GRS LX 700 Language Acquisition and Linguistic Theory Week 8b. The Critical Period Hypothesis L1A vs L2A Some properties of L1A: Fast Seemingly effortless Uniformly successful in reaching target. Some properties of L2A: Slow Hard Typically does not end in native-like ability. L1A vs L2A Few, possibly no, adults reach the level of tacit or unconscious knowledge of the grammar of the L2 that puts them on a par with native speakers. L2A often hits a brick wall after a certain point (“fossilization”), stuck with errors in variation with correct forms. L2A isn’t “equipotent”—the particular L1 and L2 pairing has an effect on the overall difficulty and problem areas. C. L1A: fast, easy, successful. A. L2A: slow, hard, failure-prone. Suggests that kids are “built to learn language” in a way that adults are not. Perhaps there is a “sensitive period” early in life where one absorbs languages? A sensitive period which ends at some point… Lenneberg 1967 Lenneberg 1967 (or Penfield and Roberts 1959) is usually considered to be the written origin of this idea that there is a “critical period” or “sensitive period” for language acquisition. He based this on several observations, including the observation that critical periods are biologically common. What makes us think there might be a critical period? Concerning L1A, there are (traumatic) cases of delayed language exposure which together seem to show that only if recovered before age 10 would normal L1 language development occur. This includes Genie (started at 13;7, learned some but stopped short of native-like attainment in morphology and syntax) What makes us think there might be a critical period? Another case of severely delayed language access (but without abuse) is Chelsea, misdiagnosed as retarded in early childhood, when in fact she was congenitally deaf—only discovered when Chelsea was 31. Chelsea’s utterances have almost no discernable structure at all; her speech was less language-like than Genie’s. How early is early enough? Isabelle (imprisoned with her mute, uneducated mother), starting at 6, rapidly caught up to normal age-levels. Jim, hearing child of deaf parents, brought into speech contact around 3;6, rapidly caught up in spoken language, reaching age-norms by 6. How early is early enough? Newport & Supalla’s study of ASL as L1 among congenitally deaf individuals, who started learning ASL at different ages. Exposure before 6 yields native competence, uniform error types (4-6 did slightly less well). Exposure after 7 yielded more errors in closed-class items, later correlated with evidence of more “holistically” (rote?) learned elements. Exposure after 12 much higher error rate and variable error types, more frozen forms. Seems clear enough There is some kind of advantage to L1A within the “sensitive period”. Is it language specific? Or is there something about overall cognitive development that can explain this? Once you get L1 within the sensitive period, is that good enough (does that “get it started”) for L2A even after the sensitive period? To reiterate… Is there a critical period for L1A? Evidence just reviewed suggests probably. Does this critical period affect L2A? Is it easier to learn an L2 inside the critical period? It is possible to learn an L2 outside the critical period? Does it just depend on having learned an L1 inside the critical period? (Kind of hard to distinguish from there being no critical period for L2A) About critical periods Just a note: It’s pretty uncontroversial that there is some decline in the ability to learn language that happens with age. Nobody disputes the fact that it’s harder to learn a second language later in life. The question is: Is this caused by an irreversible neurological change? (A critical period) Is it impossible to “learn an L2” after the end of the critical period? Or does it just get harder to learn stuff as you get older? Why does it seem to be particularly acute with language learning? About knowledge We can distinguish between two types of knowledge: language competence (acquired competence) learned linguistic knowledge The first is generally unavailable to conscious reflection. The second is quite often conscious. An L1 example of LLK is Don’t end your sentences with a preposition, which if followed threaten to result in travesties like: This is the sort of arrant pedantry up with which I will not put! About knowledge The critical period hypothesis is about obtaining acquired competence (not learned linguistic knowledge) and it makes a claim about whether an L2 speaker can obtain a native-like competence of an L2. People can always gain LLK in an L2 as well, learn rules, apply them, maybe get so practiced at it that it becomes second nature, but this still wouldn’t rise to the level of acquired competence. L2A and age of initial exposure Adults proceed through early stages of morphological and syntactic development faster than children (time and exposure constant). Older children acquire faster than younger children (morphology and syntax; time and exposure constant) Child starters outperform adult starters in the end. So, age improves rate, at least initially, but negatively affects ultimate level of attainment. Phonology—6 Other studies of phonological acquisition suggest that 6 years old is a critical one for attainment of native-like phonology. The period between 6-11 generally still results in some detectible accent. Generally tested by having native speaker judges listening (to accent, presumably) and guessing which were native speakers and which weren’t. Morphology, syntax, semantics—15 A few studies show that L2 speakers with an initial exposure prior to 15 did significantly better than L2 speakers with an initial exposure after 15 in the domain of syntax and morphology. Comprehension—10 A small set of results (Oyama 1978, Scovel 1981) suggest that ability to comprehend “masked” speech and recognize foreign accents has a discontinuity at around age 10. Several “critical periods” So it seems that there is an age-sensitivity, but it is not even language specific, it is subpart-of-language specific. Phonology—6 Morphology, syntax, semantics—15 Comprehension—10 …? Why isn’t it strange that there should be (a) critical period(s)? There are critical periods attested all over the biological world. The visual system is a favorite example. In experiments done on macaque monkeys, it was determined that there is a critical period for development of binocular vision cells in the visual cortex (tested by monocular deprivation) Recovery after CNS damage: disappointingly limited in the adult brain, but can be nearly 100% in the immature nervous system. Why isn’t it strange that there should be (a) critical period(s)? Vision studies replicated in cats (Hubel & Wiesel 1962, 1970). In fact, vision studies “replicated” in humans as well; there seems to be a visual critical period at around age 6, after which providing previously delayed visual stimuli is of no use. (Congenital opacities of the cornea; surgery performed on juveniles or adults does not restore sight) Imprinting in birds; just after birth, they “become attached” to a prominent moving object in their environment (typically, the mother). This attachment persists. But it can only be done sometimes in the first few hours, for some species. (For a demonstration, see the movie Fly Away Home) Why isn’t it strange that there should be (a) critical period(s)? …The development of form perception and the binocular vision necessary for depth perception proceed in stages after birth. Each stage culminates in one or more developmental decisions, many of which are irreversible. In each stage, appropriate sensory experiences are necessary to validate, shape, and update normal developmental processes. Consequently, the effects of sensory deprivation are most severe during a restricted and well-defined period early in postnatal life when these developmental decisions are still being made. (Kandel, Schwartz, Jessell 3d ed. 1991, p. 956) Why isn’t it strange that there should be (a) critical period(s)? …Critical periods of development generally do not have sharp time boundaries. Different layers within one region of the brain may have different critical periods of development, so that even after the critical period for one layer has passed, rearrangement of the layer may still be possible because the entire region has not yet fully developed. For example, 8 weeks after birth layer 4c in the visual cortex of the monkey is no longer affected by monocular deprivation, whereas the upper and lower layers continue to be susceptible for almost the entire first year.. (Kandel, Schwartz, Jessell 3d ed. 1991, p. 957) What might cause a critical period? Social / cognitive factors that covary with age? (an “intervening variable”); e.g., attitude, motivation, empathy, self-esteem, … Doesn’t seem to get at the uniformity across situations. Why phonology at 6, morphology at 15? Difference in the input? Unlikely to cause this big of an effect, and also unlikely to be as consistent as the facts require. Cognitive development provides other learning mechanisms which overwhelm our LAD mechanisms? Is this detectibly different? Is it even conceptually different? Brain development? Brain = mass of interconnected neurons. Certain areas of the brain have specific functions (visual cortex; auditory cortex; motor cortex) despite high levels of interconnectivity. Is the critical period due to loss of brain plasticity in the language area? Does language specifically have its own area? Localization Early evidence for localization came from aphasic patients—patients with specific linguistic deficits due to brain lesions, which could be correlated with location in an autopsy. Broca, French surgeon, 1861. Saw patient who lost had his ability to speak (could only utter the monosyllable tan except if agitated—reputedly often—when he could swear). Intelligence, comprehension spared Gradual paralysis of right side of the body. In autopsy, a lesion was discovered in what became known as “Broca’s area”—left hemisphere, frontal lobe. Broca’s area Spinning brain QuickTime™ and a Photo - JPEG decompressor are needed to see this picture. This came from here: http://brainmuseum.org/Specimens/primates/human/qtvrbrains.htm Lateralization Broca’s area is on the left hemisphere, not symmetrical. (Some very small variation with handedness—right-handed, almost exceptionless; left-handed, some variation). By now various regions of the brain have been correlated with certain kinds of aphasia Lateralization The two hemispheres of the brain also seem to have somewhat different functions. Left hemisphere generally controls the majority of language function. Right hemisphere appears to be involved in maintaining focus of attention, and also possibly prosody. Right hemisphere lesions have been known to severely affect ability to analyze metaphors, summarize complex texts, as well as disrupt prosody in otherwise normal language Child aphasia Acquired aphasia during childhood is almost never fluent (mutism), but they recover rapidly (lasting effects generally only slight word-finding and vocabulary difficulties). Recovery is faster, better than in adult acquired aphasia, but not complete. Early enough, right hemisphere can take over language functions after a serious loss in the left hemisphere, but it doesn’t do as good a job. Child aphasia Lenneberg’s summary of the results of left hemisphere lesions as a function of age: 0-3 months: no effect 21-36 months: all language accomplishments disappear; language is re-acquired with repetition of all stages. 3-10 years: aphasic symptoms, tendency for full recovery 11th year on: aphasic symptoms persist. Basis for his view that lateralization was tied to critical period. What might cause a critical period? Associated with lateralization of language processes in the process of brain development? Interesting, but the timing is probably off. Lateralization seems to be complete by around age 5, long before the syntax critical period. Maybe implicated in some way in the phonology critical period? What might cause a critical period? Brain development. Myelinization of axons precludes further connections (limits plasticity). Myelinization happens more slowly—in fact, it might miss the critical period on the other end, still going on after 15. Plus, we’d still like to know why the particular sequence we see, even if myelinization is the answer. Myelinization & neurons What might cause a critical period? Bottom line: We don’t really know. Neural development seems like a promising place to look, but there are very few things actually known about the connection between language and neurons, or even about neural development (beyond description). Johnson and Newport (1991) Aiming to test the critical period hypothesis by looking at correlations between eventual performance and age of initial exposure to the target language. In particular, they were trying to focus on whether purportedly universal properties of language exhibited in L2 show an age effect. Subjacency Johnson & Newport used grammaticality judgments to try to get at the language learners’ interlanguage competence, testing subtle contrasts that native speakers make. Their primary test looked at Subjacency violations (characterizing the possible whquestions in a language). Subjacency review Certain kinds of phrases that cannot contain the trace of a wh-movement. If you try to relate a whword at the beginning of the sentence to a trace inside one of these islands, the result is ungrammatical (or bad-sounding) sentence. *What did you ask whether John will buy — tomorrow? *Who did you see the book John gave — on the table? *What did you laugh after John brought — home? *What did John eat — and a muffin? Language variation Wh-in-situ languages tend also to allow a wh-question with the wh-word inside of an island to be asked (unlike in wh-movement languages). So, in Japanese, it is perfectly possible to ask (in Japanese): I saw [the book John gave who] on the table? I laughed [after John brought what home]? Subject-Auxiliary Inversion Johnson & Newport look at second language learners’ control of Subjacency in comparison to second language learners’ control of SubjectAuxiliary Inversion (TC movement). SAI is considered by them to be an “Englishspecific” rule (not a universal constraint like Subjacency, allowed by UG but in a sense not required by UG). Subject-Auxiliary Inversion So, what Johnson & Newport were assuming was essentially something like: When learning a language: (If the language has (wh-)movement), LAD is required to pick out the Subjacency rule and add it to the grammar of the language being built. A language may or may not opt to formulate a rule like SAI and add it to the language being built (language-particular, not provided by UG, Johnson & Newport (1991) J&N wanted to compare the ability of native speakers of Chinese (a wh-in-situ language) to learn/use Subjacency (a universal principle, provided by UG) and subjectauxiliary inversion (an English-specific rule, supposed to be part of English over and above UG). The idea is that if universal principles are provided by UG and there is a critical period, young learners (within the critical period) might have “access” to it whereas older learners might not (given that the L1 did not make use of Subjacency). J&N91: Study 1 Tested: declarative controls subjacency violations wh-questions satisfying subjacency SAI error (“English-specific”) simple wh-question controls (filter) Subjacency violations covered a number of possible settings for bounding nodes. J&N91: Study 1 results Adult learners (ChineseEnglish) did much worse (accepted ungrammatical sentences) than native speakers. L2’ers did better on SAI than on subjacency; subjacency doesn’t seem “privileged”. Response bias was ruled out; there is a slightly better than chance influence of subjacency in L2’ers. L2’ers seem to accept sentences that exemplify violations of subjacency with bounding nodes that hold in all languages. They verified that subjacency violations were by asking for answers—so we could tell where wh-words moved from. J&N91: Study 1 So, the adult learners didn’t do well at all on Subjacency tests—and not even better on Subjacency than SAI. And the actual responses didn’t seem to follow from a missetting of the bounding node parameters either. J&N91 (Study 2) Johnson & Newport looked at how second language learners fared with respect to Subjacency (“UG”) and Subject-Aux Inversion (“English-specific”), and what effect “initial age of immersion” had. They were looking for evidence of a critical period for language learning (in the form of “learning” the syntactic principle of Subjacency). J&N91 (Study 2) What’s the effect of initial age of immersion? 21 speakers ChineseEnglish with initial ages between 4-16. 21 more with initial ages between 17-25. J&N91 (Study 2) 36 34 32 30 28 Subjacency No-inversion 26 24 22 20 18 Native 4 to 7 8 to 13 14 to 16 adult J&N91 (Study 2) They conclude: Their results are incompatible with the view that nothing’s different between late and early L2 acquisition. There seems to be a more rapid drop-off of ability to use the putative universally available principle of Subjacency in one’s L2 if initial immersion is after 14 years old. But what question are we answering? White (2003, ch. 8) points out that J&N91 are really asking a different question: Does age-of-exposure/immersion affect a L2’ers achievement of the grammar of L2 that a native speaker would have? Answer seems to be yes. But does that mean the ultimate grammar (for late learners) is UG-unconstrained? Of course not, at least to the extent that there are languages for which native speakers could behave like J&N’s L2’ers. A possible interpretation Chinese allows topicalization, not derived by movement: zheben shu [[du guo pro de] ren] bu duo this book [[read ASP pro C] man] not many ‘This book, the people who read (it) aren’t many.’ If a fronted wh-phrase is reanalyzed as a kind of topic with a null resumptive, a language that looks like English without Subjacency would be the result. And it would still be UG-compliant. (Hawkins & Chan 1997) Those who disagree… Despite all of this, there are still those who maintain that there isn’t a critical period. The primary evidence brought in favor of this is that we can find isolated, rare instances of people who have learned a second language in their adult years (after a critical period should be over) who pass for native speakers on various kinds of tests. What are we to make of this kind of evidence? White & Genesee (1996) W&G are among the non-believers in a critical period. They don’t believe the results of previous studies are really representative of what level of competence is achievable. Instead, let’s find people who are likely candidates (near-natives) and test them (and compare their initial ages of immersion) White & Genesee (1996) Their subjects seem to distribute as you’d expect, though—the young learners are the near-natives, the old learners are the non-natives. Age groups Group 0-7 8-11 12-15 16+ Totals Near-native 22 7 7 9 45 Non-native 6 5 11 22 44 White & Genesee (1996) Their tests were grammaticality judgments and question formation tasks testing subjacency and also measuring reaction time. Their results from the GJ task showed that their categorizations of the subjects were right—the nearnatives performed like native speaker controls, and often significantly different from the non-native speakers. The QF task showed the same thing. White & Genesee (1996) W&G’s conclusion: It is possible for ultimate attainment to be native-like (to the point where you can’t experimentally tell a near-native from a native speaker). And there seems to be no particular effect among the near-natives of initial age of immersion. The age effect must be due to something else other than a “loss of UG”. White & Genesee (1996) Of course, English and French are a lot alike—is this an artifact of that? Did these L2’ers do so well because they could carry their parameter set over from French almost wholesale? Alluding to another study (White & Juffs 1996), W&G suggest no—Chinese not-quite-near-natives caught about the same number of ungrammatical sentences as native English speakers. So where are we? There is lots of evidence from neuroscience that some aspects of brain development are subject to critical periods. The evidence seems to show that people who start learning a second language relatively late are much less likely to approximate native speaker competence. The evidence may not quite manage to show that late learners cannot reach near-native levels. So is this inconsistent with a biological explanation? Are the “near-natives” just really good with LLK? Or, is it simply that the near-natives had to make use of something other than a (full-strength) LAD to get there? DeKeyser (2000) Adopts the familiar hypothesis that early language learning is due to unconscious, automatic, “implicit” acquisition and late language learning relies on more conscious “explicit” learning. Note: there is a similar distinction one can make between explicit and implicit knowledge (automatization, cf. driving a standard transmission car). These are two different things. One could imagine explicit learning procedures might still lead to implicit knowledge (cf. driving a standard transmission car). DeKeyser (2000) Basic prediction of the CPH: Late learners no longer have the implicit learning mechanism that early learners had. They must rely on analytic explicit learning procedures to learn language. There are individual differences between people in their analytic and verbal abilities. DeKeyser (2000) Therefore: All late-learning achievers of near-native status must have high verbal ability. Early-learning achievers of (near-)native status will not show any effect of verbal ability. Ran a Johnson & Newport-like study to see if these correlations hold. DeKeyser (2000) Tested 57 native speakers of Hungarian, all in the US for at least 10 years. Non-Indo-European, quite different from English in many respects. Almost no exposure to English prior to moving to an English-speaking country. Used modified version of Johnson & Newport’s grammaticality judgment task, then tested on a Hungarian verbal aptitude test. DeKeyser (2000) Aptitude test: Aptitude scores did not correlate with Average 4.7 of 20, std. dev. 2.79. 6 or above (+.46) was considered “high aptitude”. Resulted in 15 individuals. Age of arrival GJ test score (whole group) GJ test score (early learners only < 16) But did correlate significantly with GJ test score (late learners only ≥ 16) DeKeyser (2000) Several types of items on the GJ task. High correlation with age of arrival on: Tom working in his office right now. Tom is reading book in the bathtub. The beauty is something that lasts forever. I need to get some informations about the train schedule. What Martha is bringing to the party? Who you meet at the park every day? I want you will go to the store now. The student eats quickly his meals. DeKeyser (2000) Low correlation with age of arrival on: The dinner the man burned. The woman the policeman asked a question. The students to the movies went. Bites the dog. Knows John the answer to that question? The girl cut himself on a piece of glass. DeKeyser (2000) So, different things seem to be differently affected by the age effects, but there are significant age-of-arrival effects on many of the items. Looking now at the few late learners who did achieve a high test score, we find that they all* had high verbal aptitude scores too. *One didn’t, but DeKeyser argued that his score wasn’t representative of his analytical ability. DeKeyser (2000) Early learners got high test scores regardless of their aptitude scores; the only late learners to get high test scores had high aptitude scores. Years of schooling did not correlate with GJ scores. Exactly as predicted if post-CPH learners have to rely on more explicit learning mechanisms to learn a second language. But note! that this doesn’t mean that there isn’t some kind of implicit learning happening, only that it seems to be facilitated by explicit learning. DeKeyser (2000) Some structures, still, showed no correlation with aptitude—everybody got them, regardless of age-of-arrival, regardless of aptitude. Why? DeKeyser suggests it is a function of salience. SAI and do-support in yes-no questions (initial), pronoun gender (corrected), basic word order (initial, final). Concludes: CPH exists and constrains implicit learning mechanisms. So where are we? The onset of language takes place at early infancy, if not already at birth. At least by 6 months, infants are able to discriminate linguistic sounds (phonetic inventories, open syllables) from one another and from non-linguistic sounds. So where are we? There is an initial sensitive period for phonetic perception that is already over at 10-12 months of age but that appears to be overcomeable at least to some extent. Prior to this, children can discriminate linguistic sounds not only from the language they are learning as a native language, but also from other languages as well. After this, their ability wanes, although it seems to still be possible even for adult learners to regain the ability to distinguish non-native sounds with training or with the right experimental conditions. So where are we? Delayed first language acquisition is incomplete when the onset of language is after age 4; the later the age of onset, the less complete acquisition is likely to be. “Complete” = “as measured against the normally developing linguistic community” Newport (1990) studied congenitally deaf adults with different initial ages of exposure to ASL and found that even those whose initial age of exposure was as early as four were outperformed by those whose initial age of exposure was prior. So where are we? Late first language acquisition is less successful in the long run than equally late second language acquisition. Many studies combined show this sort of effect; it appears to be vital to learn a native language early, whereas the “window” doesn’t seem to completely close on highlysuccessful second language acquisition until quite a bit later. So where are we? More mature learners generally make faster initial progress in acquiring morphosyntactic and lexical aspects of second language. The general idea here is that more mature learners have more advanced general cognitive processes and problem-solving ability that allows them to better deal with the task of learning the morphology and syntax as a problem to be solved. Perhaps this is indicative of a role for LLK? In the long run, though, more mature learners are generally less successful. So where are we? An increasing age of onset for second language acquisition is correlated with declining ultimate attainment in pronunciation and morphosyntax across age groups, this pattern beginning typically with an onset age of 6 to 7 in childhood and continuing into adulthood. In adult learners, the association between onset age and declining outcomes is most strongly manifested in the oral aspects of second language proficiency. Learning a second language without an accent is very difficult after quite an early age. So where are we? Second language studies have not provided any real support for a critical period terminus at puberty, just somewhere. Some adult learners are capable of nearnative, if not native-like, performance in a second language, whereas some children are less successful than others. Puberty is another biologically scheduled process that is tempting to compare with a “critical period” for language acquisition. However, puberty is not itself contemporaneous with any observable linguistic milestone—it appears to be also maturational, but not directly linked to linguistic capacities. Whatever critical period there is, it seems to be somewhat “overcomable” either with effort or perhaps in terms of individual differences…?