Professor of Phonetic Sciences   University of Amsterdam   E-mail: paul.boersma@uva.nl Telephone: +31–20–5253864 (secretary).

Software:  PRAAT: doing phonetics by computer

Research:  all my writings in chronological order

My research focuses on showing (by computer simulations) how the production, comprehension and acquisition of the phonetics, phonology and morphology of a language, as well as how these change over the generations, can be explained by assuming multi-level representations in the language user’s mind.

My most recent papers in this area attempt to do this by using artificial neural networks, in which phonological categories emerge from the phonetic data, sometimes in combination with semantic data:

Earlier papers attempt do this by using Optimality Theory, which still has the advantage of being able to work with less toy-like languages. The following paper shows how in a parallel multi-level model of production phonological (i.e. “later”) considerations can influence morphological (i.e. “earlier”) choices:

Multi-level models allow us to regard the typologies of the world’s languages as emergent rather than as innate or synchronically functionalist. For example, markedness is an emergent property that follows from frequency differences in the learner’s input (and not from innate markedness constraints or from synchronically functionalist faithfulness rankings), and licensing by cue emerges from differences in auditory cue reliability in the learner’s input (and not from innate specific-over-general faithfulness rankings or from synchronically listener-oriented faithfulness rankings):

As another example, auditory dispersion in inventories of phonemes emerges from the fact that learners use in production the same constraint rankings that have optimized their comprehension (and not from innate markedness constraints or synchronically functionalist dispersion constraints). The following two papers are the one- and two-dimensional cases, respectively:

As for the emergence of categories and constraints themselves, that is discussed for Optimality Theory as well as neural networks:

Such simulations make it possible to track languages over the generations (for more, see sound change):

The above papers (if younger than 2005) rely heavily on the framework of Parallel Bidirectional Phonology and Phonetics (BiPhon), i.e. on the idea that you use the same constraint ranking as a listener and as a speaker and on the parallel multi-level evaluation of your phonology and your phonetics. Here is more information on that subject:

Most of the papers with simulations utilize the Gradual Learning Algorithm for Optimality Theory, which was defined in the following two papers:

Since 2007 we have routinely checked how the simulations behave if we use Harmonic Grammar instead of Optimality Theory. The following paper describes a proof of the learning algorithm:

Writings by subject:

• Optimality-Theoretic (OT) and neural-network (NN) modelling of bidirectional phonology and phonetics and their acquisition and evolution (1989–2021):
  • sound change (1989–2020), partly with Silke Hamann, Joe Pater and Klaas Seinhorst
  • Gradual Learning Algorithm and other OT and HG learning algorithms (1997–2016), partly with Bruce Hayes, Clara Levelt and Joe Pater
  • linguistic processes:
    • prelexical perception:
      • categorization (1997–2022), partly with Paola Escudero, Rachel Hayes, Titia Benders and Kateřina Chládková
      • OCP (1998–2003)
    • lexicon (word recognition, faithfulness ranking, lexical selection) (1999–2009), partly with Silke Hamann and Diana Apoussidou
    • production:
      • framework listener-oriented production (control loops, probabilistic faithfulness) (1997–2005), partly with Silke Hamann
      • framework Parallel Bidirectional Phonology and Phonetics (BiPhon-OT) (2005–2017), partly with Diana Apoussidou, Silke Hamann and Jan-Willem van Leussen
      • framework Parallel Bidirectional Phonology and Phonetics (BiPhon-NN) (2019–2022), with Titia Benders, Kateřina Chládková and Klaas Seinhorst
  • representations:
    • features with Kateřina Chládková, Titia Benders and Mirjam de Jonge (2011–2022)
  • paralinguistic tasks:
    • grammaticality and prototypicality (2001–2008), partly with Bruce Hayes and Silke Hamann
  • applications:
    • loanword adaptation (2000–2009), partly with Silke Hamann
    • Poverty of the Base (1997–2009)
    • nasal harmony (1998–2003)
    • metrical phonology with Diana Apoussidou (2003–2004)
• Experimental phonetics and phonology (1993–2019):
  • writings on the Praat program (1993–2014), partly with David Weenink and Ton Wempe
  • methodology (2005–2013), partly with Paola Escudero, Titia Benders and Kateřina Chládková
  • distributional learning with Karin Wanrooij (2013–2015): continuous distributions, two-month-old infants, adults, confounds
  • statistical learning with Sophie ter Schure (2016): multimodal, speech–object coupling with simulations
  • the undersampling hypothesis with Kateřina Chládková (2009)
  • Portuguese, Spanish and Swedish vowels (2009–2019), with Paola Escudero, Andréia Rauber, Kateřina Chládková, Ricardo Bion and Joppe Pelzer
  • Croatian folk singing with Gordana Kovacic (2003–2006)
• Limburgian tone (2002–2018):
  • Franconian tonogenesis (2002–2018)
  • synchronic analysis of Roermond (2011)
  • pitch accent versus tone with Björn Köhnlein (2008)

All writings in chronological order

Talks and posters