮ acquisition of existentials before universals ◮ existentials could be easier due to a simpler form-to-force

mapping is simpler: 1-to-1 for (for ♦) as opposed to 4-to-1 (for ) in Spanish; 2-to-1 (for ♦) as opposed to 6-to-1 (for ) in French

Discussion: back to the learning problem

◮ children are sensitive to the strength of the expression, and

first uses seem to involve negation + existential

Discussion: back to the learning problem

◮ children are sensitive to the strength of the expression, and

first uses seem to involve negation + existential

◮ they probably don’t learn bare forms before negated ones

Discussion: back to the learning problem

◮ children are sensitive to the strength of the expression, and

first uses seem to involve negation + existential

◮ they probably don’t learn bare forms before negated ones ◮ they learn top-down; it is unclear whether they know the force

• f the underlying item

Further work

◮ corpus study has a lot of limitations; results need to be

replicated

Further work

◮ corpus study has a lot of limitations; results need to be

replicated

◮ good news: Dieuleveut et al. (in prep.) come to similar

conclusions

Further work

◮ corpus study has a lot of limitations; results need to be

replicated

◮ good news: Dieuleveut et al. (in prep.) come to similar

conclusions

◮ ideas for follow-up experimental studies:

◮ Testing cost of weak+negation: a puppet says something that

is covered up with noise, and after having been given an appropriate context, the child has to pick between equivalent sentences ”you don’t have to leave” (¬) and ”you can stay” (♦), and ”you have to leave” () and ”you can’t stay” (¬)

◮ prediction: they pick non-negated statements more often with

weak than with strong

Further work

◮ corpus study has a lot of limitations; results need to be

replicated

◮ good news: Dieuleveut et al. (in prep.) come to similar

conclusions

◮ ideas for follow-up experimental studies:

◮ Testing cost of weak+negation: a puppet says something that

is covered up with noise, and after having been given an appropriate context, the child has to pick between equivalent sentences ”you don’t have to leave” (¬) and ”you can stay” (♦), and ”you have to leave” () and ”you can’t stay” (¬)

◮ prediction: they pick non-negated statements more often with

weak than with strong

◮ testing alternative generation with weakness: children see a

picture of an apple and a banana, they hear one of the following sentences: ”you can/don’t have to/can’t/have to have the apple”

◮ prediction: more eye movements between the fruits with weak

than with strong

Implications for typology

◮ Base syntactic order of root modals and negation:

Neg > Modal

(Hacquard 2010, Iatridou & Zeijlstra 2013) ◮ The scopal configuration Modal > Neg can be derived only

when the result is cognitively easier than the base order

◮ Neg > must (weak) harder than must > Neg (strong)

→ derived scopal configuration is possible

◮ Neg > can (strong) easier than can > Neg (weak)

→ derived scopal configuration is not possible

References I

Ackerman, B. P. (1981). Performative bias in children’s interpretations of ambiguous referential communications. Child Development, 1224–1230. Acredolo, C., & Horobin, K. (1987). Development of relational reasoning and avoidance of premature closure. Developmental Psychology, 23(1), 13. Aguirre, C. (2000). La adquisici´

• n de las categor´

ıas gramaticales en espa˜ nol. Ediciones de la Universidad Aut´

• noma de Madrid.

Barner, D., & Bachrach, A. (2010). Inference and exact numerical representation in early language development. Cognitive psychology, 60(1), 40–62. Chierchia, G., Crain, S., Guasti, M. T., Gualmini, A., & Meroni, L. (2001). The acquisition of disjunction: Evidence for a grammatical view of scalar

• implicatures. In Proceedings of the 25th boston university conference on

language development (pp. 157–168). Dieuleveut, A., van Dooren, A.-M., Cournane, A., & Hacquard, V. (in prep.). Children’s acquisition of modal force. Hirst, W., & Weil, J. (1982). Acquisition of epistemic and deontic meaning of modals. Journal of child language, 9(3), 659–666. Huang, Y. T., & Snedeker, J. (2009). Online interpretation of scalar quantifiers: Insight into the semantics–pragmatics interface. Cognitive psychology, 58(3), 376–415. Kratzer, A., & Shimoyama, J. (2002). Indeterminate pronouns: The view from japanese. L´

• pez Ornat, S. (1994). La adquisici´
• n de la lengua espa˜
• nola. Madrid: Siglo XXI.

References II

MacWhinney, B. (2000). The childes project: The database (Vol. 2). Psychology Press. Morgenstern, A., & Parisse, C. (2007). Codage et interpr´ etation du langage spontan´ e d’enfants de 1 ` a 3 ans. Corpus(6), 55–78. Noveck, I. A. (2001). When children are more logical than adults: Experimental investigations of scalar implicature. Cognition, 78(2), 165–188. Noveck, I. A., Ho, S., & Sera, M. (1996). Children’s understanding of epistemic

• modals. Journal of Child Language, 23(3), 621–643.

¨ Ozt¨ urk, O., & Papafragou, A. (2015). The acquisition of epistemic modality: From semantic meaning to pragmatic interpretation. Language Learning and Development, 11(3), 191–214. Papafragou, A. (2006). Epistemic modality and truth conditions. Lingua, 116(10), 1688–1702. Skordos, D., & Papafragou, A. (2016). Children’s derivation of scalar implicatures: Alternatives and relevance. Cognition, 153, 6–18. Tomasello, M. (2003). Constructing a language: A usage-based approach to child language acquisition. Cambridge (MA). Vila, I. (1990). Adquisici´

• n y desarrollo del lenguaje. Barcelona: Gra´
• .