Noun Categories in Children`s Production of Noun

Noun Categories in Children’s Production of Noun-Noun Compounds
Jessica Gamache, Kyle Latack, Mina Hirzel, Cristina Schmitt
Michigan State University
In word learning, children are rarely presented with unambiguous information about new
words, but rather receive an over-abundance of possibly relevant information. To produce or
comprehend noun-noun (NN) compounds, children must take information from two
independent nouns and establish a relation between them to yield a compound interpretation.
Every NN compound allows a variety of interpretations. A giraffe-mosquito (Adam 2;3) can
be interpreted as a mosquito inside a giraffe, a mosquito that bites giraffes, a long-necked
mosquito, etc. Krott et al. (2009) suggest that NN compounds are learned from frequencies of
particular nouns linked to particular relations in the input. In this paper we examine
frequencies of relations, noun categories, and particular nouns in Child and Child-Directed
Speech in order to determine what role frequency plays in children’s production of
compounds. We argue that the frequency with which children hear particular noun tokens
with particular types of compound relations is not the determining factor. What is important is
children’s knowledge of lexical information and noun categories (artifact, natural kind, etc).
Linguistic Background. The various relations that hold between the nouns in NN compounds
are not assumed to be derived from structure (Roeper & Snyder 2005, Delfitto et al. 2009,
Harley 2009). The different relations arise from the lexical properties of the two Ns, which
reduce to a small set of roles (Johnston & Busa 1996). In the word breadknife, part of our
lexical knowledge involves the fact that knives serve the telic role of cutting, and bread serves
as the object of cutting. The roles/relations are not derived from an arbitrary list (e.g., Lees
1960, Levi 1978), but rather by the lexical semantics of words, in line with the semantics of
modification. Besides the Telic role, we also consider the Constitutive role, which refers to a
material, content or part/whole relation and the Formal role, which refers to the properties that
distinguish an object within a larger domain (appearance, shape, etc.) (Pustejovsky 1991).
Acquisition Background. English children begin producing novel compounds around age 2
(Clark 1981, Snyder 1995) and by age 3-4 are adult-like in many aspects of compound
production (Clark et al. 1985). In production Hiramatsu et al. (2000) found that certain
relations, such as shape/appearance, surface late in speech, while Krott et al. (2009) found
that children were biased towards temporary, visible relations, such as location, and were best
at interpreting compounds with frequent head-relation combinations.
Methods. NN compounds were defined as two (or more) independent nouns combined
without extra morphology and were extracted from Adam (2;3-4;8) and Sarah (2;3-4;6) in the
Brown (1973) corpus in CHILDES (MacWhinney 2000). Table 1 shows coding categories.
Analyses reported here are in type, not token, counts and exclude compounds whose heads
were classified as Event, Location or Concept.
Results. Both corpora contained a high number of compounds and different N types within
those compounds (Table 2). There was little overlap (<30%) in the Ns used by child and
mother in both head and modifier positions. Contra Krott et al., children’s specific headrelation combinations differed from those of their mothers (Figure 1). If we consider N
categories, instead, children closely matched their mothers in terms of the relation used for
each head noun category (Figure 2). Comparing across N categories, the Telic (function) role
was used significantly more often with artifacts than with animate/inanimate natural kinds.
See Table 3 for results of accompanying χ2 analyses.
Conclusion. Adam and Sarah produced nearly as many not previously mentioned compounds
as their mothers, with very little overlap in the N-relations combinations used. Children are
not relying on frequencies of head-modifier combinations but are producing compounds based
on higher level generalizations about N categories and their lexical structure.
Table 1. Coding categories for compounds
Speaker to first utter compound in corpus
Child, Mother
Age of child when compound was first uttered
In months
Category of modifier noun
Human, Animate Natural Kind, Inanimate Natural
Category of head noun
Kind, Location, Event, Concept
Relation of compound
Constitutive, Formal, Telic
Table 2. Count of distinct compound and noun types by speaker of first mention
Adam
Mother
Sarah
Mother
Compound Types
NN
382
157
140
175
NNN
37
6
14
12
NNNN
2
0
0
0
Noun Types
Modifier nouns
229
130
99
113
Head nouns
194
118
101
112
Overlapping nouns
modifier: 54, head: 48
modifier: 26, head: 25
Table 3. χ2 results
Adam
Sarah
Relations by head noun category (child vs. mother)
Human
χ2(2, N = 60) = 8.267, p = .016
n.s. (p = .217)
Animate
n.s. (p = .768)
χ2(1, N = 21) = 3.360, p = .067
Inanimate
n.s. (p = .351)
n.s. (p = .361)
Artifact
n.s. (p = .194)
n.s. (p = .306)
Proportion of telic relations by head noun category (child vs. child)
Artifact X Human
χ2(1, N = 243) = 15.622, p < .001
χ2(1, N = 81) = 4.520, p = .034
Artifact X Animate
χ2(1, N = 256) = 15.306, p < .001
χ2(1, N = 87) = 7.308, p = .007
Artifact X Inanimate
χ2 (1, N = 219) = 3.007, p = .083
χ2 (1, N = 83) = 5.466, p = .019
Figure 1. Proportion of relation type by head noun category in child and mother compounds
Figure 2. Proportion of relation by head noun type in Adam and mother compounds
(only head Ns with 2+ compound types per speaker)
References
Clark (1981). In The Child’s Construction of Language. Clark et al. (1985). Child Development, 84-94.
Delfitto et al. (2009). In Proceedings of NELS, 39. Harley (2009). In Oxford Handbook of Compounding, 129144. Hiramatsu et al. (2000). In Proceedings of BUCLD. Johnston & Busa (1996). In Proceedings of the ACL
SIGLEX workshop, 77-88. Krott et al. (2009). Journal of child language, 36(1), 85-112. Lees (1960).
International Journal of American Linguistics. Levi (1978). The syntax and semantics of complex nominals.
MacWhinney (2000). The CHILDES Project. Pustejovsky (1991). Computational linguistics, 17(4), 409441.Roeper & Snyder (2005). In UG and external systems: Language, brain and computation, 155-169.
Snyder (1995). Language acquisition and language variation. Ph.D Thesis.