According to David Marr’s hierarchy of levels of understanding of a complex processing system, linguistic theory approaches language issues mostly at Marrian level 1 (“what do speakers do when they use language?”), while cognitive psychology and computational linguistics are chiefly concerned with level 2 issues (“how do they use it?”), and neurosciences with level 3 questions (“how is such behavior implemented in the brain?”).
Although David Marr originally introduced his hierarchy to emphasize that explanations at different levels can be investigated independently, over the last 25 years there has been growing interest in the potential for between-level interaction to investigate the methodological conditions for their interdisciplinary unification. With the advances in computer sciences and neuro-imaging technology, we contend that psycho-computational models can provide the middle-ground framework for inter-level cross talking.
To illustrate, a recent reconceptualization of morphological generalization as the “Cell Filling Problem” hinged on modelling the implicative structure of morphological paradigms through conditional entropy, an information-theoretic measure of inferential complexity that proves to correlate significantly with speakers’ behavior, opening new avenues for typological inquiry. Likewise, analogy-based synchronic descriptions of language systems and historical accounts of language change got a new lease of life when analogical relations and their cognitive implications were successfully operationalized in the machine learning literature.
The categorical subdivision between regularly and irregularly inflected forms advocated by dual models of word processing has been challenged by integrative, connectionist models of short-term and long-term memories, implemented as two different temporal dynamics of the same underlying process. On a complementary front, principles of naïve discrimination learning have successfully simulated human behavioral evidence based on explicitly non-morphemic word representations. At the same time, however, statistical analyses and computer simulations of speakers’ reaction times in visual word recognition challenged evidence of amorphous, holistic representations in the speakers’ mental lexicon.
In this Research Topic, we aim to take stock of the implications of current psycho-computational models of word processing for morphological theory, and make a fair assessment of current models of word knowledge and processing, and their potential for hybridization.
We welcome articles addressing one of these questions:
What are the optimal representation units of human morphological competence and how are they acquired? What role do they play in the way speakers process and store words? Do speakers combine these units in a linear way, as in chaining Markov models, or rather structure them hierarchically, as suggested by the literature on sentence processing? Do they store them in their long-term lexical repository economically, or rather multiply them redundantly, as a function of their context and use?
In addition, are these units represented as independent items, or are they mutually related as nodes in a network of paradigmatic relations? What is the contribution of lexical semantics to this picture, and what type of influence is exercised on lexical units by the communicative context where they are used referentially? What is the status of the processes combining these units into larger units? Are they implemented by a single mechanism? Or should we rather hypothesize that more than one mechanism is in place?
According to David Marr’s hierarchy of levels of understanding of a complex processing system, linguistic theory approaches language issues mostly at Marrian level 1 (“what do speakers do when they use language?”), while cognitive psychology and computational linguistics are chiefly concerned with level 2 issues (“how do they use it?”), and neurosciences with level 3 questions (“how is such behavior implemented in the brain?”).
Although David Marr originally introduced his hierarchy to emphasize that explanations at different levels can be investigated independently, over the last 25 years there has been growing interest in the potential for between-level interaction to investigate the methodological conditions for their interdisciplinary unification. With the advances in computer sciences and neuro-imaging technology, we contend that psycho-computational models can provide the middle-ground framework for inter-level cross talking.
To illustrate, a recent reconceptualization of morphological generalization as the “Cell Filling Problem” hinged on modelling the implicative structure of morphological paradigms through conditional entropy, an information-theoretic measure of inferential complexity that proves to correlate significantly with speakers’ behavior, opening new avenues for typological inquiry. Likewise, analogy-based synchronic descriptions of language systems and historical accounts of language change got a new lease of life when analogical relations and their cognitive implications were successfully operationalized in the machine learning literature.
The categorical subdivision between regularly and irregularly inflected forms advocated by dual models of word processing has been challenged by integrative, connectionist models of short-term and long-term memories, implemented as two different temporal dynamics of the same underlying process. On a complementary front, principles of naïve discrimination learning have successfully simulated human behavioral evidence based on explicitly non-morphemic word representations. At the same time, however, statistical analyses and computer simulations of speakers’ reaction times in visual word recognition challenged evidence of amorphous, holistic representations in the speakers’ mental lexicon.
In this Research Topic, we aim to take stock of the implications of current psycho-computational models of word processing for morphological theory, and make a fair assessment of current models of word knowledge and processing, and their potential for hybridization.
We welcome articles addressing one of these questions:
What are the optimal representation units of human morphological competence and how are they acquired? What role do they play in the way speakers process and store words? Do speakers combine these units in a linear way, as in chaining Markov models, or rather structure them hierarchically, as suggested by the literature on sentence processing? Do they store them in their long-term lexical repository economically, or rather multiply them redundantly, as a function of their context and use?
In addition, are these units represented as independent items, or are they mutually related as nodes in a network of paradigmatic relations? What is the contribution of lexical semantics to this picture, and what type of influence is exercised on lexical units by the communicative context where they are used referentially? What is the status of the processes combining these units into larger units? Are they implemented by a single mechanism? Or should we rather hypothesize that more than one mechanism is in place?
