Help in doing simulations
The website simulations provide examples of how the Multilink model functions. Note that for this purpose, the parameters here have been fixed to a limited set of values. We are currently optimizing the model with respect to its semantics. Thus, examples of semantic or associative priming are not intended to correspond to actual experimental results in any detail.
Realistically and smoothly simulating your experimental results at large may require somewhat different parameter settings, an adapted lexicon, and so on. For 'proper' simulations of large stimulus word lists, therefore please contact the Multilink team via Ton Dijkstra (firstname.lastname@example.org). We will be happy to help you!
You can provide two sorts of input, isolated letter strings and word pairs.
(1) Isolated letter strings that are words or non-words.
The 0: at the beginning indicates that the item is presented to Multilink on the first time cycle (counting starts at 0).
(2) Prime and target words that are orthographically or semantically related.
In the first example, the word HOUSE is presented on the first time cycle, while the word MOUSE replaces it on the fourth time cycle.
Note that you may accidentally choose words that are not in the present lexicon; they will be treated like non-words. Also note that interlingual homographs may not produce optimal results with these settings.
It is possible to present a list of several items to the model. Just put the items on separate lines.
The model can run with standard Lateral Inhibition (LI = -0.0001), strong Lateral Inhibition (LI = -0.1), or no Lateral Inhibition (LI disabled) between word form nodes (orthography and phonology).
The model can be run for different tasks. Output can be based on different representations, like Orthography in Lexical Decision or Phonology in Word Naming. In the Recognition (R) task, recognition is assumed to take place if word activation surpasses a recognition threshold of .70. Currently, the output of the R task and the Lexical Decision task are the same, because we cannot yet account for stimulus list composition and nonwords. Further note that lexical activation in the network itself does not depend on task; however, the output criteria and output codes are task dependent.
For some tasks and input materials, it is necessary to specify whether the input is Dutch or English, orthographic or phonological. This specification will be ignored if it is not task-relevant: Activation of input representations is assumed to be language-independent.
This specifies the representation that Multilink uses as response production criterion. For instance, if a Dutch word is given as Dutch input orthography and the task is word translation, English phonology (English:P) should be specified as output code. If instead you specify Dutch phonology (Dutch:P) as output code, you may see your activated Dutch input word in the graph. Similar confusions may result in a 'No response' in your .csv file.
The output can be produced if an activation threshold of .70 is surpassed (Threshold), but also on the difference between the first and second activated nodes (First Second Difference or FSDiff), and on the Luce choice rule (First All Ratio or FARatio; activation of a node relative to total activation summed across active nodes).
You can ask for three different types of output: a data file (.csv), a graph, and a matrix of activations for the input word at different time steps. Note that the activation curve and matrix file are only produced if your input consists of just one item per line.