Christiansen & Chater (C&C) offer an intriguing proposal concerning the nature of language, intended to explain fundamental aspects of language comprehension, production, learning, and evolution. We agree with the basic framework, and indeed we have offered our own theoretical approach, Good-Enough (GE) Language Processing, to capture many of the phenomena discussed in the target article, particularly those relating to both online and offline comprehension. In this commentary, we hope to expand the discussion by pointing to some of these connections and highlighting additional phenomena that C&C did not discuss but that reinforce some of their points. In addition, however, we believe the GE model is better able to explain important aspects of language comprehension that C&C consider, as well as several they leave out. Of course, no single article could be comprehensive when it comes to a field as broad and active as this one, but we believe a complete theory of language must ultimately have something to say about these important phenomena, and particularly the content of people's interpretations.
We begin, then, with a brief review of the GE approach (Ferreira et al. Reference Ferreira, Bailey and Ferraro2002). The fundamental assumption is that interpretations are often shallow and sometimes inaccurate. This idea that interpretations are shallow and underspecified is similar to C&C's suggestion that the comprehension system creates chunks that might not be combined into a single, global representation. In their model, this tendency arises from memory constraints that lead the system to build chunks at increasingly abstract levels of representation. As evidence for this assumption regarding underspecified representations, C&C might have discussed our work demonstrating that ambiguous relative clauses are often not definitively attached into the matrix structure if a failure to attach has no interpretive consequences (Swets et al. 2008; cf. Payne et al. Reference Payne, Grison, Gao, Christianson, Morrow and Stine-Morrow2014). Very much in line with C&C, Swets et al. observed that people who are asked detailed comprehension questions probing their interpretation of the ambiguous relative clause make definitive attachments, but those asked only shallow questions about superficial features of the sentence seem to leave the relative clause unattached – that is, they underspecify. This finding fits neatly with C&C's discussion of “right context effects,” where here “right context” can be broadly construed to mean the follow-on comprehension question that influences the interpretation constructed online. An important difference, however, emerges as well, and here we believe the GE framework has some advantages over Now-or-Never as a broad model of comprehension: Our framework predicts that the language user's task will have a strong effect on the composition of “chunks” and the interpretation created from them (cf. Christianson & Luke Reference Christianson and Luke2011; Lim & Christianson Reference Lim and Christianson2015). We have reported these results in production as well, demonstrating that the extent to which speaking is incremental depends on the processing demands of the speaking task (Ferreira & Swets Reference Ferreira and Swets2002). Given the importance of task effects in a range of cognitive domains, any complete model of language processing must include mechanisms for explaining how they arise.
Moreover, the idea that language processing proceeds chunk-by-chunk is not novel. C&C consider some antecedents of their proposal, but several are overlooked. For example, they argue that memory places major constraints on language processing, essentially obligating the system to chunk and interpret as rapidly as possible (what they term “eager processing”). This was a key motivation for Lyn Frazier's original garden-path model (Frazier & Rayner Reference Frazier and Rayner1982) and the parsing strategies known as minimal attachment and late closure: The parser's goal is to build an interpretation quickly and pursue the one that emerges first rather than waiting for and considering multiple alternatives. This, too, is part of C&C's proposal – that the parser cannot construct multiple representations at the same level in parallel – but the connections to the early garden-path model are not mentioned, and the incompatibility of this idea with parallel models of parsing is also not given adequate attention. Another example is work by Tyler and Warren (Reference Tyler and Warren1987), who showed that listeners form unlinked local phrasal chunks during spoken language processing and who conclude that they could find no evidence for the formation of a global sentence representation. Thus, several of these ideas have been part of the literature for many years, and evidence for them can be found in research motivated from a broad range of theoretical perspectives.
Perhaps the most critical aspect of comprehension that C&C's approach does not capture is meaning and interpretation: C&C describe an architecture that can account for some aspects of processing, but their model seems silent on the matter of the content of people's interpretations. This is a serious shortcoming given the considerable evidence for systematic misinterpretation (e.g., Christianson et al. Reference Christianson, Hollingworth, Halliwell and Ferreira2001; Reference Christianson, Williams, Zacks and Ferreira2006; Patson et al. Reference Patson, Darowski, Moon and Ferreira2009; van Gompel et al. Reference van Gompel, Pickering, Pearson and Jacob2006). In our work, we demonstrated that people who read sentences such as While Mary bathed the baby played in the crib often derive the interpretation that Mary bathed the baby, and they also misinterpret simple passives such as The dog was bitten by the man (Ferreira Reference Ferreira2003). These are not small tendencies; the effects are large, and they have been replicated in numerous studies across many different labs. For C&C, these omissions are a lost opportunity because these results are consistent with their proposed architecture. For example, misinterpretations of garden-path sentences arise in part because the parser processes sentences in thematic chunks and fails to reconcile the various meanings constructed online. Recently, we demonstrated that the misinterpretations are attributable to a failure to “clean up” the interpretive consequences of creating these chunks (Slattery et al. Reference Slattery, Sturt, Christianson, Yoshida and Ferreira2013), a finding compatible with C&C's idea that chunks are quickly recoded into more abstract levels of representation and that it is difficult to re-access the less abstract representations.
C&C's framework is exciting, and we believe it will inspire significant research. Their creative synthesis is a major achievement, and we hope we have contributed constructively to the project by pointing to areas of connection and convergence as well as by highlighting important gaps.
Christiansen & Chater (C&C) offer an intriguing proposal concerning the nature of language, intended to explain fundamental aspects of language comprehension, production, learning, and evolution. We agree with the basic framework, and indeed we have offered our own theoretical approach, Good-Enough (GE) Language Processing, to capture many of the phenomena discussed in the target article, particularly those relating to both online and offline comprehension. In this commentary, we hope to expand the discussion by pointing to some of these connections and highlighting additional phenomena that C&C did not discuss but that reinforce some of their points. In addition, however, we believe the GE model is better able to explain important aspects of language comprehension that C&C consider, as well as several they leave out. Of course, no single article could be comprehensive when it comes to a field as broad and active as this one, but we believe a complete theory of language must ultimately have something to say about these important phenomena, and particularly the content of people's interpretations.
We begin, then, with a brief review of the GE approach (Ferreira et al. Reference Ferreira, Bailey and Ferraro2002). The fundamental assumption is that interpretations are often shallow and sometimes inaccurate. This idea that interpretations are shallow and underspecified is similar to C&C's suggestion that the comprehension system creates chunks that might not be combined into a single, global representation. In their model, this tendency arises from memory constraints that lead the system to build chunks at increasingly abstract levels of representation. As evidence for this assumption regarding underspecified representations, C&C might have discussed our work demonstrating that ambiguous relative clauses are often not definitively attached into the matrix structure if a failure to attach has no interpretive consequences (Swets et al. 2008; cf. Payne et al. Reference Payne, Grison, Gao, Christianson, Morrow and Stine-Morrow2014). Very much in line with C&C, Swets et al. observed that people who are asked detailed comprehension questions probing their interpretation of the ambiguous relative clause make definitive attachments, but those asked only shallow questions about superficial features of the sentence seem to leave the relative clause unattached – that is, they underspecify. This finding fits neatly with C&C's discussion of “right context effects,” where here “right context” can be broadly construed to mean the follow-on comprehension question that influences the interpretation constructed online. An important difference, however, emerges as well, and here we believe the GE framework has some advantages over Now-or-Never as a broad model of comprehension: Our framework predicts that the language user's task will have a strong effect on the composition of “chunks” and the interpretation created from them (cf. Christianson & Luke Reference Christianson and Luke2011; Lim & Christianson Reference Lim and Christianson2015). We have reported these results in production as well, demonstrating that the extent to which speaking is incremental depends on the processing demands of the speaking task (Ferreira & Swets Reference Ferreira and Swets2002). Given the importance of task effects in a range of cognitive domains, any complete model of language processing must include mechanisms for explaining how they arise.
Moreover, the idea that language processing proceeds chunk-by-chunk is not novel. C&C consider some antecedents of their proposal, but several are overlooked. For example, they argue that memory places major constraints on language processing, essentially obligating the system to chunk and interpret as rapidly as possible (what they term “eager processing”). This was a key motivation for Lyn Frazier's original garden-path model (Frazier & Rayner Reference Frazier and Rayner1982) and the parsing strategies known as minimal attachment and late closure: The parser's goal is to build an interpretation quickly and pursue the one that emerges first rather than waiting for and considering multiple alternatives. This, too, is part of C&C's proposal – that the parser cannot construct multiple representations at the same level in parallel – but the connections to the early garden-path model are not mentioned, and the incompatibility of this idea with parallel models of parsing is also not given adequate attention. Another example is work by Tyler and Warren (Reference Tyler and Warren1987), who showed that listeners form unlinked local phrasal chunks during spoken language processing and who conclude that they could find no evidence for the formation of a global sentence representation. Thus, several of these ideas have been part of the literature for many years, and evidence for them can be found in research motivated from a broad range of theoretical perspectives.
Perhaps the most critical aspect of comprehension that C&C's approach does not capture is meaning and interpretation: C&C describe an architecture that can account for some aspects of processing, but their model seems silent on the matter of the content of people's interpretations. This is a serious shortcoming given the considerable evidence for systematic misinterpretation (e.g., Christianson et al. Reference Christianson, Hollingworth, Halliwell and Ferreira2001; Reference Christianson, Williams, Zacks and Ferreira2006; Patson et al. Reference Patson, Darowski, Moon and Ferreira2009; van Gompel et al. Reference van Gompel, Pickering, Pearson and Jacob2006). In our work, we demonstrated that people who read sentences such as While Mary bathed the baby played in the crib often derive the interpretation that Mary bathed the baby, and they also misinterpret simple passives such as The dog was bitten by the man (Ferreira Reference Ferreira2003). These are not small tendencies; the effects are large, and they have been replicated in numerous studies across many different labs. For C&C, these omissions are a lost opportunity because these results are consistent with their proposed architecture. For example, misinterpretations of garden-path sentences arise in part because the parser processes sentences in thematic chunks and fails to reconcile the various meanings constructed online. Recently, we demonstrated that the misinterpretations are attributable to a failure to “clean up” the interpretive consequences of creating these chunks (Slattery et al. Reference Slattery, Sturt, Christianson, Yoshida and Ferreira2013), a finding compatible with C&C's idea that chunks are quickly recoded into more abstract levels of representation and that it is difficult to re-access the less abstract representations.
C&C's framework is exciting, and we believe it will inspire significant research. Their creative synthesis is a major achievement, and we hope we have contributed constructively to the project by pointing to areas of connection and convergence as well as by highlighting important gaps.