Psycholinguistics James Myers May 14, 2004 Language development II: Lexical stuff and syntax OVERVIEW: 1. Word learning 2. Syntactic development 3. Morphological development 4. Summary of language development ============================================================= 1. Word learning 1.1 Earliest words (from 12 mo.s onward) are concrete, dealing with here-and-now, with bias towards words that change in response to them: e.g. "ball" better than "chair" or "tree". Idiomorphs: "personalized words" (e.g. "ba" for "cup"). Strong evidence that even something as boring as word-learning doesn't really involve simple "imitation"! 1.2 Philosophical problems with word-learning: 1.2.1 John Locke (1690): "If we will observe how children learn languages, we will find that ... people ordinarily show them the thing whereof they would have them have the idea; and then repeat to them the name that stands for it...." [quoted in Gleitman 1990] This is called an "ostensive definition" (pointing and naming), and the child language pioneer Roger Brown called using this with kids the "original word game" But is it really sufficient for teaching kids words? 1.2.2 The "gavagai" problem (Quine 1960). A "simple" Behaviorist connection between a word and a situation is not at all simple: Imagine that you're an anthropological linguist is studying an unknown culture. Your native consultant points to a rabbit hopping across a field and says "Gavagai!" What does this mean? "Rabbit"? "Hop"? "Rabbit in a field"? "Rabbit ear"? "Rabbit ear + part of field"? Problem: The hypothesis space is too large. Some hypotheses can be distinguished by further experience (e.g. "rabbit" vs. "hop"), but others are unlikely to be (e.g. "rabbit" vs. "rabbit ear") -- and others cannot be, even in principle (e.g. "rabbit" vs. "set of attached rabbit parts"). 1.3 Evidence that children do have problems in the hypothesis space: Overextension: using a word for more referents than it should (e.g. calling all-4-legged animals "dog"). Mistakes may even be purely contextual (e.g. calling a crib blanket a "nap"); child has made the wrong link between the word and the context. Underextension: using a word with a narrower meaning than it should have (e.g. using "shoes" to mean just Mom's shoes). 1.4 Do adults help kids constrain the hypothesis space? Adults use basic level terms in child-directed speech. Basic level: the lexical level at which there are broad similarities among exemplars, so not too specific, and not too general. E.g. "bird", and not "robin" or "animal". By hearing primarily basic level terms, children can learn to stick to the basic level as well (and thus avoid over- and underextensions). Adults sometimes do in fact give ostensive definitions: Crucially, 95% of the time the labels used refer to whole objects, not parts (e.g., "gavagai" means "rabbit", not "rabbit's ear") [Ninio 1980] However, most studies have been done in Ango-American culture; it is not known if ostensive definitions are REQUIRED for learning nouns. Also, ostensive definitions are used mainly for nouns, not for verbs, adjectives, etc. 1.5 Perhaps the "gavagai" problem should be solved from the child side of the mystery, not the adult side: What biases do children show in their hypotheses? (i.e., are there some types of hypotheses that kids just naturally prefer?) 1.5.1 Whole object bias (also called "object principle", e.g. in Golinkoff et al. 1994): kids assume that nouns refer to whole objects. Is this learned from the adults' behavior? If so, how do the adults learn to use it? Could it be innate? How could you tell...?? (More on such questions next week.) 1.5.2 Taxonomic bias (also called "Categorical scope principle" in Golinkoff et al. 1994): kids assume that nouns refer to entire classes of objects (a "taxonomic category") Thus they assume that nouns tend not to refer to just one specific example ("avoid underextensions") How could this be learned? Kids only hear specific examples! (The mystery of induction is at the heart of language acquisition....) An example of a test of this bias: Markman and Hutchinson (1984): In experiments, 4 and 5-year-olds extended the name for a novel object if they share characteristics (i.e. are part of a single taxomonic category). E.g. both X and Y "swim in the water". So if X is called a "dax", then Y can be called a "dax" too. So that's what kids do. But not if they are merely associated: E.g. X "catches" Y. If X is called a "dax", this doesn't mean that Y should also be called "dax". And so kids behave accordingly. This argues against a simple Behaviorist theory of word learning, where associated objects will tend to be associated with the same name. Even more interestingly, the taxonomic effect depends on the use of NAMES: it is specifically a linguistic effect: Kids were shown novel objects, as above; they were taught their properties as above. Condition A: (no word used) "I'm going to show you something. Then I want you to think carefully, and find another one." Condition B: (novel word) "I'm going to show you a dax. Then I want you to think carefully, and find another dax." Results: A: 37% category responses B: 63% category responses (sig. higher) 1.5.3 Shape bias: (an aspect of the taxonomic bias) kids assume that shape plays a more important role in defining object categories than color, size, material, etc. This seems to play a role in the structure of noun classifier systems, too (e.g. Allan 1977). But since the shape bias grows more specific and strong over time (e.g. Landau et al. 1988) it seems less likely to be purely innate. 1.5.4 Mutual exclusivity bias (called the "novel name - nameless category principle" in Golinkoff et al. 1994): kids assume that words always have different meanings: no synonyms. Markman and Wachtel (1988): 3-yr-olds were shown two objects where they know a name for one of them; if they are asked to show the "dax", they will assume that this new word refers to the object that hasn't been given a name yet. Adults assume the same thing, by the way, e.g. in morphological "blocking" effects: rob robber steal *stealer (blocked by "thief") The Mutual Exclusivity Bias is "ranked higher" than the Whole Object Bias: (also Markman and Wachtel 1988) When 3-year-olds are given a novel name for a familiar object, e.g. a hammer is called a "dax", kids assume that "dax" refers to a PART of the hammer. Thus the Whole Object Bias is violated in order to preserve the Mutual Exclusivity Bias. 1.6 How kids narrow down the hypothesis space (summary): Child-directed speech has a narrow range of word-meaning relations; maybe kids can pick up on that. Kids have (innate?) biases about what hypotheses to consider first. But are there are other clues to word meaning in the input? 1.7 Gleitman (1990) and others say Yes: Syntactic bootstrapping: using syntax to constrain possible word meanings. 1.7.1 Note: a "bootstrap" is a strap on the back of a boot. "To lift oneself up by one's bootstraps" is a figurative expression that describes something physically impossible, but metaphorically it implies that you help yourself with a difficult problem rather than relying on outside help. In child language research, the first use of this term was in "semantic bootstrapping" (see below). There is also "prosodic bootstrapping": using prosody to constrain hypotheses about syntax (see "Segmentation" in last handout). 1.7.2 Classic evidence that kids can use syntax to constrain word meanings: Brown (1957): 3 to 5-year-old kids see an adult pouring some novel spaghetti-like stuff into a strange container. Different responses if kids were asked to show "some blick" (spaghetti), "a blick" (the container) or "blicking" (pouring). 1.7.3 Gleitman's (1990) evidence came from the study of how blind kids learn verbs of seeing (e.g. "see", "look", "watch"). Problem: Blind kids learn vision-related words quite easily, although they interpret them in a "haptic" sense (i.e. related to touching). Thus "look up" means "raise the hands". But still, they learn that "look" is active, while "see" is more passive. Their parents don't seem to be helping them, since they don't say "Look at that" only when an object is close enough for the kid to touch it. Thus the blind kids seem unable to use the situational context to learn the meanings: the "original word game" is impossible here! Yet this knowledge can't be all innate, since "look" vs. "see" is a particular property of English, and must be learned. Solution: Syntactic bookstrapping. Kids (even blind kids) can use the syntactic contexts of "look" vs. "see" to learn that they are perception verbs and yet mean different things. For example, you can say "Let's see if there's food in the refrigerator," which suggests that "see" is a verb of perception or cognition. Yet "see" can also be transitive ("see that"), while "look" requires a preposition (*"look that" vs. "look at that"), and this syntactic difference suggests differences in meaning, i.e. "see" is more direct than "look." 2. Syntactic development 2.1 Stages of syntactic development: Crude measure: mean length of utterance (MLU): the number of morphemes per utterance. Roger Brown proposed too many distinct stages. Probably only two or three: One-word stage (12-24 months): presyntactic? Two-word stage (24-36 months): learning basic argument structure? Everything else: sentences get gradually longer and more complex, depending a lot on the structure of the target language. 2.2 The beginning: 2.2.1 Kids start producing words around 12 months, remember? This is the start of the "one-word" stage. Holophrase: a whole proposition expressed with one word. e.g. "water" means "I want water" or "There is water" or "Give me water" or.... Important aspects of meaning expressed by gestures and context: e.g. "Mama" + looking at milk + whining = "Mom, give me some milk!" "Ball" + throwing a ball = "I'm throwing a ball." "Box" + putting crayons in box = "I'm putting crayons in a box" 2.2.2 Do they have syntax at this stage? Maybe not, since they seem to understand sentences merely by combining the semantics of the individual words: Adult tells kid "Get Mommy's shoe"; kid picks up toy shoe and gives it to Mommy. (Barrett 1982) But maybe they do...? Hirsh-Pasek et al. (1985) [cited in Gleitman (1990)]: 17-month-old American babies, mainly in the one-word stage, who used few if any verbs. They watched two TV's which showed familiar characters from the Sesame Street TV show: Cookie Monster and Big Bird. One TV showed CM tickling BB; the other TV showed BB tickling CM. The babies heard one of these sentences: (1) "Cookie Monster is tickling Big Bird." (2) "Big Bird is tickling Cookie Monster." Results: the babies looked longer at the correct TV scene, showing that they understood the basics of English syntax: Subject-Verb-Object. Hirsh-Pasek et al. (1988) [also cited in Gleitman 1990] found the same result for 24-month-olds with more subtle syntactic cues, and without any lexical cues: "Big Bird is flexing Cookie Monster." "Big Bird is flexing with Cookie Monster." Thus young kids can use syntactic cues in comprehension even before they can produce whole sentences. 2.3 Two-word stage (about 24 months). 2.3.1 Now we definitely got syntax! How do we know? Use primarily most informative content words: "More milk", "Push truck", leaving out less informative function words like "the", "and". Critique: Function words have less stress, and may be harder to segment out; also they are heard less often in isolation; so there may be non-syntactic explanation of this fact as well. (But what do you think of this critique...?) 2.3.2 Kids follow syntactic rules, even if they aren't the same as adult's rules: (Braine 1976) e.g. "Allgone sock" is more common with some kids than "Sock allgone"; the "wrong order" appears to be "ungrammatical". By the way, this simple fact puts 2-year-olds way ahead of the so-called "signing chimpanzees". Here are some sample "sentences" produced by Nim Chimpsky (Terrace 1979): Me banana you banana me you give. You me banana me banana you. Banana me me me eat. Give orange me give eat orange me eat orange give me eat orange give me you. (High MLU, but no syntax!) 2.3.3 But how "general" are child syntactic rules? Is there a general rule that says S -> Modifier + Modifee? More hot Allgone ball Or are the rules tied to specific words? e.g. S -> "more" + X (Braine 1976) More hot More car More crayon 2.3.4 One strategy that may help kids acquire syntax: Semantic bootstrapping: using the meanings of sentences (inferred from real-life context) to deduce syntactic categories and other syntactic properties of words E.g. if you know that "fip" means to speak in some way, then you can guess that "fip" is a verb, and you can also predict that it can be used in a wide variety of constructions, e.g. "Mary fipped", "Mary fipped 'Hello!'", "Mary fipped that she hates ice cream," "Mary fipped quietly"; but not *"Mary is fip" or *"Mary fipped Fred" or *"Mary was fipped." Of course, semantics alone is not enough: e.g. "suggestion" and "advice" have similar meanings, but different syntactic behaviors; in such cases, syntactic bootstrapping may be more helpful ("a suggestion" implies "suggestions") An example of testing the semantic bootstrapping hypothesis: Gropen, Pinker, Hollander and Goldberg (1991) Issue: Do kids use exceptionless "linking rules" to figure out which things in the real world are represented by direct objects in syntax? Linking rule: a hypothetical innate rule that kids use to link semantically defined things in the real world to syntactic categories (e.g. "object" to "noun"). One such rule might be: "Link the thing that is changed by the action (which is linked to the verb) to the verb's direct object." e.g. in "cook the food", the food is changed by cooking, so it should be the syntactic direct object. Figure-object (like figure in visual perception): the object that is directly changed by the verb: "fill": figure object is a container. So in "Fill the cup with water" the figure-object is "cup" Ground-object (like ground in vision): the object that is in the "background" to the action described by the verb: In "Fill the cup with water" the ground-object is "water" The verb "pour" has basically the same "meaning", but exactly the opposite argument structure: pour: figure-object is liquid ground-object is container as in "Pour water into the cup." Subjects: Kids from around 3 years of age up to adults Teach them novel verbs, controlling for figure-object and ground-object, but without using ANY object words while teaching Meaning focuses on ground: "keating" = "move in a zigzag manner" Meaning focuses on figure: "keating" = "cause something to sag" Results: The subjects who learned the "figure" meaning made the figure into a direct object; vice versa for the other case, which implies that they "know" the above linking rule. Even the youngest kids did this -- so is this linking rule innate (as Gropen et al. claim)? 2.4 Beyond the 2-word stage: 2.4.1 Some production milestones: Express more than one argument at a time: "Daddy throw ball", not just "Daddy throw" or "Throw ball". Negation, questions, passives all have tricky aspects that must be learned, including the associated morphology (details vary a lot from language to language, of course). Complex sentences: sentences with more than one sentential constituent: Coordination: "I love Mommy and Mommy loves me." Sentential complements: "I want [to go home]" Relative clauses: "I want [the ball [you dropped]]." 2.4.2 But maybe kids "know" such complex constructions before they can produce them "correctly": "Ungrammatical" productions may actually show that a young child has complex syntax, but just not the SAME syntax as the parents: Hamburger and Crain (1992): Kids as young as 24 months produce things that act like relative clauses, even though they don't obey the adult's rules: Child's speech Adult's speech 2 years: "This [my did it]" = "This is [what I did]" 2 yrs 4 mo: "Look-a my made ... Look-a wha I made" In a sense, this American child is basically using Chinese syntax! [my did it] = [wo zuo t ] de 2.4.3 Pinker (1984): continuity hypothesis: all people (including kids) have a complete grammar; thus even funny-sounding kids are speaking a possible human language. 2.4.4 Comprehension tasks reveal children's complex syntactic knowledge even more clearly (Crain 1991): Issue: Do kids have constraints on the comprehension of pronouns and wh-words? Pretty old kids (about 4 years) [this is after they produce sentential complements] Key sentence: (S) I know who he said has the best smile. S has the syntactic analysis in (a), not (b); thus it doesn't mean the same as (S'): (a) I know who(i) he said t(i) has the best smile. (b) *I know who(i) he(i) said t(i) has the best smile. (S') I know who said he has the best smile. Task: Experimenter tells a story about familiar characters. Kermit the Frog (a puppet) then says something about the story. If he's right, the child feeds Kermit a cookie as a reward. But sometimes Kermit is wrong, so then the child punishes Kermit by feeding him a rag. "Without the rag ploy children were reluctant to say that Kermit had said anything wrong." (Kids are biased to give positive answers.) Experimenter: "Last year's winner of the best smile contest was the Joker. That makes him this year's judge of the best smile. Here are the three people in the contest: Grover, one of the Teenage Mutant Ninja Turtles, and Yogi Bear. The Joker walks over to each contestant in turn. To Grover, he says 'Pretty good, big mouth.' To Yogi Bear, 'Not the best.' Then he looks at the Ninja Turtle and says 'Look at those teeth. You've definitely got the best smile.' But Grover says 'Joker, you're wrong. I have the best smile. Look at how big my mouth is.' And Yogi Bear says 'No, I have the best smile.'" Kermit: "I know who he said has the best smile. Grover and Yogi Bear!" Child: "No!" (feeds Kermit the rag) Results: Children correctly rejected the wrong interpretation about 95% of the time. They correctly accepted the right interpretation about 87% of the time. Mystery: When and how did kids learn this syntactic constraint? Or did they already "know" it from the very beginning (as Crain claims)? 3. Morphological development 3.1 Morphological development takes the longest to finish, since it's related to word learning, which of course never stops. Kids start using grammatical morphemes at around MLU = 2.5 (usually after second birthday), but this varies by language: in some languages (e.g. Spanish), nouns, verbs and adjectives always have inflectional morphology, so kids use the inflections as soon as they start using words. 3.2 Productivity of regular inflectional morphology 3.2.1 This was the first evidence that psycholinguists found that kids are developing a grammar, and not merely "imitating": they can produce new words that they never heard before. The classic experiment: Berko (1958): "Here is a wug. Now there are two of them. There are two ...?" 3.2.2 Overregularization: applying regular inflection to irregular forms (e.g. "breaked" or "goed") The classic U-shaped learning curve: (1): 100% correct ("broke", "went", "washed") [implies rote learning of past tense forms] (2): less than 100% ("breaked", "goed", "washed") [implies the discovery of the rule] (3): kids gradually return to 100% correct ("breaked", "went", "washed"; ...then later "broke", "went", "washed"...) [implies relearning exceptions word by word] More recent evidence shows that "U" is not very deep: kids overregularize only about 2% of the verbs that they know at any give time. (Marcus et al. 1992) 3.2.3 Nativism arguments again: Marcus (1995): U-curve is not directly caused by changes in input; the proportion of regular vs. irregular verbs in parents' speech does not change throughout the child's experience. Instead, overregularization begins as soon as kids learn that inflection is obligatory in English (i.e. learn that "walked" is two morphemes and not just one). That is, kids were "ready to overregularize" from the beginning; they don't need to "learn" that there's a rule; the existence of linguistic rules is therefore "innate" (Note: Marcus is one of Pinker's students!) 3.3 Development of less productive morphology: E.g. English "leave"-"left", "sane"-"sanity". This kind of morphology takes much longer to learn; sometimes not even until adulthood! Guy and Boyd (1990): /t/ and /d/ delete more often in unsuffixed forms like "past" than in suffixed forms like "passed." What about "sort of" suffixed forms like "left" (leave) and "told" (tell)? This class of words is not treated as a separate class by variable t/d deletion until speakers are in their late 20's or early 30's, if ever! Armbruster (1978): The probability that a college-age subject will change the vowel correctly in words like "sane"-"sanity" (using fake words) is correlated with their scores on a college entrance exam (the SAT again). That is, this kind of morphology requires a high level of linguistic awareness (including experience with reading). 4. Summary of language development (age is in months): Social Speech Speech Age cognition perception production Words Syntax --- --------- ---------- ---------- ----- ------ 0 Smile,cry Perceive prosody, cat perc. of non-native contrasts 1-8 ... ... Cooing Object Use prosody permanence to find clauses 8-12 Points, Lose cat. Babbling Use prosody inten- percept. to find tional of phrases commu- nonnative nication contrasts 12- Follows Phonology First One-word 18 pointing begins words stage; can use word order to comprehend 18- Mature Vocabulary 24 theory of explosion mind 24- Two-word 36 stage >36 Gradual increase in complexity REFERENCES Allan, K. (1977). Classifiers. Language, 53, 285-311. Armbruster, T. E. (1978). The psychological reality of the Vowel Shift and Laxing rules. University of California at Irvine PhD dissertation. Barrett, M. D. (1982). The holophrastic hypothesis: Conceptual and empirical issues. Cognition, 11, 47-76. Berko, J. (1958). The child's learning of English morphology. Word, 14, 150-177. Bloom, P. (ed.) (1993). Language acquisition: Core readings. Cambridge: MIT Press. Braine, M. D. S. (1976). Children's first word combinations. Monographs of the Society for Research in Child Development, 41 (1, Serial No. 164). Brown, R. (1957). Linguistic determinism and the parts of speech. Journal of Abnormal and Social Psychology, 55, 1-5. Crain, S. (1991). Language acquisition in the absence of experience. Behavioral and Brain Sciences, 14. Reprinted in Bloom (1993), pp. 364-409. Gleitman, L. (1990). The structural sources of verb meanings. Language Acquisition, 1, 3-55. Reprinted in Bloom (ed.), 174-221. Golinkoff, R. M., Mervis, C. B., & Hirsh-Pasek, K. (1994). Early object labels: The case for a developmental lexical principles framework. Journal of Child Language, 21, 125-155. Gropen, J., Pinker, S., Hollander, M., & Goldberg, R. (1991). Affectedness and direct objects: The role of lexical semantics in the acquisition of verb argument structure. Cognition, 41. Reprinted in Bloom (1993), 285-328. Guy, G. R., & Boyd, S. (1990). The development of a morphological class. Language Variation and Change, 2, 1-18. Hamburger, H., & Crain, S. (1982). Relative acquisition. In S. Kuczai (Ed.) Language development, Vol 1: Syntax and semantics (pp. 245-274). Lawrence Erlbaum. Landau, B., Smith, L., & Jones, S. (1988). The importance of shape in early lexical learning. Cognitive Development, 3, 299-321. Marcus, G. F. (1995). The acquisition of the English past tense in children and multilayered connectionist networks. Cognition, 56, 271-279. Marcus, G. F., Pinker, S., Ullman, M., Hollander, M., Rosen, T. J., & Xu, F. (1992). Overregularization in language acquisition. Monographs of the Society for Research in Child Development 57. Markman, E. M., & Hutchinson, J. E. (1984). Children's sensitivity to constraints on word meaning: Taxonomic versus thematic relations. Cognitive Psychology, 16, 1-27. Markman, E. M., & Wachtel, G. F. (1988). Children's use of mutual exclusivity to constrain the meanings of words. Cognitive Psychology, 20, 121-157. Ninio, A. (1980). Ostensive definition in vocabulary teaching. Journal of Child Language, 7, 565-573. Pinker, S. (1984). Language learnability and language development. Harvard University Press. Quine, W. V. O. (1960). Word and object. MIT Press. Terrace, H. S. (1979). Nim. Knopf.