Christiansen & Chater (C&C) claim that memory limitations force the cognitive system to process the transient linguistic signal by compressing it. They suggest that this processing pressure influences the ultimate structure of language over the course of language evolution. Taken at face value, this proposal would lead to a degenerate linguistic structure, however. If memory constraints were the only pressure on language, languages would evolve to compress meaning into the simplest possible form – a single word (Horn Reference Horn and Schiffrin1984). But, as the authors point out, natural languages are not of this sort; they are richly structured into lexical and phrasal units of varying length. To account for this variability, we highlight the need to consider the communicative function of language. Communication serves as an important counter-pressure against compression in language processing, not just as a caveat.
Interlocutors use language with the goal of communicating information, but they also aim to minimize energetic cost (Zipf Reference Zipf1949). For the speaker, this goal implies minimizing production cost, and for the listener it implies minimizing comprehension cost. Importantly, these processing constraints have opposing cost functions (Horn Reference Horn and Schiffrin1984; Zipf Reference Zipf1949). For a producer, processing is minimized when a form is easy to say, and thus highly compressible. For the comprehender, however, processing is minimized when a form is minimally ambiguous and thus verbose. Compressing information is a useful strategy for a speaker who faces memory constraints, but it is useful only to the extent that the listener can still recover the intended meaning. This view of language use as rational action – minimizing costs while maximizing information transfer – is supported by a rich body of theoretical and empirical work (Clark Reference Clark1996; Frank & Goodman Reference Frank and Goodman2012; Goodman & Stuhlmüller Reference Goodman and Stuhlmüller2013; Grice Reference Grice, Cole and Morgan1975).
Although C&C argue that compression is the key factor in the emergence of structure, evidence at both the acquisition and evolution timescales suggests language is the product of the interaction between both compression and informativity. At the timescale of acquisition, experimental work suggests the resolution of reference in word learning is the product of communicative inferences (e.g., Baldwin Reference Baldwin1991; Reference Baldwin1993; Frank et al. Reference Frank, Goodman and Tenenbaum2009; Frank & Goodman Reference Frank and Goodman2014). And at the timescale of language evolution, a growing body of work suggests that the forms of words are also equilibria between these two pressures (Lewis & Frank Reference Lewis, Sugarman and Frank2014; Mahowald et al. Reference Mahowald, Fedorenko, Piantadosi and Gibson2012; Piantadosi et al. Reference Piantadosi, Tily and Gibson2011; Zipf Reference Zipf1936). For example, Piantadosi et al. (Reference Piantadosi, Tily and Gibson2011) found that words that are less predictable in their linguistic context are longer, suggesting that speakers may lengthen words that are surprising in order to increase time for the listener to process.
In addition to linguistic form, these pressures influence the mapping between form and meaning. An equilibrium in the structure of form-meaning mappings is one in which the listener is able to recover the intended meaning, but the speaker does not exert additional effort over-describing. A range of semantic domains reflect this equilibrium (Baddeley & Attewell Reference Baddeley and Attewell2009; Kemp & Regier Reference Kemp and Regier2012; Regier et al. Reference Regier, Kay and Khetarpal2007), and ambiguity, more generally, has been argued to reflect this communicative tradeoff (Piantadosi et al. 2012). Ambiguity is an equilibrium in cases where the listener can recover the intended meaning from the communicative context. One example is the word “some,” which has a literal meaning of “at least one and possibly all” but can be strengthened pragmatically to mean “at least one but not all” (Horn Reference Horn1972). Because its meaning is determined through communicative context, its literal semantics can overlap those of its competitor, “all.”
The key challenge associated with this broader proposal – that processing pressures influence linguistic structure – is providing direct evidence for a causal link between these two timescales. This problem is difficult to study in the laboratory because the proposed mechanism takes place over a long timescale and over multiple individual speakers. Furthermore, the presence of a causal link does not entail that phenomena in processing are directly reflected in linguistic structure – rather, entirely new properties may emerge at higher levels of abstraction from the interactions of more fundamental phenomena (Anderson Reference Anderson1972). It may, therefore, not be possible to directly extrapolate from brief communicative interactions observed in the laboratory to properties of linguistic structure.
Several recent pieces of experimental data begin to address this challenge, however. In one study, Fedzechkina et al. (Reference Fedzechkina, Jaeger and Newport2012) asked speakers to learn an artificial language that arbitrarily distinguished nouns through case-marking. Over learning sessions, speakers developed a system for marking in contexts where meanings were least predictable – a pattern reflected in the case-marking systems of natural language. Other work has used a similar paradigm to reveal the emergence of typologically prevalent patterns in the domains of word order (Culbertson et al. Reference Culbertson, Smolensky and Legendre2012; Culbertson & Newport Reference Culbertson and Newport2015) and phonology (Wilson 2008).
A particularly promising approach for exploring this causal link is through transmission chains (Kirby et al. Reference Kirby, Cornish and Smith2008; Reali & Griffiths Reference Reali and Griffiths2009). In a transmission chain, a participant learns and recalls a language, and then the recalled language becomes the learning input for a new learner. By iterating over learners, we can observe how languages change across transmission of learners over the course of language evolution. Kirby et al. (Reference Kirby, Tamariz, Cornish and Smith2015) have compared the emergence of linguistic structure in a regime that iterates over different partners of learners versus a regime where the same two partners repeatedly interact with each other. They find that linguistic structure emerges only by iterating over different partners, demonstrating the unique contribution of cross-generational learning to the emergence of structure. Others have begun to use this paradigm to link the interaction of processing pressures to the emergence of communicative regularities in semantic structure (Carstensen et al. Reference Carstensen, Xu, Smith, Regier, Noelle, Dale, Warlaumont, Yoshimi, Matlock, Jennings and Maglio2015; Lewis & Frank Reference Lewis, Frank, Noelle, Dale, Warlaumont, Yoshimi, Matlock, Jennings and Magli2015).
In sum, the consequences of memory constraints are likely a critical factor in shaping language structure. But an additional important constraint is the pressure to communicate informatively, and this constraint should not be overlooked in accounting for linguistic structure.
Christiansen & Chater (C&C) claim that memory limitations force the cognitive system to process the transient linguistic signal by compressing it. They suggest that this processing pressure influences the ultimate structure of language over the course of language evolution. Taken at face value, this proposal would lead to a degenerate linguistic structure, however. If memory constraints were the only pressure on language, languages would evolve to compress meaning into the simplest possible form – a single word (Horn Reference Horn and Schiffrin1984). But, as the authors point out, natural languages are not of this sort; they are richly structured into lexical and phrasal units of varying length. To account for this variability, we highlight the need to consider the communicative function of language. Communication serves as an important counter-pressure against compression in language processing, not just as a caveat.
Interlocutors use language with the goal of communicating information, but they also aim to minimize energetic cost (Zipf Reference Zipf1949). For the speaker, this goal implies minimizing production cost, and for the listener it implies minimizing comprehension cost. Importantly, these processing constraints have opposing cost functions (Horn Reference Horn and Schiffrin1984; Zipf Reference Zipf1949). For a producer, processing is minimized when a form is easy to say, and thus highly compressible. For the comprehender, however, processing is minimized when a form is minimally ambiguous and thus verbose. Compressing information is a useful strategy for a speaker who faces memory constraints, but it is useful only to the extent that the listener can still recover the intended meaning. This view of language use as rational action – minimizing costs while maximizing information transfer – is supported by a rich body of theoretical and empirical work (Clark Reference Clark1996; Frank & Goodman Reference Frank and Goodman2012; Goodman & Stuhlmüller Reference Goodman and Stuhlmüller2013; Grice Reference Grice, Cole and Morgan1975).
Although C&C argue that compression is the key factor in the emergence of structure, evidence at both the acquisition and evolution timescales suggests language is the product of the interaction between both compression and informativity. At the timescale of acquisition, experimental work suggests the resolution of reference in word learning is the product of communicative inferences (e.g., Baldwin Reference Baldwin1991; Reference Baldwin1993; Frank et al. Reference Frank, Goodman and Tenenbaum2009; Frank & Goodman Reference Frank and Goodman2014). And at the timescale of language evolution, a growing body of work suggests that the forms of words are also equilibria between these two pressures (Lewis & Frank Reference Lewis, Sugarman and Frank2014; Mahowald et al. Reference Mahowald, Fedorenko, Piantadosi and Gibson2012; Piantadosi et al. Reference Piantadosi, Tily and Gibson2011; Zipf Reference Zipf1936). For example, Piantadosi et al. (Reference Piantadosi, Tily and Gibson2011) found that words that are less predictable in their linguistic context are longer, suggesting that speakers may lengthen words that are surprising in order to increase time for the listener to process.
In addition to linguistic form, these pressures influence the mapping between form and meaning. An equilibrium in the structure of form-meaning mappings is one in which the listener is able to recover the intended meaning, but the speaker does not exert additional effort over-describing. A range of semantic domains reflect this equilibrium (Baddeley & Attewell Reference Baddeley and Attewell2009; Kemp & Regier Reference Kemp and Regier2012; Regier et al. Reference Regier, Kay and Khetarpal2007), and ambiguity, more generally, has been argued to reflect this communicative tradeoff (Piantadosi et al. 2012). Ambiguity is an equilibrium in cases where the listener can recover the intended meaning from the communicative context. One example is the word “some,” which has a literal meaning of “at least one and possibly all” but can be strengthened pragmatically to mean “at least one but not all” (Horn Reference Horn1972). Because its meaning is determined through communicative context, its literal semantics can overlap those of its competitor, “all.”
The key challenge associated with this broader proposal – that processing pressures influence linguistic structure – is providing direct evidence for a causal link between these two timescales. This problem is difficult to study in the laboratory because the proposed mechanism takes place over a long timescale and over multiple individual speakers. Furthermore, the presence of a causal link does not entail that phenomena in processing are directly reflected in linguistic structure – rather, entirely new properties may emerge at higher levels of abstraction from the interactions of more fundamental phenomena (Anderson Reference Anderson1972). It may, therefore, not be possible to directly extrapolate from brief communicative interactions observed in the laboratory to properties of linguistic structure.
Several recent pieces of experimental data begin to address this challenge, however. In one study, Fedzechkina et al. (Reference Fedzechkina, Jaeger and Newport2012) asked speakers to learn an artificial language that arbitrarily distinguished nouns through case-marking. Over learning sessions, speakers developed a system for marking in contexts where meanings were least predictable – a pattern reflected in the case-marking systems of natural language. Other work has used a similar paradigm to reveal the emergence of typologically prevalent patterns in the domains of word order (Culbertson et al. Reference Culbertson, Smolensky and Legendre2012; Culbertson & Newport Reference Culbertson and Newport2015) and phonology (Wilson 2008).
A particularly promising approach for exploring this causal link is through transmission chains (Kirby et al. Reference Kirby, Cornish and Smith2008; Reali & Griffiths Reference Reali and Griffiths2009). In a transmission chain, a participant learns and recalls a language, and then the recalled language becomes the learning input for a new learner. By iterating over learners, we can observe how languages change across transmission of learners over the course of language evolution. Kirby et al. (Reference Kirby, Tamariz, Cornish and Smith2015) have compared the emergence of linguistic structure in a regime that iterates over different partners of learners versus a regime where the same two partners repeatedly interact with each other. They find that linguistic structure emerges only by iterating over different partners, demonstrating the unique contribution of cross-generational learning to the emergence of structure. Others have begun to use this paradigm to link the interaction of processing pressures to the emergence of communicative regularities in semantic structure (Carstensen et al. Reference Carstensen, Xu, Smith, Regier, Noelle, Dale, Warlaumont, Yoshimi, Matlock, Jennings and Maglio2015; Lewis & Frank Reference Lewis, Frank, Noelle, Dale, Warlaumont, Yoshimi, Matlock, Jennings and Magli2015).
In sum, the consequences of memory constraints are likely a critical factor in shaping language structure. But an additional important constraint is the pressure to communicate informatively, and this constraint should not be overlooked in accounting for linguistic structure.