Titone and Tiv (Reference Titone and Tiv2022) describe recent research and new questions to illustrate the power of the systems framework of bilingualism. But this framework does more than suggest new questions. My purpose here is to present 3 examples of how multicausal frameworks can illuminate longstanding topics in bilingualism.
• A systems framework disrupts the illusion of causality that has obscured the relationship between age of immigration and L2 ultimate attainment.
• A systems framework can prevent missteps in language-policy, as in the case of profoundly deaf individuals growing up without an accessible language.
• A systems framework, and particularly network science, can organize myriad factors to facilitate agent-based modeling of the decision to invest resources in learning a new language vs. maintaining expertise in a known language.
Critical (or sensitive periods) for L2 acquisition: the illusion that age is causal
When humans learn of two associated events, the brain's meaning-making machinery rapidly activates existing causal schemas to explain the association. Consider graphs showing the reliable, strong decrease in ultimate L2 attainment as a function of age of acquisition (AoA) (e.g., Vanhove, Reference Vanhove2013). One outcome is intuition that a brain-based sensitive period for L2 acquisition caused the decline.
Training students to be skeptical of causal illusions is common in psychology (Huck, Reference Huck1979). Consider the statement that children exposed to the arts end up with higher math and science scores than students not exposed to the arts. My students will eventually think of third variables that are correlated with the X and Y, such as parental or school financial resources differing for students exposed vs. not exposed to the arts. Some complain: “But you never said other things could be different!” Indeed.
Two situations inhibit the cognitive-miser solution that everything else is the same, and therefore, Y was caused by X. Having a stake in a non-causal outcome compels decision makers to search for alternative explanations; an example is rejecting the genetic essentialism suggested by data that black children have lower IQ scores than white children (Flynn, Reference Flynn2008).
The other situation is having a wealth of knowledge about what other factors vary with the X and Y variables. Here is where the social infusion aspect of the systems framework of bilingualism mutes the inference that brain changes are the cause of AoA effects. Familiarity with the nested spheres of influence surrounding language learners (as in Figure 1 from Titone & Tiv, Reference Titone and Tiv2022) alert us to how children resist learning a second language if they are able to use their L1 to communicate; we connect with information that immigrant children encounter diverse support for L2 learning while adults typically do not; we know that the language directed at children is more comprehensible than language directed to adults (Caldwell-Harris & MacWhinney, Reference Caldwell-Harris and MacWhinneyunder review). Nonetheless, Montrul wrote (Reference Montrul2020, p. 16) “Actual language loss is rare in adults, but it is not in children exposed to the same environmental conditions…” Even experts can forget that age organizes language learning, and thus the environment for learning is not held constant for child and adult learners.
The systems framework prompts us to adopt a multi-causal approach. Brain changes may well be a part of the explanation for age effects, but language learning is a complex dynamical system, with myriad moving parts. All the embedded social spheres described by Titone and Tiv (Reference Titone and Tiv2022) have their influence.
Deaf children's English reading abilities are superior if they don't use sign language
Antia, Lederberg, Easterbrooks, Schick, Branum-Martin, Connor, and Webb (Reference Antia, Lederberg, Easterbrooks, Schick, Branum-Martin, Connor and Webb2020) measured English reading ability in deaf children who were trained in only spoken English compared to those who learned only sign language. Graphs depicting this data (see Antia et al.'s Reference Antia, Lederberg, Easterbrooks, Schick, Branum-Martin, Connor and Webb2020 supplementary files) visually convey a compelling conclusion: the best route for achieving English literacy is to promote spoken language learning and restrict or forbid using sign language. That inference has influenced policy for decades (Lane, Reference Lane1992).
Deaf children are never randomly assigned to sign-only vs. spoken-only language acquisition treatment. Instead, some deaf children have sufficient residual hearing or for somewhat mysterious reasons can succeed with lip reading. These children then succeed, at least to some degree, in acquiring spoken English as their first language. Children without that residual hearing or lipreading abilities cannot easily extract language information from interacting with speech. They can have normal cognitive and social development when exposed to sign language from birth (Hoffmeister, Reference Hoffmeister, Chamberlain, Morford and Mayberry2000). Once in the public school system, these children learn English as a second language via print, a notoriously hard task (Caldwell-Harris, Reference Caldwell-Harris, Enns, Henner and McQuarrie2021).
Titone and Tiv (Reference Titone and Tiv2022) mention political and economic interests as part of the nested spheres of influence; awareness of the role these have played in deaf education (Lane, Reference Lane1992) can alter inferences drawn from simple correlational data concerning deaf children with different learning histories.
An organizational framework to facilitate agent-based modeling
The prior examples involved how a multicausal framework can disrupt single-factor reasoning. My next case involves how the nested spheres of social influence (Figure 1 in Titone & Tiv, Reference Titone and Tiv2022; and a related framework (Atkinson, Byrnes, Doran, Duff, Ellis, Hall, Johnson, Lantolf, Larsen–Freeman, Negueruela, Norton, Ortega, Schumann, Swain & Tarone, Reference Atkinson, Byrnes, Doran, Duff, Ellis, Hall, Johnson, Lantolf, Larsen–Freeman, Negueruela, Norton, Ortega, Schumann, Swain and Tarone2016) can organize complex computer modeling endeavors. An example is my own attempt to create an agent-based model of language learning following immigration (Caldwell-Harris, Reference Caldwell-Harris, Diallo, Wildman, LeRon Shults and Tolk2019).
No agent-based model exists of how multilinguals choose to invest in L1 vs. L2. Titone and Tiv (Reference Titone and Tiv2022) discuss social features of language use, which fit with network models (Vitevitch, Reference Vitevitch2019), such as learning via social interaction and being influenced by other social actors. Agent based modeling has these and other characteristics, making it a good fit to agent-based modeling, such as diffusion/adoption effects (Bonabeau, Reference Bonabeau2002). Language learning involves positive and negative feedback loops, such that poor progress early in learning can spiral into negativity and avoidance, while early success can accelerate learning. An implemented model would provide simulations to test theories about why language learning is difficult for people occupying different linguistic ecosystems (Caldwell-Harris, Reference Caldwell-Harris, Diallo, Wildman, LeRon Shults and Tolk2019).
I argued for the relevance of pertinent factors, but organizing entities, state variables, scales to represent brains, learning, family, neighborhoods, and societal values proved too challenging to aim for an implemented model. I used the Overview, Design Concepts and Details protocol (ODD; Grimm, Berger, Bastiansen, Eliassen, Ginot, Giske, Goss-Custard, Grand, Heinz, Huse, Huth, Jepsen, Jørgensen, Mooij, Müller, Pe'er, Piou, Railsback, Robbins, Robbins, Rossmanith, Rüger, Strand, Souissi, Stillman, Vabø, Visser & Deangelis, Reference Grimm, Berger, Bastiansen, Eliassen, Ginot, Giske, Goss-Custard, Grand, Heinz, Huse, Huth, Jepsen, Jørgensen, Mooij, Müller, Pe'er, Piou, Railsback, Robbins, Robbins, Rossmanith, Rüger, Strand, Souissi, Stillman, Vabø, Visser and Deangelis2006) to organize a narrative description. I identified outcome variables (frequency of use and fluency in the two languages), frequent actors, and defined rules for initiating/continuing conversation, and rules for agents to move to new locations. Titone and Tiv's (Reference Titone and Tiv2022) framework suggests that an additional organizing feature would be to use the nested spheres of influence, and to draw on complex variables such as language entropy to organize both a description and implementation.
Summary
Titone and Tiv (Reference Titone and Tiv2022) have been forward thinking in describing diverse new research and new questions that exemplify systems thinking about multilingual experiences. I make the case here that longstanding questions will also benefit from being incorporated into a multicausal, dynamic framework.
Titone and Tiv (Reference Titone and Tiv2022) describe recent research and new questions to illustrate the power of the systems framework of bilingualism. But this framework does more than suggest new questions. My purpose here is to present 3 examples of how multicausal frameworks can illuminate longstanding topics in bilingualism.
• A systems framework disrupts the illusion of causality that has obscured the relationship between age of immigration and L2 ultimate attainment.
• A systems framework can prevent missteps in language-policy, as in the case of profoundly deaf individuals growing up without an accessible language.
• A systems framework, and particularly network science, can organize myriad factors to facilitate agent-based modeling of the decision to invest resources in learning a new language vs. maintaining expertise in a known language.
Critical (or sensitive periods) for L2 acquisition: the illusion that age is causal
When humans learn of two associated events, the brain's meaning-making machinery rapidly activates existing causal schemas to explain the association. Consider graphs showing the reliable, strong decrease in ultimate L2 attainment as a function of age of acquisition (AoA) (e.g., Vanhove, Reference Vanhove2013). One outcome is intuition that a brain-based sensitive period for L2 acquisition caused the decline.
Training students to be skeptical of causal illusions is common in psychology (Huck, Reference Huck1979). Consider the statement that children exposed to the arts end up with higher math and science scores than students not exposed to the arts. My students will eventually think of third variables that are correlated with the X and Y, such as parental or school financial resources differing for students exposed vs. not exposed to the arts. Some complain: “But you never said other things could be different!” Indeed.
Two situations inhibit the cognitive-miser solution that everything else is the same, and therefore, Y was caused by X. Having a stake in a non-causal outcome compels decision makers to search for alternative explanations; an example is rejecting the genetic essentialism suggested by data that black children have lower IQ scores than white children (Flynn, Reference Flynn2008).
The other situation is having a wealth of knowledge about what other factors vary with the X and Y variables. Here is where the social infusion aspect of the systems framework of bilingualism mutes the inference that brain changes are the cause of AoA effects. Familiarity with the nested spheres of influence surrounding language learners (as in Figure 1 from Titone & Tiv, Reference Titone and Tiv2022) alert us to how children resist learning a second language if they are able to use their L1 to communicate; we connect with information that immigrant children encounter diverse support for L2 learning while adults typically do not; we know that the language directed at children is more comprehensible than language directed to adults (Caldwell-Harris & MacWhinney, Reference Caldwell-Harris and MacWhinneyunder review). Nonetheless, Montrul wrote (Reference Montrul2020, p. 16) “Actual language loss is rare in adults, but it is not in children exposed to the same environmental conditions…” Even experts can forget that age organizes language learning, and thus the environment for learning is not held constant for child and adult learners.
The systems framework prompts us to adopt a multi-causal approach. Brain changes may well be a part of the explanation for age effects, but language learning is a complex dynamical system, with myriad moving parts. All the embedded social spheres described by Titone and Tiv (Reference Titone and Tiv2022) have their influence.
Deaf children's English reading abilities are superior if they don't use sign language
Antia, Lederberg, Easterbrooks, Schick, Branum-Martin, Connor, and Webb (Reference Antia, Lederberg, Easterbrooks, Schick, Branum-Martin, Connor and Webb2020) measured English reading ability in deaf children who were trained in only spoken English compared to those who learned only sign language. Graphs depicting this data (see Antia et al.'s Reference Antia, Lederberg, Easterbrooks, Schick, Branum-Martin, Connor and Webb2020 supplementary files) visually convey a compelling conclusion: the best route for achieving English literacy is to promote spoken language learning and restrict or forbid using sign language. That inference has influenced policy for decades (Lane, Reference Lane1992).
Deaf children are never randomly assigned to sign-only vs. spoken-only language acquisition treatment. Instead, some deaf children have sufficient residual hearing or for somewhat mysterious reasons can succeed with lip reading. These children then succeed, at least to some degree, in acquiring spoken English as their first language. Children without that residual hearing or lipreading abilities cannot easily extract language information from interacting with speech. They can have normal cognitive and social development when exposed to sign language from birth (Hoffmeister, Reference Hoffmeister, Chamberlain, Morford and Mayberry2000). Once in the public school system, these children learn English as a second language via print, a notoriously hard task (Caldwell-Harris, Reference Caldwell-Harris, Enns, Henner and McQuarrie2021).
Titone and Tiv (Reference Titone and Tiv2022) mention political and economic interests as part of the nested spheres of influence; awareness of the role these have played in deaf education (Lane, Reference Lane1992) can alter inferences drawn from simple correlational data concerning deaf children with different learning histories.
An organizational framework to facilitate agent-based modeling
The prior examples involved how a multicausal framework can disrupt single-factor reasoning. My next case involves how the nested spheres of social influence (Figure 1 in Titone & Tiv, Reference Titone and Tiv2022; and a related framework (Atkinson, Byrnes, Doran, Duff, Ellis, Hall, Johnson, Lantolf, Larsen–Freeman, Negueruela, Norton, Ortega, Schumann, Swain & Tarone, Reference Atkinson, Byrnes, Doran, Duff, Ellis, Hall, Johnson, Lantolf, Larsen–Freeman, Negueruela, Norton, Ortega, Schumann, Swain and Tarone2016) can organize complex computer modeling endeavors. An example is my own attempt to create an agent-based model of language learning following immigration (Caldwell-Harris, Reference Caldwell-Harris, Diallo, Wildman, LeRon Shults and Tolk2019).
No agent-based model exists of how multilinguals choose to invest in L1 vs. L2. Titone and Tiv (Reference Titone and Tiv2022) discuss social features of language use, which fit with network models (Vitevitch, Reference Vitevitch2019), such as learning via social interaction and being influenced by other social actors. Agent based modeling has these and other characteristics, making it a good fit to agent-based modeling, such as diffusion/adoption effects (Bonabeau, Reference Bonabeau2002). Language learning involves positive and negative feedback loops, such that poor progress early in learning can spiral into negativity and avoidance, while early success can accelerate learning. An implemented model would provide simulations to test theories about why language learning is difficult for people occupying different linguistic ecosystems (Caldwell-Harris, Reference Caldwell-Harris, Diallo, Wildman, LeRon Shults and Tolk2019).
I argued for the relevance of pertinent factors, but organizing entities, state variables, scales to represent brains, learning, family, neighborhoods, and societal values proved too challenging to aim for an implemented model. I used the Overview, Design Concepts and Details protocol (ODD; Grimm, Berger, Bastiansen, Eliassen, Ginot, Giske, Goss-Custard, Grand, Heinz, Huse, Huth, Jepsen, Jørgensen, Mooij, Müller, Pe'er, Piou, Railsback, Robbins, Robbins, Rossmanith, Rüger, Strand, Souissi, Stillman, Vabø, Visser & Deangelis, Reference Grimm, Berger, Bastiansen, Eliassen, Ginot, Giske, Goss-Custard, Grand, Heinz, Huse, Huth, Jepsen, Jørgensen, Mooij, Müller, Pe'er, Piou, Railsback, Robbins, Robbins, Rossmanith, Rüger, Strand, Souissi, Stillman, Vabø, Visser and Deangelis2006) to organize a narrative description. I identified outcome variables (frequency of use and fluency in the two languages), frequent actors, and defined rules for initiating/continuing conversation, and rules for agents to move to new locations. Titone and Tiv's (Reference Titone and Tiv2022) framework suggests that an additional organizing feature would be to use the nested spheres of influence, and to draw on complex variables such as language entropy to organize both a description and implementation.
Summary
Titone and Tiv (Reference Titone and Tiv2022) have been forward thinking in describing diverse new research and new questions that exemplify systems thinking about multilingual experiences. I make the case here that longstanding questions will also benefit from being incorporated into a multicausal, dynamic framework.
Conflict of interest
The author declares no conflict of interest.