Developing a meaningful account of the acquisition of the English past tense

ESRC Grant reference ES/I021612/1

Duration: June 2011 – December 2014

Principal Investigator: Dr Anna Theakston, University of Manchester, LuCiD Child Study Centre 
Co-Investigator: Dr Anna Woollams, University of Manchester
Research Associates: Dr Grzegorz Krajewski & Dr Joanna Moy
Research Assistant: Sarah Keeble

OVERVIEW

The aim of this study was to compare children’s performance in producing English past tense verb forms across two different tasks, the first a stem-based elicitation task with a strong emphasis on phonological representations, and the second a novel and more ecologically valid video inflection task with an inherently higher load on semantic information. The purpose was to enhance our understanding of the nature of children’s linguistic representations and how they might change over development, and to cast light on theoretical models of morphological acquisition and processing.

In the stem-based task, children were presented with the progressive form of a verb (e.g. jumping) and asked to produce the past tense form (i.e. jumped). In the second video-based task, children watched an animation and were asked to describe what had happened, thus they were not given the progressive form of the target verb.  Two tasks were used because we wanted to investigate whether error rates and types might vary according to task (as has been previously observed in some adult populations). We predicted that children would be more likely to produce certain kinds of errors (e.g. overregularisation errors, go-ed) when presented with the progressive verb form (e.g. going) because this could activate the stem form of the verb (e.g. go) which is then used erroneously to create a past tense form.

We created a standard picture task to elicit around 300 English verbs, and an animation task for around 120 English verbs. Approx. 900 children aged 2yrs 6months - 5yrs 5months participated in the study across the two tasks, with each child completing a subset of the test items (30 for children aged 3½ -5½, 20 for children aged  2½ - 3½.  Our results demonstrate that the rate at which children produce correct and error responses varies as a function of their age, the verb, and the task, reflecting different routes to access or generate past tense forms as a function of context.  These data revealing interesting differences between verbs will provide a benchmark against which to compare the performance of children in other studies, and those with language disorders.

In addition, we developed a wide range of predictor variables for past tense verb performance encoding various kinds of frequency, phonological, and semantic information for individual verbs, including a number of novel measures. Our aim was to discover which factors are most important in determining children’s correct use and errors across verbs, to build a more complete model of past tense acquisition than has been possible previously.  Our results reveal complex interactions between the variables that change over the course of development, providing insights into how the system builds up.

Finally, we developed a connectionist model to simulate the children’s data. The model consisted of three input/output layers corresponding to input phonology, output phonology, and input/output semantics (what the model is “hearing”, “saying”, and the verb meanings presented to the model respectively) and one large bank of hidden units connecting them. The two tasks were simulated as either “inflection from form” (standard picture inflection task) or “inflection from meaning” (video-inflection task) whereby the model was presented with the present tense/semantic representation of the verb respectively. The model accurately reproduced the task and developmental differences seen in the children, demonstrating that a unified frequency-sensitive system of phonological and semantic networks can explain the variable performance we see in children at different developmental stages.

Together, these new methods, resources and theoretical insights provide much needed evidence of how children’s knowledge of morphological systems develops, and important tools for the further investigation of critical questions in language acquisition research.

 

PRACTITIONER RESOURCES

Problems in producing past tense forms accurately are widely recognised in children with language impairments.  Our data and materials may therefore be helpful to Speech and Language Therapists wanting to gain an insight into the variability of performance across verbs in the typically developing population, or looking for new resources to elicit past tense forms from children.

Practitioners may find some of the following resources useful to:

  • Compare a child’s performance on a range of verbs against data from typically developing children to identify areas of difficulty
  • Decide on a sample of verbs to elicit from children to test their knowledge of past tense inflections, or to use as training in therapeutic interventions
  • Create a set of test materials (pictures & animations).

Selecting test items

You can select test items by looking at the frequency with which different verbs occur in the past tense in the language addressed to children.  Often children make fewer mistakes with high frequency words, so you might want to select verbs from across the frequency range. Here you can find a frequency ordered list of 300 different verbs based on how often they occurred in their past tense form in a large sample of speech from 12 mothers addressed to their children who were aged between two and three years old. Full transcripts of these interactions are available via the CHILDES website and can be downloaded by clicking on Manchester.zip.

Another way to select test items is based on how accurately they are produced by typically developing children at different ages. Some verbs tend to be produced correctly, others attract higher error rates. You might want to select a range of verbs based on typical accuracy of use. Here you can find typically developing children’s accuracy rates for these same 300 verbs (data taken from our stem-elicitation task described above).  Children’s responses are grouped by age for 2½ - 3½ yrs; 3½ - 4½ yrs; 4½ - 5½ yrs.

Elicitation materials

Here you can download past tense elicitation scripts and associated elicitation pictures.

Materials for the video elicitation task and comparable accuracy rates will be made available in due course. Please contact Anna Theakston if you would like further information.

OUTPUTS

  • Moy, J., Theakston, A., Krajewski, G. & Woollams, A. (2014) The impact of task context on children’s acquisition of inflection. 20th Architectures and Mechanisms for Language Processing (AMLaP), Edinburgh, UK. 
  • Woollams, A., Keeble, S.,  Krajewski, G., Moy, J. & Theakston, A. (2014). Inflection from form versus meaning: Developmental task differencesCognitive Science Society Conference, Canada. 
  • Theakston, A., Krajewski, G., Keeble, S., & Woollams, A. (2013) Frequency, phonological and semantic factors in the development of the English past tense. Cognitive Science Society Conference, Berlin, Germany. 
  • Woollams, A., Moy, J., Keeble, S.,  Krajewski, G., Moy, J. & Theakston, A. (2013) Inflection from form versus meaning: Developmental and computational evidence Cognitive Science Society Conference Berlin, Germany, 
  • Theakston, A., Krajewski, G., Keeble, S. & Woollams, A. (2014).  The role of lexical aspect in past tense production in children. Paper presented at the 5th Uk Cognitive Linguistics Society conference.
  • Theakston, A., Krajewski, G., Keeble, S. & Woollams, A. (2014).  Verb semantics and the acquisition of the English past tense. Poster presented at the 13th International Association for the Study of Child Language, Amsterdam.
  • Theakston, A., Krajewski, G., Keeble, S. & Woollams, A. (2014).  Morphological errors in the production of the English past tense: The contribution of frequency, phonological, semantic and contextual factors. Paper given at the 13th International Association for the Study of Child Language, Amsterdam, 2014.
  • Theakston, A., Woollams, A., Krajewski, G., Keeble, S. (2013).  Task effects in the production of the English past tense: Stem retrieval versus direct access from meaning. Paper presented at the Child Language Seminar, Manchester.
  • Theakston, A., Krajewski, G., Keeble, S. & Woollams, A. (2013).  Prior discourse effects in the production of the English past tense - the role of priming in children's errors. Poster presented at Boston University Child Language Development conference.
  • Krajewski, G. Keeble, S., Woollams, A. & Theakston, A. (2012) The acquisition of the English past tense. Paper presented at 14th Annual Conference of the Chester (North-West) Language Development Group.
  • Woollams, A. (2012). Explorations in Past-Tense Inflection:  From decline to development. Invited Colloquium, Department of Psychology, Royal Holloway.
  • Woollams, A. (2011). Exploiting variation in language processing: A cognitive neuroscience approach. Colloquium, Department of Psychology, University of Melbourne.