Naive Discriminative Learning

Description

Naive discriminative learning implements learning and classification models based on the Rescorla-Wagner equations and their equilibrium equations.

Naive discriminative learning implements classification models based on the Rescorla-Wagner equations and the equilibrium equations of the Rescorla-Wagner equations. This package provides three kinds of functionality: (1) discriminative learning based directly on the Rescorla-Wagner equations, (2) a function implementing the naive discriminative reader, and a model for silent (single-word) reading, and (3) a classifier based on the equilibrium equations. The functions and datasets for the naive discriminative reader model make it possible to replicate the simulation results for Experiment 1 of Baayen et al. (2011). The classifier is provided to allow for comparisons between machine learning (svm, TiMBL, glm, random forests, etc.) and discrimination learning. Compared to standard classification algorithms, naive discriminative learning may overfit the data, albeit gracefully.

Details

The DESCRIPTION file:

Index of help topics:

RescorlaWagner          Implementation of the Rescorla-Wagner
                        equations.
acts2probs              Calculate probability matrix from activation
                        matrix, as well as predicted values
anova.ndlClassify       Analysis of Model Fit for Naive Discriminatory
                        Reader Models
crosstableStatistics    Calculate statistics for a contingency table
cueCoding               code a vector of cues as n-grams
danks                   Example data from Danks (2003), after Spellman
                        (1996).
dative                  Dative Alternation
estimateActivations     Estimation of the activations of outcomes
                        (meanings)
estimateWeights         Estimation of the association weights using the
                        equilibrium equations of Danks (2003) for the
                        Rescorla-Wagner equations.
estimateWeightsCompact
                        Estimation of the association weights using the
                        equilibrium equations of Danks (2003) for the
                        Rescorla-Wagner equations using a compact
                        binary event file.
learn                   Count cue-outcome co-occurences needed to run
                        the Danks equations.
learnLegacy             Count cue-outcome co-occurrences needed to run
                        the Danks equations.
lexample                Lexical example data illustrating the
                        Rescorla-Wagner equations
modelStatistics         Calculate a range of goodness of fit measures
                        for an object fitted with some multivariate
                        statistical method that yields probability
                        estimates for outcomes.
ndl-package             Naive Discriminative Learning
ndlClassify             Classification using naive discriminative
                        learning.
ndlCrossvalidate        Crossvalidation of a Naive Discriminative
                        Learning model.
ndlCuesOutcomes         Creation of dataframe for Naive Discriminative
                        Learning from formula specification
ndlStatistics           Calculate goodness of fit statistics for a
                        naive discriminative learning model.
ndlVarimp               Permutation variable importance for
                        classification using naive discriminative
                        learning.
numbers                 Example data illustrating the Rescorla-Wagner
                        equations as applied to numerical cognition by
                        Ramscar et al. (2011).
orthoCoding             Code a character string (written word form) as
                        letter n-grams
plot.RescorlaWagner     Plot function for the output of
                        'RescorlaWagner'.
plot.ndlClassify        Plot function for selected results of
                        'ndlClassify'.
plurals                 Artificial data set used to illustrate the
                        Rescorla-Wagner equations and naive
                        discriminative learning.
predict.ndlClassify     Predict method for ndlClassify objects
random.pseudoinverse    Calculate an approximation of the pseudoinverse
                        of a matrix.
serbian                 Serbian case inflected nouns.
serbianLex              Serbian lexicon with 1187 prime-target pairs.
serbianUniCyr           Serbian case inflected nouns (in Cyrillic
                        Unicode).
serbianUniLat           Serbian case inflected nouns (in Latin-alphabet
                        Unicode).
summary.ndlClassify     A summary of a Naive Discriminatory Learning
                        Model
summary.ndlCrossvalidate
                        A summary of a crossvalidation of a Naive
                        Discriminatory Reader Model
think                   Finnish 'think' verbs.

For more detailed information on the core Rescorla-Wagner equations, see the functions RescorlaWagner and plot.RescorlaWagner, as well as the data sets danks, numbers (data courtesy of Michael Ramscar), and lexample (an example discussed in Baayen et al. 2011).

The functions for the naive discriminative learning (at the user level) are estimateWeights and estimateActivations. The relevant data sets are serbian, serbianUniCyr,serbianUniLat, and serbianLex. The examples for serbianLex present the full simulation for Experiment 1 of Baayen et al. (2011).

Key functionality for the user is provided by the functions orthoCoding, estimateWeights, and estimateActivations. orthoCoding calculates the letter n-grams for character strings, to be used as cues. It is assumed that meaning or meanings (separated by underscores if there are more then one) are available as outcomes. The frequency with which each (unique) combination of cues and outcomes occurs are required. For some example input data sets, see: danks, plurals, serbian, serbianUniCyr and serbianUniLat.

The function estimateWeights estimates the association strengths of cues to outcomes, using the equilibrium equations presented in Danks (2003). The function estimateActivations estimates the activations of outcomes (meanings) given cues (n-grams).

The Rcpp-based learn and learnLegacy functions use a C++ function to compute the conditional co-occurrence matrices required in the equilibrium equations. These are internally used by estimateWeights and should not be used directly by users of the package.

The key function for naive discriminative classification is ndlClassify; see data sets think and dative for examples.

Author(s)

NA

Maintainer: Tino Sering <konstantin.sering@uni-tuebingen.de>

Author Contributions: Initial concept by R. Harald Baayen with contributions from Petar Milin and Peter Hendrix. First R coding done by R. Harald Baayen.

Initial R package development until version 0.1.6 by Antti Arppe. Initial documentation by Antti Arppe. Initial optimizations in C by Petar Milin and Antti Arppe.

Classification functionality developed further by Antti Arppe.

In version 0.2.14 to version 0.2.16, improvements to the NDL algorithm by Petar Milin and Cyrus Shaoul. In version 0.2.14 to version 0.2.16, improved performance optimizations (C++ and Rcpp) by Cyrus Shaoul.

From version 0.2.17 onwards bug fixes and cran compliance by Tino Sering.

References

Baayen, R. H. and Milin, P. and Filipovic Durdevic, D. and Hendrix, P. and Marelli, M., An amorphous model for morphological processing in visual comprehension based on naive discriminative learning. Psychological Review, 118, 438-482.

Baayen, R. H. (2011) Corpus linguistics and naive discriminative learning. Brazilian Journal of Applied Linguistics, 11, 295-328.

Arppe, A. and Baayen, R. H. (in prep.) Statistical classification and principles of human learning.

Examples

## Not run: 
# Rescorla-Wagner
data(lexample)

lexample$Cues <- orthoCoding(lexample$Word, grams=1)
lexample.rw <- RescorlaWagner(lexample, nruns=25, traceCue="h",
   traceOutcome="hand")
plot(lexample.rw)
mtext("h - hand", 3, 1)

data(numbers)

traceCues <- c( "exactly1", "exactly2", "exactly3", "exactly4", "exactly5",
   "exactly6", "exactly7", "exactly10", "exactly15")
traceOutcomes <- c("1", "2", "3", "4", "5", "6", "7", "10", "15")

ylimit <- c(0,1)
par(mfrow=c(3,3), mar=c(4,4,1,1))

for (i in 1:length(traceCues)) {
  numbers.rw <- RescorlaWagner(numbers, nruns=1, traceCue=traceCues[i],
     traceOutcome=traceOutcomes[i])
  plot(numbers.rw, ylimit=ylimit)
  mtext(paste(traceCues[i], " - ", traceOutcomes[i], sep=""), side=3, line=-1,
    cex=0.7)
}
par(mfrow=c(1,1))

# naive discriminative learning (for complete example, see serbianLex)
# This function uses a Unicode dataset.
data(serbianUniCyr)
serbianUniCyr$Cues <- orthoCoding(serbianUniCyr$WordForm, grams=2)
serbianUniCyr$Outcomes <- serbianUniCyr$LemmaCase
sw <- estimateWeights(cuesOutcomes=serbianUniCyr,hasUnicode=T)

desiredItems <- unique(serbianUniCyr["Cues"])
desiredItems$Outcomes=""
activations <- estimateActivations(desiredItems, sw)$activationMatrix
rownames(activations) <- unique(serbianUniCyr[["WordForm"]])

syntax <- c("acc", "dat", "gen", "ins", "loc", "nom", "Pl",  "Sg") 
activations2 <- activations[,!is.element(colnames(activations), syntax)]
head(rownames(activations2),50)
head(colnames(activations2),8)

image(activations2, xlab="word forms", ylab="meanings", xaxt="n", yaxt="n")
mtext(c("yena", "...", "zvuke"), side=1, line=1, at=c(0, 0.5, 1),  adj=c(0,0,1))
mtext(c("yena", "...", "zvuk"), side=2, line=1, at=c(0, 0.5, 1),   adj=c(0,0,1))

# naive discriminative classification
data(think)
think.ndl <- ndlClassify(Lexeme ~ Person + Number + Agent + Patient + Register,
   data=think)
summary(think.ndl)
plot(think.ndl, values="weights", type="hist", panes="multiple")
plot(think.ndl, values="probabilities", type="density")

## End(Not run)

Implementation of the Rescorla-Wagner equations.

Description

RescorlaWagner implements an iterative simulation based on the Rescorla-Wagner equations. Given a data frame specifying cues, outcomes, and frequencies, it calculates, for a given cue-outcome pair, the temporal sequence of developing weights.

Usage

RescorlaWagner(cuesOutcomes, traceCue="h", traceOutcome="hand",
   nruns=1, random=TRUE, randomOrder = NA, alpha=0.1, lambda=1,
   beta1=0.1, beta2=0.1)

Arguments

Details

The equilibrium weights (Danks, 2003) are also estimated.

Value

An object of the class "RescorlaWagner", being a list with the following components:

weightvector

A numeric vector with the weights for all nruns*sum(dat[,"Frequency"]) training trials.

equilibriumWeight

The weight of the cue-outcome link at equilibrium.

traceCue

A character string specifying the trace cue.

traceOutcome

A character string specifying the trace outcome.

Author(s)

R. H. Baayen and Antti Arppe

References

Danks, D. (2003). Equilibria of the Rescorla-Wagner model. Journal of Mathematical Psychology, 47 (2), 109-121.

Rescorla, R. A., & Wagner, A. R. (1972). A theory of Pavlovian conditioning: Variations in the effectiveness of reinforcement and nonreinforcement. In Black, A. H., & Prokasy, W. F. (Eds.), Classical conditioning II: Current research and theory (pp. 64-99). New York: Appleton-Century-Crofts.

See Also

orthoCoding, plot.RescorlaWagner, numbers

Examples

data(lexample)
lexample$Cues <- orthoCoding(lexample$Word, grams=1)
lexample.rw <- RescorlaWagner(lexample, nruns=25, 
   traceCue="h", traceOutcome="hand")
plot(lexample.rw)

data(numbers)
traceCues=c( "exactly1", "exactly2", "exactly3", "exactly4",
   "exactly5", "exactly6", "exactly7", "exactly10", "exactly15")
traceOutcomes=c("1", "2", "3", "4", "5", "6", "7", "10", "15")
ylimit=c(0,1)
par(mfrow=c(3,3),mar=c(4,4,1,1))
     
for(i in 1:length(traceCues)) {
   numbers.rw <- RescorlaWagner(numbers, nruns=1,
      traceCue=traceCues[i], traceOutcome=traceOutcomes[i])
    plot(numbers.rw, ylimit=ylimit)
    mtext(paste(traceCues[i], " - ", traceOutcomes[i], sep=""), 
       side=3, line=-1, cex=0.7)
  }
par(mfrow=c(1,1))

Calculate probability matrix from activation matrix, as well as predicted values

Description

acts2probs takes the activation matrix returned by ndlClassify and calculates the matrix of probabilities for the estimated activation matrix, as well as the predicted values of the response variable.

Usage

acts2probs(acts)

Arguments

Details

Probabilities in p are obtained by adding the absolute value of the minimum activation to the activation matrix, and renorming. The selection rule used to produce predicted is to choose for each instance in the data the outcome value that has received the highest probability estimate.

Value

A list with the following components:

p

a matrix with the probabilities of the levels of the response variable for each observation.

predicted

a character vector with predicted values.

Author(s)

Harald Baayen and Antti Arppe

References

Arppe, A. and Baayen, R. H. (in prep.). Statistical classification and principles of human learning.

See Also

See also ndlClassify.

Examples

data(think)
think.ndl <- ndlClassify(Lexeme ~ Person + Number + Agent + Register, data=think)
pdata <- acts2probs(think.ndl$activationMatrix)

Analysis of Model Fit for Naive Discriminatory Reader Models

Description

Compute an analysis of individual variable contributions or model comparisons for one or more Naive Discriminatory Reader model fits.

Usage

 ## S3 method for class 'ndlClassify'
anova(object, ..., statistic = "deviance", test = "Chisq")

Arguments

Details

Currently, comparison of the terms of a single model or multiple models is only implemented based on the deviance statistic.

Specifying a single object gives a sequential analysis of deviance table for that fit. That is, the reductions in the residual deviance as each term of the formula is added in turn are given in as the rows of a table, plus the residual deviances themselves.

If more than one object is specified, the table has a row for the residual degrees of freedom and deviance for each model. For all but the first model, the change in degrees of freedom and deviance is also given. (This only makes statistical sense if the models are nested.) It is conventional to list the models from smallest to largest, but this is up to the user.

The table will contain test statistics (and P values) comparing the reduction in deviance for the row to the residuals. Only a comparison of models or contributions of their components by the chi-squared test has been implemented.

The comparison between two or more models by anova or anova.ndlClassifylist will only be valid if they are fitted to the same dataset. If anova.ndlClassifylist detects this, it will stop and report an error.

Value

An object of class "anova" inheriting from class "data.frame".

Author(s)

Antti Arppe

References

Arppe, A. and Baayen, R. H. (in prep.) Statistical classification and principles of human learning.

See Also

ndlClassify

Examples

data(think)
set.seed(314)
think <- think[sample(1:nrow(think),500),]

think.ndl1 <- ndlClassify(Lexeme ~ Agent * Person, data=think)
anova(think.ndl1)

think.ndl2 <- ndlClassify(Lexeme ~ Agent * Person + Patient, data=think)
anova(think.ndl1, think.ndl2)

Calculate statistics for a contingency table

Description

crosstableStatistics takes a contingency table of observed vs. predicted values for a binary or polytomous response variable as input, and calculates a range of statistics about prediction accuracy.

Usage

crosstableStatistics(ctable)

Arguments

Value

A list with the following components:

accuracy

Overall prediction accuracy

recall.predicted

Recall of prediction for each outcome value

precision.predicted

Precision of prediction for each outcome value

lambda.prediction

lambda for prediction accuracy (improvement over baseline of always predicting mode)

tau.classification

tau for classification accuracy (improvement over baseline of homogeneous distribution of predicted outcomes)

d.lambda.prediction

d(lambda): used for calculating P(lambda)

d.tau.classification

d(tau): used for calculating P(tau)

p.lambda.prediction

P(lambda): probability of reaching lambda by chance

p.tau.classification

P(tau): probability of reaching tau by chance

Author(s)

Antti Arppe and Harald Baayen

References

Arppe, A. 2008. Univariate, bivariate and multivariate methods in corpus-based lexicography – a study of synonymy. Publications of the Department of General Linguistics, University of Helsinki, No. 44. URN: http://urn.fi/URN:ISBN:978-952-10-5175-3.

Arppe, A. and Baayen, R. H. (in prep.). Statistical classification and principles of human learning.

Menard, Scott (1995). Applied Logistic Regression Analysis. Sage University Paper Series on Quantitative Applications in the Social Sciences 07-106. Thousand Oaks: Sage Publications.

See Also

See also modelStatistics, ndlStatistics, ndlClassify.

Examples

ctable <- matrix(c(30, 10, 5, 60), 2, 2)
crosstableStatistics(ctable)

code a vector of cues as n-grams

Description

cueCoding codes a vector of cues into unigrams, bigrams, ..., n-grams, with unigrams as default.

Usage

cueCoding(cues = c("hello", "world"), maxn=1, adjacent=FALSE)

Arguments

Value

A vector of cue n-grams, one for each word in the input vector cues. Each n-gram vector lists the constituent unigrams, bigrams, etc., separated by underscores.

Author(s)

Antti Arppe and Harald Baayen

References

Arppe, A. and Baayen, R. H. (in prep.). Statistical classification and principles of human learning.

See Also

See also ndlClassify, ndlCuesOutcomes, ndlVarimp, ndlCrossvalidate.

Examples

# Cues from the \code{think} data: Person, Number, Register
cues <- c("First", "Plural", "hs95")
cueCoding(cues, maxn=1)
cueCoding(cues, maxn=2)

Example data from Danks (2003), after Spellman (1996).

Description

Data of Spellman (1996) used by Danks (2003) to illustrate the equilibrium equations for the Rescorla-Wagner model. There are two liquids (red and blue) that are potentially fertilizers, and the experimental participant is given the rates at which flowers bloom for the four possible conditions (no liquid, red liquid, blue liquid, and both liquids).

Usage

data(danks)

Format

A data frame with 8 observations on the following 3 variables.

Cues

A character vector specifying the cues. The pots in which the flowers are grown, and the color of the fertilizer. Individual cues are separated by underscores.

Outcomes

A character vector specifying whether plants flowered (y or n).

Frequency

A numeric vector specifying the frequency of flowering.

Details

For details, see Danks (2003: 112).

Source

B. A. Spellman, (1996). Conditionalizing causality. In Shanks, D. R., Holyoak, K. J., & Medin, D. L. (Eds.), Causal learning: the psychology of learning and motivation, Vol. 34 (pp. 167-206). San Diego, CA: Academic Press.

References

D. Danks (2003), Equilibria of the Rescorla-Wagner model. Journal of Mathematical Psychology 47, 109-121.

B. A. Spellman, (1996). Conditionalizing causality. In Shanks, D. R., Holyoak, K. J., & Medin, D. L. (Eds.), Causal learning: the psychology of learning and motivation, Vol. 34 (pp. 167-206). San Diego, CA: Academic Press.

Examples

data(danks)
estimateWeights(cuesOutcomes=danks)

Dative Alternation

Description

Data describing the realization of the dative as NP or PP in the Switchboard corpus and the Treebank Wall Street Journal collection.

Usage

data(dative)

Format

A data frame with 3263 observations on the following 15 variables.

Speaker

a factor coding speaker; available only for the subset of spoken English.

Modality

a factor with levels spoken, written.

Verb

a factor with the verbs as levels.

SemanticClass

a factor with levels a (abstract: 'give it some thought'), c (communication: 'tell, give me your name'), f (future transfer of possession: 'owe, promise'), p (prevention of possession: 'cost, deny'), and t (transfer of possession: 'give an armband, send').

LengthOfRecipient

a numeric vector coding the number of words comprising the recipient.

AnimacyOfRec

a factor with levels animate and inanimate for the animacy of the recipient.

DefinOfRec

a factor with levels definite and indefinite coding the definiteness of the recipient.

PronomOfRec

a factor with levels nonpronominal and pronominal coding the pronominality of the recipient.

LengthOfTheme

a numeric vector coding the number of words comprising the theme.

AnimacyOfTheme

a factor with levels animate and inanimate coding the animacy of the theme.

DefinOfTheme

a factor with levels definite and indefinite coding the definiteness of the theme.

PronomOfTheme

a factor with levels nonpronominal and pronominal coding the pronominality of the theme.

RealizationOfRecipient

a factor with levels NP and PP coding the realization of the dative.

AccessOfRec

a factor with levels accessible, given, and new coding the accessibility of the recipient.

AccessOfTheme

a factor with levels accessible, given, and new coding the accessibility of the theme.

References

Bresnan, J., Cueni, A., Nikitina, T. and Baayen, R. H. (2007) Predicting the dative alternation, in Bouma, G. and Kraemer, I. and Zwarts, J. (eds.), Cognitive Foundations of Interpretation, Royal Netherlands Academy of Sciences, 69-94.

Examples

## Not run: 
data(dative)
out <- which(is.element(colnames(dative), c("Speaker","Verb")))
dative <- dative[,-out]
dative.ndl <- ndlClassify(RealizationOfRecipient ~ ., data=dative)
ndlStatistics(dative.ndl)


## End(Not run) 

Estimation of the activations of outcomes (meanings)

Description

estimateActivations is used to estimate the activations for outcomes (meanings) using the equilibrium association strengths (weights) for the Rescorla-Wagner model.

Usage

estimateActivations(cuesOutcomes, weightMatrix, unique=FALSE, ...)

Arguments

Details

The activation of an outcome is defined as the sum of the weights on the incoming links from active cues. When the input (the Cues in cuesOutcomes) contain elements that are not present in the rownames of the weightMatrix, such new cues are added to the weightMatrix with zero entries. The set of exemplars in cuesOutcomes may contain rows with identical cue sets but different outcome sets. Consequently, for such rows, identical vectors of activations of outcomes are generated. In the activation matrix returned by estimateActivations, such duplicate entries are removed.

For examples of how the cuesOutcomes data frame should be structured, see the data sets danks, plurals, and serbian. For examples of how the weightMatrix should be structured, see the corresponding output of estimateWeights.

Value

A list with the following components:

activationMatrix

A matrix with as dimensions, for rows, the number of exemplars (by-row cue sets, typically word forms), and for columns, the number of unique outcomes (meanings), specifying the activation of a meaning given the cues in the input for a given exemplar.

newCues

A vector of cues encountered in cuesOutcomes which were not present in weightMatrix.

...

Control arguments to be passed along from ndlClassify, and/or ndlCrossvalidate.

Author(s)

R. H. Baayen & Antti Arppe

References

Baayen, R. H. and Milin, P. and Filipovic Durdevic, D. and Hendrix, P. and Marelli, M., An amorphous model for morphological processing in visual comprehension based on naive discriminative learning. Psychological Review, 118, 438-482.

See Also

estimateWeights, danks, plurals, serbian

Examples

  data(serbian)
  serbian$Cues <- orthoCoding(serbian$WordForm, grams=2)
  serbian$Outcomes <- serbian$LemmaCase
  sw <- estimateWeights(cuesOutcomes=serbian)
  sw[1:5,1:6]
  activations <- estimateActivations(unique(serbian["Cues"]), sw)$activationMatrix
  rownames(activations) <- unique(serbian[["WordForm"]])
  activations[1:5,1:6]

  syntax <- c("acc", "dat", "gen", "ins", "loc", "nom", "Pl", "Sg") 
  activations2 <- activations[,!is.element(colnames(activations),syntax)]
  head(rownames(activations2), 50)
  head(colnames(activations2), 8)
  image(activations2, xlab="word forms", ylab="meanings", xaxt="n", yaxt="n")
  mtext(c("yena", "...", "zvuke"), side=1, line=1, at=c(0, 0.5, 1), adj=c(0,0,1))
  mtext(c("yena", "...", "zvuk"), side=2, line=1, at=c(0, 0.5, 1), adj=c(0,0,1))

Estimation of the association weights using the equilibrium equations of Danks (2003) for the Rescorla-Wagner equations.

Description

A function to estimate the weights (associative strengths) for cue-outcome pairs when learning is in equilibrium, using the equilibrium equations for the Rescorla-Wagner model of Danks (2003).

Usage

estimateWeights(cuesOutcomes, removeDuplicates=TRUE, saveCounts=FALSE,
verbose=FALSE, trueCondProb=TRUE, addBackground=FALSE, hasUnicode=FALSE, ...)

Arguments

Details

Using Rcpp, a C++ based implementation processes all of the data in RAM. The module will check the amount of RAM you have available in your system and warn you if the amount of RAM is insufficient to build your model.

For examples of how the cuesOutcomes data frame should be structured, see the data sets danks, plurals, and serbian. N.B. Empty Cues or Outcomes (effectively having length = 0), e.g. Cues or Outcomes strings with an initial or final underscore or two immediately adjacent underscores, will result in an error.

Value

A matrix with cue-to-outcome association strengths. Rows are cues, and columns are outcomes. Rows and columns are labeled. If addBackground=T, a row named "Environ" will be added to the output.

Acknowledgements

The assistance of Uwe Ligges in getting the C function cooc to work within the R framework is greatly appreciated. This C function was removed in version 0.2.0 and replaced with the C++ function by Cyrus Shaoul.

Note

Add a note here.

Author(s)

Cyrus Shaoul, R. H. Baayen and Petar Milin, with contributions from Antti Arppe and Peter Hendrix.

References

Baayen, R. H. and Milin, P. and Filipovic Durdevic, D. and Hendrix, P. and Marelli, M. (2011), An amorphous model for morphological processing in visual comprehension based on naive discriminative learning. Psychological Review, 118, 438-482.

See Also

estimateActivations, ndlCuesOutcomes, danks, plurals, serbian

Examples

  data(danks)
  estimateWeights(cuesOutcomes=danks)

  data(plurals)
  plurals$Cues <- orthoCoding(plurals$WordForm, grams=1)
  round(estimateWeights(cuesOutcomes=plurals),2)
  
  data(serbian)
  serbian$Cues <- orthoCoding(serbian$WordForm, grams=2)
  serbian$Outcomes <- serbian$LemmaCase
  sw <- estimateWeights(cuesOutcomes=serbian)
  round(sw[1:5,1:6],2)

Estimation of the association weights using the equilibrium equations of Danks (2003) for the Rescorla-Wagner equations using a compact binary event file.

Description

A function to estimate the weights (associative strengths) for cue-outcome pairs when learning is in equilibrium, using the equilibrium equations for the Rescorla-Wagner model of Danks (2003) using a compact binary event file.

Usage

estimateWeightsCompact(datasource, removeDuplicates=TRUE,
saveCounts=FALSE, verbose=FALSE, MaxEvents=100000000000000,
trueCondProb=TRUE, addBackground=FALSE, ...)

Arguments

Details

Using Rcpp, a C++ based implementation processes all of the data RAM. The module will check the amount of RAM you have available in your system and warn you of RAM is insufficient to build your model.

Value

A matrix with cue-to-outcome association strengths. Rows are cues, and columns are outcomes. Rows and columns are labeled. If addBackground=T, a row named "Environ" will be added to the output.

Acknowledgements

Thanks to all the beta testers of the ndl package.

Note

Add a note here.

Author(s)

Cyrus Shaoul, R. H. Baayen and Petar Milin, with contributions from Antti Arppe and Peter Hendrix.

References

Baayen, R. H. and Milin, P. and Filipovic Durdevic, D. and Hendrix, P. and Marelli, M., (2011) An amorphous model for morphological processing in visual comprehension based on naive discriminative learning. Psychological Review, 118, 438-482.

See Also

estimateActivations

Examples

  message("This module requires data in a non-portable format to
demonstrate how it works.")

Count cue-outcome co-occurences needed to run the Danks equations.

Description

An internal function to count cue-outcome co-occurrences.

Usage

learn(data,RemoveDuplicates,verbose,MaxEvents,addBackground)

Arguments

Details

This function calls an Rcpp function of the same name to process the data in the compact data format.

Value

A list of two matrices with cue-cue coocurrences and cue-outcome cooccurrences and a vector with background rates.

Acknowledgements

Thanks to all the testers!

Note

No temporary files are used.

Author(s)

Cyrus Shaoul

References

Baayen, R. H. and Milin, P. and Filipovic Durdevic, D. and Hendrix, P. and Marelli, M., An amorphous model for morphological processing in visual comprehension based on naive discriminative learning. Psychological Review, 118, 438-482.

See Also

estimateActivations, ndlCuesOutcomes, estimateWeightsCompact, danks, plurals, serbian

Examples

#None (internal function)

Count cue-outcome co-occurrences needed to run the Danks equations.

Description

An internal function to count cue-outcome co-occurrences.

Usage

learnLegacy(DFin,RemoveDuplicates,verbose)

Arguments

Details

This function calls an Rcpp function of the same name to process the data in the DFin data frame.

Value

A list of two matrices with cue-cue co-occurrences and cue-outcome co-occurrences.

Acknowledgements

Thanks to all the testers out there! Martijn, you know who you are.

Note

No temporary files are used.

Author(s)

Cyrus Shaoul

References

Baayen, R. H. and Milin, P. and Filipovic Durdevic, D. and Hendrix, P. and Marelli, M., An amorphous model for morphological processing in visual comprehension based on naive discriminative learning. Psychological Review, 118, 438-482.

See Also

estimateActivations, ndlCuesOutcomes, estimateWeights, danks, plurals, serbian

Examples

#None (internal function)

Lexical example data illustrating the Rescorla-Wagner equations

Description

Ten monomorphemic and inflected English words with fictive frequencies, and meanings.

Usage

data(lexample)

Format

A data frame with 10 observations on the following 3 variables:

Word

A character vector specifying word forms

Frequency

A numeric vector with the – fictive – frequencies of occurrence of the words

Outcomes

A character vector specifying the meaning components of the words, separated by underscores

Details

This example lexicon is used in Baayen et al. (2011) (table 8, figure 4) to illustrate the Rescorla-Wagner equations.

References

Baayen, R. H., Milin, P., Filipovic Durdevic, D., Hendrix, P. and Marelli, M. (2011), An amorphous model for morphological processing in visual comprehension based on naive discriminative learning. Psychological Review, 118, 438-482.

See Also

RescorlaWagner, orthoCoding

Examples

## Not run: 
data(lexample)
lexample$Cues <- orthoCoding(lexample$Word, grams=1)
par(mfrow=c(2,2))
lexample.rw <- RescorlaWagner(lexample, nruns=25, traceCue="h",traceOutcome="hand")
plot(lexample.rw)
mtext("h - hand", 3, 1)

lexample.rw <- RescorlaWagner(lexample, nruns=25, traceCue="s",traceOutcome="plural")
plot(lexample.rw)
mtext("s - plural", 3, 1)

lexample.rw <- RescorlaWagner(lexample, nruns=25, traceCue="a",traceOutcome="as")
plot(lexample.rw)
mtext("a - as", 3, 1)

lexample.rw <- RescorlaWagner(lexample, nruns=25, traceCue="s",traceOutcome="as")
plot(lexample.rw)
mtext("s - as", 3, 1)
par(mfrow=c(1,1))

## End(Not run)

Calculate a range of goodness of fit measures for an object fitted with some multivariate statistical method that yields probability estimates for outcomes.

Description

modelStatistics calculates a range of goodness of fit measures.

Usage

  modelStatistics(observed, predicted, frequency=NA, p.values,
     n.data, n.predictors, outcomes=levels(as.factor(observed)),
     p.normalize=TRUE, cross.tabulation=TRUE, 
     p.zero.correction=1/(NROW(p.values)*NCOL(p.values))^2)

Arguments

Value

A list with the following components:

loglikelihood.null

Loglikelihood for null model

loglikelihood.model

Loglikelihood for fitted model

deviance.null

Null deviance

deviance.model

Model deviance

R2.likelihood

(McFadden's) R-squared

R2.nagelkerke

Nagelkerke's R-squared

AIC.model

Akaike's Information Criterion

BIC.model

Bayesian Information Criterion

C

index of concordance C (for binary response variables only)

crosstable

Crosstabulation of observed and predicted outcomes, if cross.tabulation=TRUE

crosstableStatistics(crosstable)

Various statistics calculated on crosstable with crosstableStatistics, if cross.tabulation=TRUE

Author(s)

Antti Arppe and Harald Baayen

References

Arppe, A. 2008. Univariate, bivariate and multivariate methods in corpus-based lexicography – a study of synonymy. Publications of the Department of General Linguistics, University of Helsinki, No. 44. URN: http://urn.fi/URN:ISBN:978-952-10-5175-3.

Arppe, A., and Baayen, R. H. (in prep.) Statistical modeling and the principles of human learning.

Hosmer, David W., Jr., and Stanley Lemeshow 2000. Applied Regression Analysis (2nd edition). New York: Wiley.

See Also

See also ndlClassify, ndlStatistics, crosstableStatistics.

Examples

data(think)
think.ndl <- ndlClassify(Lexeme ~ Agent + Patient, data=think)
probs <- acts2probs(think.ndl$activationMatrix)$p
preds <- acts2probs(think.ndl$activationMatrix)$predicted
n.data <- nrow(think)
n.predictors <- nrow(think.ndl$weightMatrix) *
   ncol(think.ndl$weightMatrix)
modelStatistics(observed=think$Lexeme, predicted=preds, p.values=probs,
   n.data=n.data, n.predictors=n.predictors)

Classification using naive discriminative learning.

Description

ndlClassify uses the equilibrium equations of Danks (2003) for the Rescorla-Wagner model (1972) to estimate association strengths (weights) for cues (typically levels of factorial predictors) to outcomes (typically a binary or polytomous response variable). Given the association strengths, the probability of a response level is obtained by summation over the weights on active incoming links.

Usage

ndlClassify(formula, data, frequency=NA, variable.value.separator="", ...)

## S3 method for class 'ndlClassify'
print(x, max.print=10, ...)

Arguments

Details

Classification by naive discriminative learning.

Value

A list of the class "ndlClassify" with the following components:

activationMatrix

A matrix specifying for each row of the input data frame the activations (probabilities) of the levels of the response variable (nrow observations by nlevels of response variable).

weightMatrix

A matrix specifying for each cue (predictor value) the association strength (weight) to each outcome (level of the response variable) (number of distinct predictor values by number of response levels).

cuesOutcomes

The input data structure for naive discriminative learning created by ndlCuesOutcomes based on the data argument (number of observations by 3: Frequency, Cues, Outcomes).

call

The call matched to fit the resulting "ndlClassify" object.

formula

The formula specified for fitting the resulting "ndlClassify" object.

data

The supplied data argument, excluding all elements not specified for the modeling task in formula and frequency.

Author(s)

R. H. Baayen and Antti Arppe

References

Baayen, R. H. and Milin, P. and Filipovic Durdevic, D. and Hendrix, P. and Marelli, M., An amorphous model for morphological processing in visual comprehension based on naive discriminative learning. Psychological Review, 118, 438-482.

Danks, D. (2003). Equilibria of the Rescorla-Wagner model. Journal of Mathematical Psychology, 47 (2), 109-121.

Rescorla, R. A., & Wagner, A. R. (1972). A theory of Pavlovian conditioning: Variations in the effectiveness of reinforcement and nonreinforcement. In Black, A. H., & Prokasy, W. F. (Eds.), Classical conditioning II: Current research and theory (pp. 64-99). New York: Appleton-Century-Crofts.

Arppe, A. and Baayen, R. H. (in prep.) Statistical classification and principles of human learning.

See Also

summary.ndlClassify, plot.ndlClassify, anova.ndlClassify, predict.ndlClassify, ndlCuesOutcomes, estimateWeights, cueCoding

Examples

data(think)
set.seed(314)
think <- think[sample(1:nrow(think),500),]
think.ndl <- ndlClassify(Lexeme ~ (Person * Number * Agent) + Register,
   data=think)
summary(think.ndl)

## Not run: 
think.ndl.SA <- ndlClassify(Lexeme ~ (Polarity + Voice + Mood + Person +
  Number + Covert + ClauseEquivalent + Agent + Patient + Manner + Time +
  Modality1 + Modality2 + Source + Goal + Quantity + Location +
  Duration + Frequency + MetaComment + ReasonPurpose + Condition +
  CoordinatedVerb)^2 + Author + Section, data=think)
summary(think.ndl.SA)

## End(Not run)

## Not run: 
data(dative)
out <- which(is.element(colnames(dative), c("Speaker","Verb")))
dative <- dative[-out]
dative.ndl <- ndlClassify(RealizationOfRecipient ~ ., data=dative)
summary(dative.ndl)


## End(Not run)

Crossvalidation of a Naive Discriminative Learning model.

Description

ndlCrossvalidate undertakes a crossvalidation of a Naive Discriminative Learning model fitted using ndlClassify.

Usage

ndlCrossvalidate(formula, data, frequency=NA, k=10, folds=NULL, ...)

Arguments

Details

Crossvalidation of a Naive Discriminative Learning model.

Value

A list of the class "ndlCrossvalidate" with the following components:

call

The call matched by ndlCrossvalidate

formula

The formula specified for ndlCrossvalidate

fits

A list of individual fits resulting from ndlCrossvalidate

k

The number of folds, by default equal to 10

n.total

The sum frequency of data points

n.train

The size of the training set

n.test

The size of of the testing set

folds

A list with the folds used in the crossvalidation; either selected at random by ndlCrossvalidate or provided by the user.

Author(s)

Antti Arppe

References

Baayen, R. H. and Milin, P. and Filipovic Durdevic, D. and Hendrix, P. and Marelli, M., An amorphous model for morphological processing in visual comprehension based on naive discriminative learning. Psychological Review, 118, 438-482.

Arppe, A. and Baayen, R. H. (in prep.). Statistical modeling and the principles of human learning.

See Also

summary.ndlCrossvalidate, ndlStatistics, ndlCuesOutcomes, cueCoding, estimateWeights, estimateActivations

Examples

data(think)
set.seed(314)
think <- think[sample(1:nrow(think),500),]
think.cv5 <- ndlCrossvalidate(Lexeme ~ Agent + Patient, data=think, k=5)
summary(think.cv5)
rm(think)

## Not run: 
data(think)
think.cv10 <- ndlCrossvalidate(Lexeme ~ Person + Number + Agent + Patient + Register,
   data=think, k=10)
summary(think.cv10)

## End(Not run)
## Not run: 
library(languageR)
data(finalDevoicing)
finDev.cv10 <- ndlCrossvalidate(Voice ~ Onset1Type + Onset2Type + VowelType *
   ConsonantType * Obstruent + Nsyll + Stress, data=finalDevoicing, k=10)
summary(finDev.cv10)


## End(Not run)

Creation of dataframe for Naive Discriminative Learning from formula specification

Description

ndlCuesOutcomes creates a dataframe for fitting a naive discriminative classification model with ndlClassify, using the specified formula and provided data.

Usage

ndlCuesOutcomes(formula, data, frequency=NA, 
  numeric2discrete=function(x) Hmisc::cut2(x,g=g.numeric), g.numeric=2,
  check.values=TRUE, ignore.absent=NULL, variable.value.separator="", ...)

Arguments

Details

Creates a dataframe to be used for fitting a Naive Discriminatory Learning classifier model.

Value

A dataframe with the following columns:

Frequency

Frequency with which the specific Cues and Outcomes co-occur.

Cues

A character vector of sets of Cues per instance, with Cues separated by underscore ‘_’.

Outcomes

A character vector of Outcomes per instance.

Author(s)

R. H. Baayen and Antti Arppe

References

Arppe, A. and Baayen, R. H. (in prep.) Statistical modeling and the principles of human learning.

See Also

cueCoding, ndlClassify

Examples

data(think)
set.seed(314)
think <- think[sample(1:nrow(think),500),]
think.CuesOutcomes <- ndlCuesOutcomes(Lexeme ~ (Person * Number * Agent) + Register, 
data=think)
head(think.CuesOutcomes)

## Not run: 
data(dative)
dative.cuesOutcomes <- ndlCuesOutcomes(RealizationOfRecipient ~ LengthOfRecipient +
   LengthOfTheme, data=dative, numeric2discrete=NULL)
table(dative.cuesOutcomes$Cues)

dative.cuesOutcomes1 <- ndlCuesOutcomes(RealizationOfRecipient ~ LengthOfRecipient +
   LengthOfTheme, data=dative)
table(dative.cuesOutcomes1$Cues)

dative.cuesOutcomes2 <- ndlCuesOutcomes(RealizationOfRecipient ~ LengthOfRecipient +
   LengthOfTheme, data=dative, g.numeric=3)
table(dative.cuesOutcomes2$Cues)


## End(Not run)

Calculate goodness of fit statistics for a naive discriminative learning model.

Description

ndlStatistics takes an Naive Discriminary Learning model object as generated by ndlClassify and calculates a range of goodness of fit statistics using modelStatistics.

Usage

ndlStatistics(ndl, ...)

Arguments

Value

A list with the following components:

n.data

sum frequency of data points

df.null

degrees of freedom of the Null model

df.model

degrees of freedom of the fitted model

statistics

a list of various measures of goodness of fit calculated with modelStatistics

Author(s)

Antti Arppe and Harald Baayen

References

Arppe, A. and Baayen, R. H. (in prep.) Statistical modeling and the principles of human learning.

See Also

See also ndlClassify, modelStatistics.

Examples

data(think)
set.seed(314)
think <- think[sample(1:nrow(think),500),]
think.ndl <- ndlClassify(Lexeme ~ Agent + Patient, data=think)
ndlStatistics(think.ndl)

## Not run: 
data(dative)
dative.ndl <- ndlClassify(RealizationOfRecipient ~ AnimacyOfRec + DefinOfRec +
   PronomOfRec + AnimacyOfTheme + DefinOfTheme + PronomOfTheme, data=dative)
ndlStatistics(dative.ndl)

## End(Not run)

Permutation variable importance for classification using naive discriminative learning.

Description

ndlVarimp uses permutation variable importance for naive discriminative classification models, typically the output of ndlClassify.

Usage

ndlVarimp(object, verbose=TRUE)

Arguments

Details

Variable importance is assessed using predictor permutation. Currently, conditional permutation variable importance (as for varimp for random forests in the party package) is not implemented.

Value

A list with two numeric vectors:

concordance

For binary response variables, a named vector specifying for each predictor the index of concordance when that predictor is permuted. For polytomous response variables, NA.

accuracy

A named vector specifying for each predictor the accuracy of the model with that predictor permuted.

Author(s)

R. H. Baayen and Antti Arppe

References

R. Harald Baayen (2011). Corpus linguistics and naive discriminative learning. Brazilian journal of applied linguistics, 11, 295-328.

Carolin Strobl, Anne-Laure Boulesteix, Thomas Kneib, Thomas Augustin and Achim Zeileis (2008). Conditional Variable Importance for Random Forests. BMC Bioinformatics, 9, 307.

See Also

summary.ndlClassify, plot.ndlClassify, anova.ndlClassify, ndlCuesOutcomes, estimateWeights, cueCoding

Examples

## Not run: 
data(dative)
dative <- dative[!is.na(dative$Speaker),-2]
dative.ndl <- ndlClassify(RealizationOfRecipient ~ ., data=dative)
dative.varimp <- ndlVarimp(dative.ndl)

library(lattice)
dotplot(sort(summary(dative.ndl)$statistics$accuracy-dative.varimp$accuracy), 
   xlab="permutation variable importance")

## End(Not run)

Example data illustrating the Rescorla-Wagner equations as applied to numerical cognition by Ramscar et al. (2011).

Description

The data used in simulation 3 of Ramscar et al. (2011) on numerical cognition.

Usage

data(lexample)

Format

A data frame with 10 observations on the following 3 variables.

Cues

A character vector specifying cues for quantities, separated by underscores.

Frequency

The frequencies with which the numbers appear in the COCA corpus.

Outcomes

A character vector specifying numerical outcomes associated with the input quantities.

Details

The cues represent learning trials with objects of the same size, shape and color. The numeric cues represent the presence of at least one subset of the specified size. The cues exactlyn represent the presence of exactly n objects. We are indebted to Michael Ramscar to making this data set available for inclusion in the package.

References

Michael Ramscar, Melody Dye, Hanna Muenke Popick & Fiona O'Donnell-McCarthy (2011), The Right Words or Les Mots Justes? Why Changing the Way We Speak to Children Can Help Them Learn Numbers Faster. Manuscript, Department of Psychology, Stanford University.

Examples

data(numbers)

traceCues=c( "exactly1", "exactly2", "exactly3", "exactly4",
"exactly5", "exactly6", "exactly7", "exactly10", "exactly15")
traceOutcomes=c("1", "2", "3", "4", "5", "6", "7", "10", "15")

ylimit=c(0,1)
par(mfrow=c(3,3),mar=c(4,4,1,1))

for (i in 1:length(traceCues)){
  numbers.rw = RescorlaWagner(numbers, nruns=1,
    traceCue=traceCues[i],traceOutcome=traceOutcomes[i])
  plot(numbers.rw, ylimit=ylimit)
  mtext(paste(traceCues[i], " - ", traceOutcomes[i], sep=""), 
    side=3, line=-1, cex=0.7)
}
par(mfrow=c(1,1))

Code a character string (written word form) as letter n-grams

Description

orthoCoding codes a character string into unigrams, bigrams, ..., n-grams, with as default bigrams as the substring size. If tokenization is not at the letter/character level, a token separator can be provided.

Usage

orthoCoding(strings=c("hel.lo","wor.ld"), grams = c(2), tokenized = F, sepToken = '.') 

Arguments

Value

A vector of grams (joined by underscores), one for each word in the input vector words.

Author(s)

Cyrus Shaoul, Peter Hendrix and Harald Baayen

References

Baayen, R. H. and Milin, P. and Filipovic Durdevic, D. and Hendrix, P. and Marelli, M., An amorphous model for morphological processing in visual comprehension based on naive discriminative learning. Psychological Review, 118, 438-482.

See Also

See also estimateWeights.

Examples

#Default
orthoCoding(tokenize=FALSE)
#With tokenizing on a specific character
orthoCoding(tokenize=TRUE)

#Comparing different n-gram sizes
data(serbian) 
serbian$Cues=orthoCoding(serbian$WordForm, grams=2)
head(serbian$Cues)
serbian$Cues=orthoCoding(serbian$WordForm, grams=c(2,4))
head(serbian$Cues)

Plot function for the output of RescorlaWagner.

Description

This function graphs the Rescorla-Wagner weights for a cue-outcome pair against learning time.

Usage

## S3 method for class 'RescorlaWagner'
plot(x, asymptote=TRUE, xlab="t", ylab="weight", ylimit=NA, ...)

Arguments

Value

A plot is produced on the graphics device.

Author(s)

R. H. Baayen and Antti Arppe

References

Danks, D. (2003). Equilibria of the Rescorla-Wagner model. Journal of Mathematical Psychology, 47 (2), 109-121.

Rescorla, R. A., & Wagner, A. R. (1972). A theory of Pavlovian conditioning: Variations in the effectiveness of reinforcement and nonreinforcement. In Black, A. H., & Prokasy, W. F. (Eds.), Classical conditioning II: Current research and theory (pp. 64-99). New York: Appleton-Century-Crofts.

See Also

RescorlaWagner, orthoCoding

Examples

data(lexample)
lexample$Cues <- orthoCoding(lexample$Word, grams=1)
lexample.rw <- RescorlaWagner(lexample, nruns=25, 
   traceCue="h", traceOutcome="hand")
plot(lexample.rw)
mtext("h - hand", 3, 1)

# Full example

## Not run: 
par(mfrow=c(2,2))
lexample.rw <- RescorlaWagner(lexample, nruns=25, 
   traceCue="h", traceOutcome="hand")
plot(lexample.rw)
mtext("h - hand", 3, 1)

lexample.rw <- RescorlaWagner(lexample, nruns=25, 
   traceCue="s", traceOutcome="plural")
plot(lexample.rw)
mtext("s - plural", 3, 1)

lexample.rw <- RescorlaWagner(lexample, nruns=25, 
   traceCue="a", traceOutcome="as")
plot(lexample.rw)
mtext("a - as", 3, 1)

lexample.rw <- RescorlaWagner(lexample, nruns=25, 
   traceCue="s", traceOutcome="as")
plot(lexample.rw)
mtext("s - as", 3, 1)
par(mfrow=c(1,1))

## End(Not run)

Plot function for selected results of ndlClassify.

Description

This function presents visually the estimated weights or expected probabilities for a model fitted with ndlClassify

Usage

## S3 method for class 'ndlClassify'
plot(x, values="weights", ...)

## S3 method for class 'ndlWeights'
plot(x, type="density", predictors=NULL, outcomes=NULL,
panes="single", lty=NULL, col=NULL, mfrow=NULL, main=NULL,
legend.position="topright", ...)

## S3 method for class 'ndlProbabilities'
plot(x, type="density", select="all",
panes="single", lty=NULL, col=NULL, pch=NULL, mfrow=NULL,
main=NULL, legend.position="topright", ...)

Arguments

Value

A plot of the selected type is produced on the graphics device.

Author(s)

Antti Arppe and R. H. Baayen

References

Arppe, A. and Baayen, R. H. (in prep.)

See Also

ndlClassify, acts2probs

Examples

## Not run: 

data(think)
think.ndl <- ndlClassify(Lexeme ~ Agent + Patient + Section, data=think)

plot(think.ndl, values="weights")
plot(think.ndl, values="weights", type="hist", panes="multiple")
plot(think.ndl, values="weights", type="density", panes="multiple")
plot(think.ndl, values="weights", type="density", panes="multiple",
   predictors="Section*")
plot(think.ndl, values="weights", type="density", panes="multiple",
   predictors="Patient*")
plot(think.ndl, values="weights", type="hist", panes="multiple", col=1:4)
plot(think.ndl, values="weights", type="density", panes="single",
   outcomes=c("ajatella","miettia","pohtia","harkita"))

plot(think.ndl, values="probabilities")
plot(think.ndl, values="probabilities", panes="multiple")
plot(think.ndl, values="probabilities", select="max")
plot(think.ndl, values="probabilities", select=c(1:3))
plot(think.ndl, values="probabilities", panes="multiple", select=c(1:3))
plot(think.ndl, values="probabilities", type="sort", legend.position="topleft")
plot(think.ndl, values="probabilities", type="sort", pch=".",
   legend.position="topleft")
plot(think.ndl, values="probabilities", type="sort", pch=".", panes="multiple")

## End(Not run)

Artificial data set used to illustrate the Rescorla-Wagner equations and naive discriminative learning.

Description

Data set with 10 English words of different (ad hoc) frequencies, each with a lexical meaning and a grammatical meaning.

Usage

data(plurals)

Format

A data frame with 10 observations on the following 3 variables:

WordForm

A character vector of word forms (cues).

Frequency

A numeric vector of frequencies.

Outcomes

A character vector of meanings (outcomes). Meanings are separated by underscores. The NIL meaning is ignored.

Source

Baayen, R. H. and Milin, P. and Filipovic Durdevic, D. and Hendrix, P. and Marelli, M., An amorphous model for morphological processing in visual comprehension based on naive discriminative learning. Psychological Review, 118, 438-482.

References

Baayen, R. H. and Milin, P. and Filipovic Durdevic, D. and Hendrix, P. and Marelli, M., An amorphous model for morphological processing in visual comprehension based on naive discriminative learning. Psychological Review, 118, 438-482.

Examples

data(plurals)
plurals$Cues <- orthoCoding(plurals$WordForm, grams=1)
estimateWeights(cuesOutcomes=plurals)

Predict method for ndlClassify objects

Description

Obtains predictions on the basis of a fitted "ndlClassify" object on data already incorporated in the object or on new data with the same predictors as the originally fitted model object.

Usage

## S3 method for class 'ndlClassify'
predict(object, newdata=NULL, frequency=NA,
   type="choice", ...)

Arguments

Details

If newdata is omitted the predictions are based on the data used for the fit.

Value

a vector predicted, or matrix of activations activations, or a matrix of predictions probabilities.

Author(s)

Antti Arppe

References

Arppe, A. and Baayen, R. H. (in prep.) Statistical classification and principles of human learning.

See Also

ndlClassify, estimateActivations, acts2probs

Examples

data(think)
think.ndl <- ndlClassify(Lexeme ~ Agent + Patient, data=think[1:300,])
head(predict(think.ndl, type="choice"))
predict(think.ndl, newdata=think[301:320,], type="probs")
predict(think.ndl, newdata=think[301:320,], type="acts")

Calculate an approximation of the pseudoinverse of a matrix.

Description

An internal function that uses an approximation of the SVD using the first k singular values of A to calculate the pseudo-inverse. Only used when the cue-cue matrix contains more than 20,000 cues.

Usage

random.pseudoinverse(m, verbose=F, k = 0)

Arguments

Details

This idea was proposed by Gunnar Martinsson Associate Professor and Director of Graduate Studies Department of Applied Mathematics, University of Colorado at Boulder http://amath.colorado.edu/faculty/martinss/ And with ideas from: Yoel Shkolnisky and his Out-of-Core SVD code: https://sites.google.com/site/yoelshkolnisky/software

Value

The approximate pseudoinverse of the input matrix

Acknowledgements

Thanks to Gunnar for his help with this!

Note

No temporary files are used.

Author(s)

Cyrus Shaoul

References

"Finding structure with randomness: Probabilistic algorithms for constructing approximate matrix decompositions" Nathan Halko, Per-Gunnar Martinsson, Joel A. Tropp http://arxiv.org/abs/0909.4061

See Also

estimateWeights, estimateWeightsCompact,

Examples

#None (internal function)

Serbian case inflected nouns.

Description

3240 case-inflected Serbian nouns and their frequencies, for 270 different masculine, feminine and neuter noun lemmas.

Usage

data(serbian)

Format

A data frame with 3240 observations on the following 3 variables:

WordForm

A character vector specifying the inflected word forms.

LemmaCase

A character vector specifying lemma (meaning), case, and number.

Frequency

A numeric vector specifying the frequency of each word form.

Details

Frequencies were taken from the Frequency Dictionary of Contemporary Serbian Language (Kostic, 1999). The 270 lemmas comprise the set of nouns for which each different case form appears at least once in this resource.

Source

Kostic, D. (1999). Frekvencijski recnik savremenog srpskog jezika (Frequency Dictionary of Contemporary Serbian Language). Institute for Experimental Phonetics and Speech Pathology & Laboratory of Experimental Psychology, University of Belgrade, Serbia.

References

Kostic, D. (1999). Frekvencijski recnik savremenog srpskog jezika (Frequency Dictionary of Contemporary Serbian Language). Institute for Experimental Phonetics and Speech Pathology & Laboratory of Experimental Psychology, University of Belgrade, Serbia.

Baayen, R. H., Milin, P., Filipovic Durdevic, D., Hendrix, P. and Marelli, M. (2011), An amorphous model for morphological processing in visual comprehension based on naive discriminative learning. Psychological Review, 118, 438-482.

See Also

See also serbianLex, estimateActivations.

Examples

data(serbian)
serbian$Cues <- orthoCoding(serbian$WordForm, grams=2)
serbian$Outcomes <- serbian$LemmaCase
sw <- estimateWeights(cuesOutcomes=serbian)
sw[1:5,1:5]
desiredItems <- unique(serbian["Cues"])
desiredItems$Outcomes=""
activations <- estimateActivations(desiredItems, sw)$activationMatrix
rownames(activations) <- unique(serbian[["WordForm"]])
activations[1:5,1:6]

Serbian lexicon with 1187 prime-target pairs.

Description

The 1187 prime-target pairs and their lexical properties used in the simulation study of Experiment 1 of Baayen et al. (2011).

Usage

data(serbianLex)

Format

A data frame with 1187 observations on the following 14 variables:

Target

A factor specifying the target noun form

Prime

A factor specifying the prime noun form

PrimeLemma

A factor specifying the lemma of the prime

TargetLemma

A factor specifying the target lemma

Length

A numeric vector with the length in letters of the target

WeightedRE

A numeric vector with the weighted relative entropy of the prime and target inflectional paradigms

NormLevenshteinDist

A numeric vector with the normalized Levenshtein distance of prime and target forms

TargetLemmaFreq

A numeric vector with log frequency of the target lemma

PrimeSurfFreq

A numeric vector with log frequency of the prime form

PrimeCondition

A factor with prime conditions, levels: DD, DSSD, SS

CosineSim

A numeric vector with the cosine similarity of prime and target vector space semantics

IsMasc

A vector of logicals, TRUE if the noun is masculine.

TargetGender

A factor with the gender of the target, levels: f, m, and n

TargetCase

A factor specifying the case of the target noun, levels: acc, dat, nom

MeanLogObsRT

Mean log-transformed observed reaction time

References

Baayen, R. H., Milin, P., Filipovic Durdevic, D., Hendrix, P. and Marelli, M. (2011), An amorphous model for morphological processing in visual comprehension based on naive discriminative learning. Psychological Review, 118, 438-482.

Examples

# calculate the weight matrix for the full set of Serbian nouns
data(serbian)
serbian$Cues <- orthoCoding(serbian$WordForm, grams=2)
serbian$Outcomes <- serbian$LemmaCase
sw <- estimateWeights(cuesOutcomes=serbian)

# calculate the meaning activations for all unique word forms

desiredItems <- unique(serbian["Cues"])
desiredItems$Outcomes <- ""
activations <- estimateActivations(desiredItems, sw)$activationMatrix
rownames(activations) <- unique(serbian[["WordForm"]])
activations <- activations + abs(min(activations))
activations[1:5,1:6]

# calculate simulated latencies for the experimental materials

data(serbianLex)
syntax <- c("acc", "dat", "gen", "ins", "loc", "nom", "Pl", "Sg")
we <- 0.4 # compound cue weight
strengths <- rep(0, nrow(serbianLex))
for(i in 1:nrow(serbianLex)) {
   target <- serbianLex$Target[i]
   prime <- serbianLex$Prime[i]
   targetLemma <- as.character(serbianLex$TargetLemma[i])
   primeLemma <- as.character(serbianLex$PrimeLemma[i])
   targetOutcomes <- c(targetLemma, primeLemma, syntax)
   primeOutcomes <- c(targetLemma, primeLemma, syntax)
   p <- activations[target, targetOutcomes]
   q <- activations[prime, primeOutcomes]
   strengths[i] <- sum((q^we)*(p^(1-we)))
}
serbianLex$SimRT <- -strengths
lengthPenalty <- 0.3
serbianLex$SimRT2 <- serbianLex$SimRT + 
  (lengthPenalty * (serbianLex$Length>5))

cor.test(serbianLex$SimRT, serbianLex$MeanLogObsRT)
cor.test(serbianLex$SimRT2, serbianLex$MeanLogObsRT)

serbianLex.lm <- lm(SimRT2 ~ Length +  WeightedRE*IsMasc + 
      NormLevenshteinDist + TargetLemmaFreq + 
      PrimeSurfFreq + PrimeCondition, data=serbianLex)
summary(serbianLex.lm)

Serbian case inflected nouns (in Cyrillic Unicode).

Description

3240 case-inflected Serbian nouns and their frequencies, for 270 different masculine, feminine and neuter noun lemmas, written using the Cyrillic alphabet and encoded in UTF-8.

Usage

data(serbianUniCyr)

Format

A data frame with 3240 observations on the following 3 variables:

WordForm

A character vector specifying the inflected word forms encoded in UTF-8.

LemmaCase

A character vector specifying lemma (meaning), case, and number.

Frequency

A numeric vector specifying the frequency of each word form.

Details

Frequencies were taken from the Frequency Dictionary of Contemporary Serbian Language (Kostic, 1999). The 270 lemmas comprise the set of nouns for which each different case form appears at least once in this resource.

Source

Kostic, D. (1999). Frekvencijski recnik savremenog srpskog jezika (Frequency Dictionary of Contemporary Serbian Language). Institute for Experimental Phonetics and Speech Pathology & Laboratory of Experimental Psychology, University of Belgrade, Serbia.

References

Kostic, D. (1999). Frekvencijski recnik savremenog srpskog jezika (Frequency Dictionary of Contemporary Serbian Language). Institute for Experimental Phonetics and Speech Pathology & Laboratory of Experimental Psychology, University of Belgrade, Serbia.

Baayen, R. H., Milin, P., Filipovic Durdevic, D., Hendrix, P. and Marelli, M. (2011), An amorphous model for morphological processing in visual comprehension based on naive discriminative learning. Psychological Review, 118, 438-482.

See Also

See also serbian, serbianLex, estimateActivations.

Examples

## Not run: 
data(serbianUniCyr)
serbianUniCyr$Cues <- orthoCoding(serbianUniCyr$WordForm, grams=2)
serbianUniCyr$Outcomes <- serbianUniCyr$LemmaCase
sw <- estimateWeights(cuesOutcomes=serbianUniCyr)
sw[1:5,1:5]
desiredItems <- unique(serbianUniCyr["Cues"])
desiredItems$Outcomes=""
activations <- estimateActivations(desiredItems, sw)$activationMatrix
rownames(activations) <- unique(serbianUniCyr[["WordForm"]])
activations[1:5,1:6]

## End(Not run)

Serbian case inflected nouns (in Latin-alphabet Unicode).

Description

3240 case-inflected Serbian nouns and their frequencies, for 270 different masculine, feminine and neuter noun lemmas, written using the Latin alphabet and encoded in UTF-8.

Usage

data(serbianUniLat)

Format

A data frame with 3240 observations on the following 3 variables:

WordForm

A character vector specifying the inflected word forms encoded in UTF-8.

LemmaCase

A character vector specifying lemma (meaning), case, and number.

Frequency

A numeric vector specifying the frequency of each word form.

Details

Frequencies were taken from the Frequency Dictionary of Contemporary Serbian Language (Kostic, 1999). The 270 lemmas comprise the set of nouns for which each different case form appears at least once in this resource.

Source

Kostic, D. (1999). Frekvencijski recnik savremenog srpskog jezika (Frequency Dictionary of Contemporary Serbian Language). Institute for Experimental Phonetics and Speech Pathology & Laboratory of Experimental Psychology, University of Belgrade, Serbia.

References

Kostic, D. (1999). Frekvencijski recnik savremenog srpskog jezika (Frequency Dictionary of Contemporary Serbian Language). Institute for Experimental Phonetics and Speech Pathology & Laboratory of Experimental Psychology, University of Belgrade, Serbia.

Baayen, R. H., Milin, P., Filipovic Durdevic, D., Hendrix, P. and Marelli, M. (2011), An amorphous model for morphological processing in visual comprehension based on naive discriminative learning. Psychological Review, 118, 438-482.

See Also

See also serbian, serbianLex, estimateActivations.

Examples

data(serbianUniLat)
serbianUniLat$Cues <- orthoCoding(serbianUniLat$WordForm, grams=2)
serbianUniLat$Outcomes <- serbianUniLat$LemmaCase
sw <- estimateWeights(cuesOutcomes=serbianUniLat)
sw[1:5,1:5]
desiredItems <- unique(serbianUniLat["Cues"])
desiredItems$Outcomes=""
activations <- estimateActivations(desiredItems, sw)$activationMatrix
rownames(activations) <- unique(serbianUniLat[["WordForm"]])
activations[1:5,1:6]

A summary of a Naive Discriminatory Learning Model

Description

A summarization method for an object of the class "ndlClassify".

Usage

## S3 method for class 'ndlClassify'
summary(object, ...)

## S3 method for class 'summary.ndlClassify'
print(x, digits = max(3, getOption("digits") - 3), max.print=10, ...)

Arguments

Details

Calculates descriptive statistics of a fitted Naive Discriminatory Learning model and prints a nice summary of the key results.

Value

summary.ndlClassify returns an object of the class "summary.ndlClassify", a list with the following components:

call

The call matched to fit the "ndlClassify" object.

formula

The formula specified for the "ndlClassify" object.

weights

The estimated weights.

statistics

A range of descriptive statistics calculated with ndlStatistics.

Author(s)

Antti Arppe

References

Arppe, A. and Baayen, R. H. (in prep.)

See Also

ndlClassify, ndlStatistics, modelStatistics

Examples

## For examples see examples(ndlClassify).

A summary of a crossvalidation of a Naive Discriminatory Reader Model

Description

A summarization method for an object of the class "ndlCrossvalidate".

Usage

## S3 method for class 'ndlCrossvalidate'
summary(object, ...)

## S3 method for class 'summary.ndlCrossvalidate'
print(x, digits = max(3, getOption("digits") - 3), ...)

Arguments

Details

Calculates overall descriptive statistics of the crossvalidation of a fitted Naive Discriminatory Reader model and prints a nice summary of the key results.

Value

summary.ndlCrossvalidate returns an object of the class "summary.ndlCrossvalidate", a list with the following components:

call

The call matched to fit the "ndlCrossvalidate" object.

formula

The formula specified for the "ndlCrossvalidate" object.

statistics.summary

The means, minima and maxima of a range descriptive statistics for the fit and performance of individual folds; see ndlStatistics.

crosstable.summary

The means of the crosstabulation of observed and predicted outcomes for the held-out test data.

recall.predicted.summary

The means of the recall values for the individual outcomes predicted with the held-out test data.

precision.predicted.summary

The means of the precision values for the individual outcomes predicted with the held-out test data.

statistics.all

All the values for a range descriptive statistics for the fit and performance of individual folds on the held-out test data; see ndlStatistics.

k

The number of folds.

n.total

The sum frequency of all data points in data.

n.train

The sum frequency of data points used for training the individual models (excluding the individual folds).

n.test

The sum frequency of data points in the individual held-out folds used for testing the individual models.

Author(s)

Antti Arppe

References

Arppe, A. and Baayen, R. H. (in prep.)

See Also

ndlCrossvalidate, ndlClassify, ndlStatistics

Examples

## For examples see examples(ndlCrossvalidate).

Finnish ‘think’ verbs.

Description

3404 occurrences of four synonymous Finnish ‘think’ verbs (‘ajatella’: 1492; ‘mietti\"a’: 812; ‘pohtia’: 713; ‘harkita’: 387) in newspaper and Internet newsgroup discussion texts

Usage

data(think)

Format

A data frame with 3404 observations on the following 27 variables:

Lexeme

A factor specifying one of the four ‘think’ verb synonyms

Polarity

A factor specifying whether the ‘think’ verb has negative polarity (Negation) or not (Other)

Voice

A factor specifying whether the ‘think’ verb is in the Passive voice or not (Other)

Mood

A factor specifying whether the ‘think’ verb is in the Indicative or Conditional mood or not (Other)

Person

A factor specifying whether the ‘think’ verb is in the First, Second, Third person or not (None)

Number

A factor specifying whether the ‘think’ verb is in the Plural number or not (Other)

Covert

A factor specifying whether the agent/subject of the ‘think’ verb is explicitly expressed as a syntactic argument (Overt), or only as a morphological feature of the ‘think’ verb (Covert)

ClauseEquivalent

A factor specifying whether the ‘think’ verb is used as a non-finite clause equivalent (ClauseEquivalent) or as a finite verb (FiniteVerbChain)

Agent

A factor specifying the occurrence of Agent/Subject of the ‘think’ verb as either a Human Individual, Human Group, or as absent (None)

Patient

A factor specifying the occurrence of the Patient/Object argument among the semantic or structural subclasses as either an Human Individual or Group (IndividualGroup), Abstraction, Activity, Communication, Event, an ‘etta’ (‘that’) clause (etta_CLAUSE), DirectQuote, IndirectQuestion, Infinitive, Participle, or as absent (None)

Manner

A factor specifying the occurrrence of the Manner argument as any of its subclasses Generic, Negative (sufficiency), Positive (sufficiency), Frame, Agreement (Concur or Disagree), Joint (Alone or Together), or as absent (None)

Time

A factor specifying the occurrence of Time argument (as a moment) as either of its subclasses Definite, Indefinite, or as absent (None)

Modality1

A factor specifying the main semantic subclasses of the entire Verb chain as either indicating Possibility, Necessity, or their absense (None)

Modality2

A factor specifying minor semantic subclasses of the entire Verb chain as indicating either a Temporal element (begin, end, continuation, etc.), External (cause), Volition, Accidental nature of the thinking process, or their absense (None)

Source

A factor specifying the occurrence of a Source argument or its absense (None)

Goal

A factor specifying the occurrence of a Goal argument or its absence (None)

Quantity

A factor specifying the occurrence of a Quantity argument, or its absence (None)

Location

A factor specifying the occurrence of a Location argument, or its absence (None)

Duration

A factor specifying the occurrence of a Duration argument, or its absence (None)

Frequency

A factor specifying the occurrence of a Frequency arument, or its absence (None)

MetaComment

A factor specifying the occurrence of a MetaComment, or its absence (None)

ReasonPurpose

A factor specifying the occurrence of a Reason or Purpose argument (ReasonPurpose), or their absence (None)

Condition

A factor specifying the occurrence of a Condition argument, or its absence (None)

CoordinatedVerb

A factor specifying the occurrence of a Coordinated Verb (in relation to the ‘think’ verb: CoordinatedVerb), or its absence (None)

Register

A factor specifying whether the ‘think’ verb occurs in the newspaper subcorpus (hs95) or the Internet newsgroup discussion corpus (sfnet)

Section

A factor specifying the subsection in which the ‘think’ verb occurs in either of the two subcorpora

Author

A factor specifying the author of the text in which the ‘think’ verb occurs, if that author is identifiable – authors in the Internet newgroup discussion subcorpus are anonymized; unidentifiable/unknown author designated as (None)

Details

The four most frequent synonyms meaning ‘think, reflect, ponder, consider’, i.e. ‘ajatella, miettia, pohtia, harkita’, were extracted from two months of newspaper text from the 1990s (Helsingin Sanomat 1995) and six months of Internet newsgroup discussion from the early 2000s (SFNET 2002-2003), namely regarding (personal) relationships (sfnet.keskustelu.ihmissuhteet) and politics (sfnet.keskustelu.politiikka). The newspaper corpus consisted of 3,304,512 words of body text (i.e. excluding headers and captions as well as punctuation tokens), and included 1,750 examples of the studied ‘think’ verbs. The Internet corpus comprised 1,174,693 words of body text, yielding 1,654 instances of the selected ‘think’ verbs. In terms of distinct identifiable authors, the newspaper sub-corpus was the product of just over 500 journalists and other contributors, while the Internet sub-corpus involved well over 1000 discussants. The think dataset contains a selection of 26 contextual features judged as most informative.

For extensive details of the data and its linguistic and statistical analysis, see Arppe (2008). For the full selection of contextual features, see the amph (2008) microcorpus.

Source

amph 2008. A micro-corpus of 3404 occurrences of the four most common Finnish THINK lexemes, ‘ajatella, miettia, pohtia, and harkita’, in Finnish newspaper and Internet newsgroup discussion texts, containing extracts and linguistic analysis of the relevant context in the original corpus data, scripts for processing this data, R functions for its statistical analysis, as well as a comprehensive set of ensuing results as R data tables. Compiled and analyzed by Antti Arppe. Available on-line at URL: http://www.csc.fi/english/research/software/amph/

Helsingin Sanomat 1995. ~22 million words of Finnish newspaper articles published in Helsingin Sanomat during January–December 1995. Compiled by the Research Institute for the Languages of Finland [KOTUS] and CSC – IT Center for Science, Finland. Available on-line at URL: http://www.csc.fi/kielipankki/

SFNET 2002-2003. ~100 million words of Finnish internet newsgroup discussion posted during October 2002 – April 2003. Compiled by Tuuli Tuominen and Panu Kalliokoski, Computing Centre, University of Helsinki, and Antti Arppe, Department of General Linguistics, University of Helsinki, and CSC – IT Center for Science, Finland. Available on-line at URL: http://www.csc.fi/kielipankki/

References

Arppe, A. 2008. Univariate, bivariate and multivariate methods in corpus-based lexicography – a study of synonymy. Publications of the Department of General Linguistics, University of Helsinki, No. 44. URN: http://urn.fi/URN:ISBN:978-952-10-5175-3.

Arppe, A. 2009. Linguistic choices vs. probabilities – how much and what can linguistic theory explain? In: Featherston, Sam & Winkler, Susanne (eds.) The Fruits of Empirical Linguistics. Volume 1: Process. Berlin: de Gruyter, pp. 1-24.

Examples

## Not run: 
data(think)
think.ndl = ndlClassify(Lexeme ~ Person + Number + Agent + Patient + Register,
   data=think)
summary(think.ndl)
plot(think.ndl)

## End(Not run)