jIAM: Interactive Activation Models in JavaScript

Walter van Heuven > Research > jIAM


jIAM_animation

jIAM makes it possible to run simulations with a number of Interactive Activation models in a web browser. Currently, jIAM implements the following models:

  • Interactive Activation (IA) model (McCelland & Rumelhart, 1981, 1988; Rumelhart & McClelland, 1982). Two versions of the IA model can be created. The default version of the IA model is identical to the original IA model, with feature, letter and word layers. The other version of the IA model has no feature layer, just letter and word layers. The output of both models is similar when the input contains only letters. Unlike the original implementation of the IA model, the model can handle words of any length, although parameters need to be adjusted when the words are not 4 letters long.
  • Bilingual Interactive Activation (BIA) model (Dijkstra & van Heuven, 1998; van Heuven et al., 1998). The BIA model can have two or more lexicons and the feature layer can be omitted.
  • BIA+ model (orthography only and without task/decision system) (Dijkstra & van Heuven, 2002). The BIA+ model can have two or more lexicons and the feature layer can be omitted.

jIAM uses a neural network JavasScript library IANE (Interactive Activate Network Engine) that I have developed for creating connectionist models. This library is based on the IANE libraries that I have written in Java and Python. The IANE library and the interactive activation models in jIAM are written in JavaScript. For the graphical interface I used a combination of JavaScript, HTML5, and Tumult Hype.

jIAM works in modern web browsers on OS X, Windows, Linux, iOS, and Android. For best performance use Safari or Firefox.

Click here to start jIAM

How to run a simulation

To create the standard IA model click first on 'Create Model'. Next, enter a string of 4 characters in the input field, then press the 'Start' button to run the simulation. A graph will be presented after the simulation with the activations of the active nodes. Detailed information about the active nodes and the connections in the network are presented when you click on the 'Nodes' button. To change to a different model or to adjust parameters and other model settings click on 'Settings'.

Special input characters

'_'Absence and presence (letter feature) units off
'.'Absence units on and presence units off
'*'Letter feature units of letter K and R on (ambiguous R-K character)
'#'Mask character; Features of the letters X and O are turned on

Lexicons and Parameters

The default lexicon for the IA model in jIAM is an English 4-letter word lexicon (word.lis, which is the lexicon provided with the PDP handbook). However, you can use your own lexicon in jIAM as well. Prepare a text file with two columns separated by space or tab. The first column should contain words (letter strings) and the second column their word frequencies. When the text file contains resting-level activations use ".lis" or ".rla" in the filename. Words can either have all the same length or they can have different lengths. jIAM will add spaces to words to align all words in the lexicon, just as in the DRC model (Coltheart et al., 2001). Bottom-up letter-to-word excitation and inhibition parameters should be changed when words are longer or shorter than 4 letters, see for example, Grainger and Jacobs (1998) and Loncke et al. (2009) for how these parameters could be adjusted. Please note that jIAM will be much slower when using large lexicons (> 5000 words), in particular when creating the network. For best preformance use Safari or Firefox. jIAM works also on iOS devices and lexicons can be imported via iCloud, Google Drive, and DropBox.

A word identification threshold can be set at the word level (e.g., 0.70) so that the simulation will stop when a word node reaches this threshold. Furthermore, to increase the temporal resolution of the model you can change the integration rate / stepsize in the model from 1.0 to 0.1 as proposed by Davis (2003) and Davis and Lupker (2006).

A detailed description of the IA model can be found in McClelland and Rumelhart (1981), Rumelhart and McClelland (1982), and in the PDP handbook (McClelland & Rumelhart, 1988; Chapter 7).

BIA and BIA+

The lexicons included with jIAM for BIA and BIA+ (dutch4.fpm and english4.fpm) are the same as used in the simulations with BIA reported in Dijkstra and van Heuven (1998), van Heuven et al. (1998) and Dijkstra et al. (1998). Please note that the BIA+ model in jIAM is only a partly implemented version of BIA+ (no phonology, semantics and task/decision system). The two models in jIAM are basically the same except that BIA+ does not have the option for top-down inhibition from language nodes to words.

References

Coltheart, M., Rastle, K., Perry, C., Langdon, R., & Ziegler, J. (2001). DRC: A dual route cascaded model of visual word recognition and reading aloud. Psychological Review, 108, 204–256.

Davis, C. J. (2003). Factors underlying masked priming effects in com- petitive network models of visual word recognition. In S. Kinoshita & S. J. Lupker (Eds.), Masked priming: The state of the art (pp. 121–170), Hove, England: Psychology Press.

Davis, C.J., & Lupker, S.J. (2006). Masked inhibitory priming in English: Evidence for lexical inhibition. Journal of Experimental Psychology: Human Perception and Performance, 32, 668-687.

Dijkstra, T., & van Heuven, W.J.B. (1998). The BIA-model and bilingual word recognition. In J. Grainger, & A. M. Jacobs (Eds.), Localist connectionist approaches to human cognition (pp. 189-225). Mahwah, NJ: Erlbaum (Scientific psychology series. Edited volumes).

Dijkstra, T., & van Heuven, W.J.B. (2002). The architecture of the bilingual word recognition system: From identification to decision. Bilingualism: Language and Cognition, 5, 175-197.

Dijkstra, T., van Heuven, W.J.B., & Grainger, J. (1998). Simulating cross-language competition with the bilingual interactive activation model. Psychologica Belgica, 38, 177-196.

Grainger, J., & Jacobs, A. M. (1996). Orthographic processing in visual word recognition: A multiple read- out model. Psychological Review, 103, 518–565.

Loncke, M., Martensen, H., van Heuven, W.J.B., & Sandra, D. (2009). Who is dominating the Dutch neighborhood? On the role of subsyllabic units in Dutch nonword reading. The Quarterly Journal of Experimental Psychology, 62, 140-154.

McClelland, J. L., & Rumelhart, D. E. (1981). An interactive activation model of context effects in letter perception, Part 1: An account of basic findings. Psychological Review, 88, 375-405.

McClelland, J. L., & Rumelhart, D. E. (1988). Explorations in Parallel Distributed Processing: A Handbook of Models, Programs, and Exercises. Cambridge, MA: MIT Press.

Rumelhart, D. E. & McClelland, J. L. (1982). An interactive activation model of context effects in letter perception: Part 2. The contextual enhancement effect and some test ans extensions of the model. Psychological Review, 89, 60-94.

van Heuven, W.J.B., Dijkstra, T., & Grainger, J. (1998). Orthographic neighborhood effects in bilingual word recognition. Journal of Memory and Language, 39, 458-483.