Introduction to the Naive Discriminative Learning Package

Organisation
Fabian Tomaschek (University of Tübingen, fabian.tomaschek@uni-tuebingen.de)

Workshop time and place
The workshop will be held in Cologne on June 14, 2017, as part of the Phonetics and Phonology in Europe (PaPE) conference.

For whom is this workshop
Due to its implementation, NDL is capable to learn many-to-one as well as one-to-many relations. This enables all researches who investigate, for example, perception and production of synonyms and homonyms, of idioms, of phonetic variation, morphological mapping and processing.
NDL enables you  to create computational learning models and test their predictions. NDL has proven to be predictive for effects in simple naming tasks, priming tasks and lexical decision tasks.

Computational models offer today’s linguists a possibility to formalize their theoretical assumptions and draw precise predictions about the speaker’s and listener’s linguistic behavior in experiments such as lexical response times or identification rates and in corpora such as phonetic durations or formant values. Based on learning algorithms tested repeatedly in animal learning behavior  (Rescorla & Wagner, 1972),  the Naive Discriminative Learning Package for R (NDL) represents such a possibility.

Workshop description
After having introduced the theoretical background on NDL, the workshop will allow participants to construct event-based dataframes in order to model cue-to-outcome association in a hands-on manner. Furthemore, techniques will be shown how to predict linguistic behavior such as response times, phonetic durations or lexical categorization.

How NDL works
NDL calculates association strengths between an input layer and an output layer, the former representing cues and the latter outcomes of e.g. perceptual processes. Crucially, like speakers, NDL is sensitive to the statistical properties of words (e.g. Aylett & Turk, 2004; Jurafsky et al., 2000). Different learning environments as well as different levels of speech production or perception can be constructed by different combinations of cues and outcomes. I.e. input units can be represented among others by letters or phones, output units can contain word forms, lexemes or grammatical categories. This allows us to create different models of linguistic processing.

Modeling learning can be based on two different input sources: Either individual events such as corpora of transcribed spoken language, e.g. Buckeye, Kiel, etc.; or corpora of linguistic forms containing counts, e.g. CELEX. So far, NDL has been used to predict response times in lexical decision tasks, neural behavior, phonetic productions, dialectal distances etc. (e.g. Baayen et. al., 2011; Wieling et al., 2014; Augurzky et al., 2014).

Requirements and Participants
The number of participants is limited to 30. Submission is based on a “first come first attend” basis. Since this is not a programming course, participants should have programming abilities as well as knowledge of statistical modeling in R (knowledge of dataframes, accessing and indexing variables, usage of for-loops, etc.). Participants interested in an introduction to R are refered to my introduction to R (http://fabian-tomaschek.com/useful-stuff/). Participants are encouraged to prepare their own experimental and corpus data.

Registration
Registration via email to fabian.tomaschek@uni-tuebingen.de

References
Augurzky, P; Riester, A., Tomaschek, F. (2014). Segmental effects on prosody: Moddeling German argument structure. In: Leemann, A. et al. (eds.) Trends in Phonetics and Phonology. Studies from German speaking Europe. Frankfurt am Main / Bern: Lang.

Aylett, M. & Turk, A. (2004): The Smooth Signal Redundancy Hypothesis: A Functional Explanation for Relationships between Redundancy, Prosodic Prominence, and Duration in Spontaneous Speech. Language and Speech, 47 (1), p.31-56.

Baayen H. et al (2011). An amorphous model for morphological processing in vusal comprehension based on naoive discriminative learning. Psychological Review 118, 438-482.

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

Jurafski, Daniel, Alan Bell, Michelle Gregory  and William D. Raymond (2000): Probabilistic Relations between Words: Evidence from Reduction in Lexical Production. In: Bybee, Joan and Paul Hopper (eds.). Frequency and the emergence of linguistic structure. Amsterdam: John Benjamins

Ramscar Michael et al (2010): The Effects of Feature-Label-Order and Their Implications
for Symbolic Learning

Ramscar, Michael, Melody Dye, and Stewart McCauley. (2013) Error and expectation in language learning: The curious absence of ‘mouses’ in adult speech. Language, 89(4), 760-793.

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.

Wieling, Martijn, Simonetta Montemagni, John Nerbonne and R. Harald Baayen (2014). Lexical differences between Tuscan dialects and standard Italian: Accounting for geographic and socio-demographic variation using generalized additive mixed modeling. Language, 90(3), 669-692.