A complementary systems account of word learning: neural and behavioural evidence

Abstract

In this paper we present a novel theory of the cognitive and neural processes by which adults learn new spoken words. This proposal builds on neurocomputational accounts of lexical processing and spoken word recognition and complementary learning systems (CLS) models of memory. We review evidence from behavioural studies of word learning that, consistent with the CLS account, show two stages of lexical acquisition: rapid initial familiarization followed by slow lexical consolidation. These stages map broadly onto two systems involved in different aspects of word learning: (i) rapid, initial acquisition supported by medial temporal and hippocampal learning, (ii) slower neocortical learning achieved by offline consolidation of previously acquired information. We review behavioural and neuroscientific evidence consistent with this account, including a meta-analysis of PET and functional Magnetic Resonance Imaging (fMRI) studies that contrast responses to spoken words and pseudowords. From this meta-analysis we derive predictions for the location and direction of cortical response changes following familiarization with pseudowords. This allows us to assess evidence for learning-induced changes that convert pseudoword responses into real word responses. Results provide unique support for the CLS account since hippocampal responses change during initial learning, whereas cortical responses to pseudowords only become word-like if overnight consolidation follows initial learning.

Figure 1.

Neural and functional organization of systems involved in representing and learning spoken words. (a) Left temporal lobe regions involved in perceiving and comprehending spoken words (based on Hickok & Poeppel 2004; Davis & Johnsrude 2007) and their interactions with medial temporal systems for word learning. (b) Functional organization of the Distributed Cohort Model (Gaskell & Marslen-Wilson , ; depicted within the grey box) with additional connections to hippocampal/episodic memory system for learning new words. In both diagrams, rapid cortico-cortico connections are shown with solid lines, and slower, cortico-hippocampal connections are shown with broken lines. Dotted lines with open arrow-heads show recurrent connections involved in maintaining acoustic-phonetic representations in echoic memory.

Figure 2.

Impact of initial learning and sleep-associated consolidation on lexical representations and word recognition. (a) Speech waveforms for tokens of an existing word (cathedral) and a new word (cathedruke) with a marker showing the approximate time point at which the acoustic-phonetic input for cathedral diverges from all other known words (uniqueness point, cf. Marslen-Wilson 1984). The uniqueness point for cathedral is markedly later (orange line versus blue line) if the new word cathedruke must also be ruled out. (b) Lexical organization of these words after learning and before or after sleep-associated consolidation. Before sleep (blue box) the strongest lexical competitor for cathedral is cathartic, hence the uniqueness point is reached once this word can be ruled out (early uniqueness point shown in a). After sleep (orange box), the addition of a new lexical competitor is such that additional speech input is required to rule out cathedruke (late uniqueness point shown in a). (c) Pause detection response times showing the impact of learning and consolidation on lexical competition (data replotted from Dumay & Gaskell 2007). Two groups of participants were trained on novel words (e.g. cathedruke) at either 8.00 h or 20.00 h and tested on matched items with and without new lexical competitors 0, 12 or 24 h after training. Responses to existing words were significantly delayed by competition from newly learned words only for those conditions in which sleep intervened between training and testing (orange bars).

Figure 3.

Activation Likelihood Estimation (ALE) map derived from 97 peak voxels from 11 functional imaging studies comparing neural responses to spoken words and pseudowords. ALE maps are thresholded at p < 0.05 FDR corrected, and only clusters larger than 100 mm³ are shown. Additional activation for pseudowords compared with words (red) and words compared with pseudowords (blue) is shown (a) rendered onto the left hemisphere, (b) displayed on an axial and (c) multiple sagittal slices of the MNI canonical brain (z and x coordinates as shown). Note inferior temporal and fusiform activation for word >pseudoword (circled in orange) is largely hidden in the rendering but apparent on the axial slice and sagittal slices x = −48, −40 and −32. (d,e) Response profiles showing predicted changes in neural responses owing to familarization with pseudowords: (d) Within regions that initially show an additional response to pseudowords (red in figure 3a–c, e.g. STG, posterior inferior frontal gyrus). For these regions we predict a diminished response to pseudowords after training. (d) Predicted response within regions that show an additional response to real words (anterior MTG, anterior fusiform, supramarginal gyrus, blue in figure 3a–c), we predict an increased response to pseudowords following training. (f) fMRI responses in a region of the STG overlapping with areas shown in red in figure 3a–c (replotted from Davis et al. 2009). An equivalent, additional response to pseudowords was seen for items that were untrained at the time of scanning or trained but not consolidated (i.e. learned on the same day as scanning). However, there was a significant lexicality by consolidation interaction with a reduced response to pseudowords that were trained and consolidated (i.e. learned on the day before scanning).