In their keynote article, Polinsky and Scontras (Polinsky & Scontras, Reference Polinsky and Scontras2019) highlight intriguing behavioral similarities between heritage language use and monolingual development (i.e., avoiding ambiguity, resisting irregularity, and shrinking structure). Moreover, just like monolingual children when compared to monolingual adults, heritage language speakers seem to have more limited input and more limited cognitive resources available to deploy. As Polinsky and Scontras note, heritage language speakers may well be in a state of frozen development, incomplete acquisition, or “developmental arrest”. This latter term is also used in the Williams Syndrome literature (Landau & Ferrara, Reference Landau and Ferrara2013) to describe what's occurring in an atypically-developing monolingual population that reaches around a five-year-old level of language knowledge. Because of this, the Williams Syndrome population may serve as another useful comparison population for heritage language speakers.
Given all this, I wonder if we can leverage existing comparative, behavioral, and computational approaches that are currently deployed to understand monolingual development in order to better understand heritage languages. We could then see how much the divergent attainment in heritage language speakers (when it occurs) resembles intermediate stages of monolingual development or the arrested development of Williams Syndrome populations. For instance, since we often have a reasonable idea about what's acquired earlier vs. later in monolingual children, does this serve as a reasonable marker of things acquired vs. not acquired in heritage language speakers, the way it seems to for Williams Syndrome populations? If so, this would allow us to better predict which phenomena are likely to be resilient vs. vulnerable in heritage language speakers.
Importantly, non-adult behavior in children can sometimes be mitigated with experiments designed to lessen the cognitive load, and so allow speakers to demonstrate underlying linguistic knowledge (e.g., Conroy, Takahashi, Lidz & Phillips, Reference Conroy, Takahashi, Lidz and Phillips2009; Viau & Lidz, Reference Viau and Lidz2011). These same approaches may therefore yield a clearer picture of heritage language speaker knowledge. Also, it may be that the non-monolingual behavior we see in heritage language speakers matches what a rational speaker, with adult-like cognitive resources, would do, given the input that heritage language speakers encounter. If so, we again have a way to predict heritage language knowledge – their perception of the input would be a key component of their observed learning outcomes. Below I sketch out a few instances of each developmental approach, based on the concrete examples Polinsky and Scontras discuss of heritage language use.
Comparative approaches
In heritage language speakers, Polinsky and Scontras note cases of overregularization, overly-strong tendencies compared to monolingual adults (e.g., a strengthened Subject bias when interpreting Spanish null pronouns), and non-adult interpretations (e.g., Russian verb ellipsis). Do monolingual children show these same behaviors for these same items? If they do, we often know something about (i) how long it takes children to recover and converge on the adult behavior, and (ii) what factors are believed to determine that recovery (e.g., perceiving variation as unpredictable can cause children to strengthen a probabilistic tendency: Hudson Kam & Newport, Reference Hudson Kam and Newport2009). Importantly, “how long” can often be translated into “how much data is required”. Given this, we can get more precise estimates of the input quantity necessary to support recovery from non-adult behavior in monolingual children; with reasonable samples of heritage language input, we can then see if this necessary quantity is typically available to heritage language speakers.
Behavioral approaches
Supportive contexts can allow children to display underlying linguistic knowledge that's hidden when more standard experimental setups are used (e.g., see Conroy et al., Reference Conroy, Takahashi, Lidz and Phillips2009 for a striking asymmetry in monolingual child pronoun interpretation with vs. without supportive pragmatic context). For heritage language speakers, the Chinese scope-taking and Japanese topic-marking behavior may be worth investigating using child behavioral techniques. Specifically for scope ambiguity, Viau and Lidz (Reference Viau and Lidz2011) demonstrate how supportive pragmatic context and structural priming allow monolingual children to display adult-like knowledge. Perhaps Chinese and Japanese heritage language speakers have more adult-like knowledge, but can't access it without additional contextual support.
Computational approaches
Computational tools can be deployed when we have realistic data samples of sufficient size, e.g., from CHILDES (MacWhinney, Reference MacWhinney2000) for monolingual child interactions. With reasonable samples of heritage language input distributions, we could use a variety of computational tools to predict learning behavior on the basis of that input (e.g., see Pearl, Reference Pearl and Sprousein press). More concretely, if we're interested in why heritage language speakers make or don't make certain generalizations and we have realistic input samples, we can use techniques that consider the average retrieval time (e.g., the Tolerance Principle: Yang, Reference Yang2016) or required storage space (e.g., information theoretical approaches: Chater, Clark, Goldsmith & Perfors, Reference Chater, Clark, Goldsmith and Perfors2015) with vs. without the generalization. On the basis of that input, these techniques will predict whether a rational learner (who prioritizes retrieval time or storage space) would make the generalization. We can then compare this to actual heritage language speaker behavior.
For example, perhaps overregularization or a simpler structure is perfectly rational, given the input heritage language speakers encounter. This strikes me as particularly relevant for the issue of morphology perception, where heritage language speakers may not perceive all the morphology in the input correctly all the time – or perhaps prefer to rely on some morphology more. With noisy input, a simpler structure may well be the rational generalization. More generally, we can also use computational tools to determine whether noisy input perception or preference for some information types over others (or both) is compatible with observed behavior (see Gagliardi, Feldman & Lidz, Reference Gagliardi, Feldman and Lidz2017 for a clear example of this with children).
Closing thoughts
I am excited by these possibilities for better understanding what's happening in heritage language speakers using developmental tools that already exist. One notable resource that may be worth creating is a repository of heritage language input samples, which would enable us to deploy the computational techniques especially and provide explanatory power for how heritage language speakers know what they do.
In their keynote article, Polinsky and Scontras (Polinsky & Scontras, Reference Polinsky and Scontras2019) highlight intriguing behavioral similarities between heritage language use and monolingual development (i.e., avoiding ambiguity, resisting irregularity, and shrinking structure). Moreover, just like monolingual children when compared to monolingual adults, heritage language speakers seem to have more limited input and more limited cognitive resources available to deploy. As Polinsky and Scontras note, heritage language speakers may well be in a state of frozen development, incomplete acquisition, or “developmental arrest”. This latter term is also used in the Williams Syndrome literature (Landau & Ferrara, Reference Landau and Ferrara2013) to describe what's occurring in an atypically-developing monolingual population that reaches around a five-year-old level of language knowledge. Because of this, the Williams Syndrome population may serve as another useful comparison population for heritage language speakers.
Given all this, I wonder if we can leverage existing comparative, behavioral, and computational approaches that are currently deployed to understand monolingual development in order to better understand heritage languages. We could then see how much the divergent attainment in heritage language speakers (when it occurs) resembles intermediate stages of monolingual development or the arrested development of Williams Syndrome populations. For instance, since we often have a reasonable idea about what's acquired earlier vs. later in monolingual children, does this serve as a reasonable marker of things acquired vs. not acquired in heritage language speakers, the way it seems to for Williams Syndrome populations? If so, this would allow us to better predict which phenomena are likely to be resilient vs. vulnerable in heritage language speakers.
Importantly, non-adult behavior in children can sometimes be mitigated with experiments designed to lessen the cognitive load, and so allow speakers to demonstrate underlying linguistic knowledge (e.g., Conroy, Takahashi, Lidz & Phillips, Reference Conroy, Takahashi, Lidz and Phillips2009; Viau & Lidz, Reference Viau and Lidz2011). These same approaches may therefore yield a clearer picture of heritage language speaker knowledge. Also, it may be that the non-monolingual behavior we see in heritage language speakers matches what a rational speaker, with adult-like cognitive resources, would do, given the input that heritage language speakers encounter. If so, we again have a way to predict heritage language knowledge – their perception of the input would be a key component of their observed learning outcomes. Below I sketch out a few instances of each developmental approach, based on the concrete examples Polinsky and Scontras discuss of heritage language use.
Comparative approaches
In heritage language speakers, Polinsky and Scontras note cases of overregularization, overly-strong tendencies compared to monolingual adults (e.g., a strengthened Subject bias when interpreting Spanish null pronouns), and non-adult interpretations (e.g., Russian verb ellipsis). Do monolingual children show these same behaviors for these same items? If they do, we often know something about (i) how long it takes children to recover and converge on the adult behavior, and (ii) what factors are believed to determine that recovery (e.g., perceiving variation as unpredictable can cause children to strengthen a probabilistic tendency: Hudson Kam & Newport, Reference Hudson Kam and Newport2009). Importantly, “how long” can often be translated into “how much data is required”. Given this, we can get more precise estimates of the input quantity necessary to support recovery from non-adult behavior in monolingual children; with reasonable samples of heritage language input, we can then see if this necessary quantity is typically available to heritage language speakers.
Behavioral approaches
Supportive contexts can allow children to display underlying linguistic knowledge that's hidden when more standard experimental setups are used (e.g., see Conroy et al., Reference Conroy, Takahashi, Lidz and Phillips2009 for a striking asymmetry in monolingual child pronoun interpretation with vs. without supportive pragmatic context). For heritage language speakers, the Chinese scope-taking and Japanese topic-marking behavior may be worth investigating using child behavioral techniques. Specifically for scope ambiguity, Viau and Lidz (Reference Viau and Lidz2011) demonstrate how supportive pragmatic context and structural priming allow monolingual children to display adult-like knowledge. Perhaps Chinese and Japanese heritage language speakers have more adult-like knowledge, but can't access it without additional contextual support.
Computational approaches
Computational tools can be deployed when we have realistic data samples of sufficient size, e.g., from CHILDES (MacWhinney, Reference MacWhinney2000) for monolingual child interactions. With reasonable samples of heritage language input distributions, we could use a variety of computational tools to predict learning behavior on the basis of that input (e.g., see Pearl, Reference Pearl and Sprousein press). More concretely, if we're interested in why heritage language speakers make or don't make certain generalizations and we have realistic input samples, we can use techniques that consider the average retrieval time (e.g., the Tolerance Principle: Yang, Reference Yang2016) or required storage space (e.g., information theoretical approaches: Chater, Clark, Goldsmith & Perfors, Reference Chater, Clark, Goldsmith and Perfors2015) with vs. without the generalization. On the basis of that input, these techniques will predict whether a rational learner (who prioritizes retrieval time or storage space) would make the generalization. We can then compare this to actual heritage language speaker behavior.
For example, perhaps overregularization or a simpler structure is perfectly rational, given the input heritage language speakers encounter. This strikes me as particularly relevant for the issue of morphology perception, where heritage language speakers may not perceive all the morphology in the input correctly all the time – or perhaps prefer to rely on some morphology more. With noisy input, a simpler structure may well be the rational generalization. More generally, we can also use computational tools to determine whether noisy input perception or preference for some information types over others (or both) is compatible with observed behavior (see Gagliardi, Feldman & Lidz, Reference Gagliardi, Feldman and Lidz2017 for a clear example of this with children).
Closing thoughts
I am excited by these possibilities for better understanding what's happening in heritage language speakers using developmental tools that already exist. One notable resource that may be worth creating is a repository of heritage language input samples, which would enable us to deploy the computational techniques especially and provide explanatory power for how heritage language speakers know what they do.