Papers, etc.
Comments very welcome!
Biases in Harmonic Grammar: the road to restrictive learning.
Karen Jesney & Anne-Michelle Tessier. to appear. Natural Language and Linguistic Theory. (summary)
Harmonic Grammar with linear programming: from linear systems to linguistic typology.
Christopher Potts, Joe Pater, Karen Jesney, Rajesh Bhatt & Michael Becker. to appear. Phonology. (ROA-984 | summary)
Learning distributions over underlying representations.
Karen Jesney, Joe Pater & Robert Staubs. 2009. Paper presented at the Northeast Computational Phonology Circle (NECPhon 3). Cambridge, MA: MIT. October 2009. (handout | summary)
Uniformity effects as a consequence of learning with lexical constraints.
Karen Jesney. 2009. Poster presented at the KNAW Colloquium on Language Acquisition and Optimality Theory. Amsterdam. July 2009. (handout | summary)
Licensing in multiple contexts: an argument for Harmonic Grammar.
Karen Jesney. 2009/to appear. In Proceedings of the 45th Meeting of the Chicago Linguistic Society (CLS 45). Chicago, IL: University of Chicago. (paper | conference handout | summary)
Positional Faithfulness, non-locality, and the Harmonic Serialism solution.
Karen Jesney. 2008/to appear. In Suzi Lima, Kevin Mullin & Brian Smith (eds.), Proceedings of the 39th Meeting of the North East Linguistics Society (NELS 39). Amherst, MA: GLSA. (paper ROA-1018 | conference handout | summary)
Gradual learning and faithfulness: consequences of ranked vs. weighted constraints.
Karen Jesney & Anne-Michelle Tessier. 2009. In Anisa Schardl, Martin Walkow & Muhammad Abdurrahman (eds.), Proceedings of the 38th Meeting of the North East Linguistics Society (NELS 38), volume 1, 375-388. Amherst, MA: GLSA. (paper ROA-967 | 2007 conference handout | summary)
Restrictiveness in gradual learning of Harmonic Grammar.
Karen Jesney & Anne-Michelle Tessier. 2007. Paper presented at the North East Computational Phonology Circle (NECPhon 1). Amherst, MA: University of Massachusetts. November 2007. (handout | summary)
Re-evaluating learning biases in Harmonic Grammar.
Karen Jesney & Anne-Michelle Tessier. 2007. In Michael Becker (ed.), University of Massachusetts Occasional Papers 36: Papers in Theoretical and Computational Phonology. Amherst, MA: GLSA. (paper | summary)
The locus of variation in weighted constraint grammars.
Karen Jesney. 2007. Poster presented at the Workshop on Variation, Gradience and Frequency in Phonology. Stanford, CA: Stanford University. July 2007. (handout | MaxEnt calculator | summary)
Phonological acquisition as weighted constraint interaction.
Joe Pater, Karen Jesney & Anne-Michelle Tessier. 2007. In Alyona Belikova, Luisa Meroni & Mari Umeda (eds.), Proceedings of the 2nd Conference on Generative Approaches to Language Acquisition – North America (GALANA 2), 339-350. Somerville, MA: Cascadilla Proceedings Project. (proceedings | erratum | summary)
Child chain shifts as faithfulness to input prominence.
Karen Jesney. 2007. In Alyona Belikova, Luisa Meroni & Mari Umeda (eds.), Proceedings of the 2nd Conference on Generative Approaches to Language Acquisition – North America (GALANA 2), 188-199. Somerville, MA: Cascadilla Proceedings Project. (proceedings | 2006 conference handout | summary)
Chain Shift in Phonological Acquisition.
Karen Jesney. 2005. MA thesis. Calgary, AB: University of Calgary. (thesis | summary)
Resolving hierarchy conflict: local obviation in Blackfoot.
Heather Bliss & Karen Jesney. 2005. Calgary Papers in Linguistics, 26: 92-116. (download here | summary)
Stridency and differential substitution in two dialects of French.
Karen Jesney. 2005. Paper presented at the Language Research Centre Graduate Student Forum. Calgary, AB: University of Calgary.
(handout | summary)
Dialect, stridency and differential substitution.
Karen Jesney. 2004. Poster presented at the 1st Conference on Generative Approaches to Language Acquisition – North America (GALANA 1). Honolulu, HI: University of Hawai'i. December 2004. (summary)
The Use of Global Foreign Accent Rating in Studies of L2 Acquisition.
Karen Jesney. 2004. Calgary, AB: University of Calgary Language Research Centre Reports. (download here | summary)
A Review of the Literature on Second Language Learning.
John Archibald, Sylvie Roy, Sandra Harmel & Karen Jesney. 2004. Calgary, AB: Calgary Language Research Centre Reports. (download here | summary)
Summaries
Biases in Harmonic Grammar: the road to restrictive learning. [to appear, with Anne-Michelle Tessier]
In the Optimality-Theoretic learnability and acquisition literature it has been proposed that certain classes of constraints must be biased toward particular rankings (e.g., Markedness >> IO-Faithfulness; Specific IO-Faithfulness >> General IO-Faithfulness). While sometimes difficult to implement efficiently or comprehensively, these biases are necessary to explain how learners acquire the most restrictive grammar consistent with positive evidence from the target language, and how innovative patterns emerge during the course of child phonological development. This paper demonstrates that altering the mode of constraint interaction from strict ranking as in Optimality Theory to additive weighting as in Harmonic Grammar (HG) reduces the number of classes of constraints that must be distinguished by such biases. Using weighted constraints and a version of the Gradual Learning Algorithm (GLA), the only distinction needed is between Output-based constraints, which must be biased toward high weights, and Input- Output-based constraints, which must be biased toward the lowest weights possible. We implement this distinction within the HG-GLA model by assigning different initial weights and plasticity values to the two classes of constraints. This implementation suffices to ensure that restrictive grammars are learned, and also predicts the emergence of a variety of attested intermediate stages during the course of acquisition. [This paper revises and extends our 2007 UMOP 36 paper: Reevaluating learning biases in Harmonic Grammar.]
Harmonic Grammar with linear programming: from linear systems to linguistic typology. [to appear, with Christopher Potts, Joe Pater, Rajesh Bhatt & Michael Becker] (ROA-984)
Harmonic Grammar (HG) is a model of linguistic constraint interaction in which well-formedness is based on the sum of weighted constraint violations. We show how linear programming algorithms can be used to determine whether there is a weighting for a set of constraints that fits a set of linguistic data. The associated software package OT-Help provides a practical tool for studying large and complex linguistic systems in the HG framework and comparing the results with those of OT. We first describe the translation from Harmonic Grammars to systems solvable by linear programming algorithms. We then develop an HG analysis of ATR harmony in Lango that is, we argue, superior to the existing OT and rule-based treatments. We further highlight the usefulness of OT-Help, and the analytic power of HG, with a set of studies of the predictions HG makes for phonological typology.
Learning distributions over underlying representations. [2009, with Joe Pater & Robert Staubs] (handout)
This paper argues that if measures are in place to ensure restrictiveness (e.g., a Markedness > Faithfulness bias), a distinction may not need to be drawn between learning allomorphy and learning “regular” underlying representations. Rather, in both cases a range of URs corresponding to the set of observed surface allomorphs may exist; a grammar that includes weighted lexical constraints (Boersma 1999, Apoussidou 2006) can then select from among the URs on each iteration of Eval. Given a model of grammar that yields variation, this approach extends to patterns that cannot modeled by relying on abstract URs. Using simulations, we demonstrate that a batch learner incorporating these assumptions achieves considerable success in matching the training data and extends in a restrictive fashion to unobserved types of input forms.
Uniformity effects as a consequence of learning with lexical constraints. [2009] (handout)
Output-Output Faithfulness constraints (Benua 1997) have been used in the OT acquisition literature to model patterns where a child allows marked output structures to appear just when this enables uniformity of stem exponence to be maintained (see, e.g., Hayes 2004, Tessier 2007). Drawing on novel data from the speech of Trevor (Compton & Streeter 1977, Pater 1997), I show that this approach is not adequate. Markedness violations can be driven by affix uniformity as well. I thus propose that such uniformity patterns are instead due to lexical constraints which require that a given meaning be associated with a particular underlying form (e.g., Apoussidou 2007, Boersma 1999). Error-driven learning simulations show that the probability of emergent uniformity effects in this model is affected by the distribution of the input data.
Licensing in multiple contexts: an argument for Harmonic Grammar. [2009] (paper | conference handout)
Positional licensing constraints in Optimality Theory predict that it should not be possible for a language to independently license a given feature in multiple contexts – e.g., in onsets and in initial syllables. Basic positional faithfulness constraints in OT allow licensing in multiple contexts, but preclude licensing effects that require double privilege – e.g., cases where licensing is limited to the onset of an initial syllable. Evidence from natural language shows that both of these configurations occur. In weighted constraint models such as Harmonic Grammar (Smolensky & Legendre 2006) positional licensing constraints can capture both of these patterns, with the result that positional faithfulness constraints are not needed to model a range of attested asymmetries. A more limited and basic constraint set may thus be possible if constraints are weighted rather than ranked. This leads both to less redundancy in the theory and to more restrictive typological predictions. [This work was also presented at the Spring 2009 MIT-UMass Meeting in Phonology (MUMM 3).]
Positional Faithfulness, non-locality, and the Harmonic Serialism solution. [2008/to appear] (paper ROA-1018 | conference handout)
Pathologies arise in parallel OT when a positional faithfulness constraint and a conflicting markedness constraint dominate the constraints responsible for determining which segments, if any, occupy the privileged position. Under such rankings, an underlying featural contrast can be displaced onto the output prosodic structure. (The possibility of such effects was first noted in Beckman 1998:37fn.) Here I show that the problem is general one in parallel OT, and that the resulting pathological patterns are opaque and can be highly non-local. The solution in Harmonic Serialism (McCarthy 2006, 2007, 2008) comes not from altering the ranking, but from reinterpreting the basis upon which privilege is established. Rather than being based on the prosodification in the output candidate, privilege is argued here to stem from the prosodic structure associated with the input to the current step in the derivation. This approach allows the desired effects of positional faithfulness to be seen, without also predicting opaque and non-local patterns that fall outside the range of what is attested in human language.
Gradual learning and faithfulness: consequences of ranked vs. weighted constraints. [2009, with Anne-Michelle Tessier] (paper ROA-967 | 2007 conference handout)
This paper investigates a class of stages in L1 phonological acquisition where children faithfully produce marked structures only in privileged positions. We present one such stage, referred here to here as an Intermediate Faith (IF) stage, using data from the acquisition of Hebrew reported by Bat-El (2007). The privileged domain in this case is defined morphologically (noun vs. non-noun). We then show how a gradual, on-line learner using weighted constraints as in Harmonic Grammar naturally passes through IF stages, which we model as gang effects between general and specific Faithfulness constraints. Finally, we compare the performance of a ranked constraint learner to that of the HG system developed here. The issues of restrictiveness and biases in Harmonic Grammar learning are further developed in our paper Biases in Harmonic Grammar: the road to restrictive learning (above).
Restrictiveness in gradual learning of Harmonic Grammar. [2007, with Anne-Michelle Tessier] (handout)
This presentation argues that restrictive learning in Harmonic Grammar requires one overarching bias – Output-oriented constraints (Markedness and OO-Faith) above Input-Output-oriented constraints (IO-Faith). Output-oriented constraints can begin with any high weight, but IO-Faithfulness constraints must begin with a weight of zero. Furthermore, the plasticity of IO-Faith constraints must be less than that of Output-oriented constraints, so that their values move away from zero at a relatively slow pace, favouring low values.
Re-evaluating learning biases in Harmonic Grammar. [2007, with Anne-Michelle Tessier, revised and extended as: Biases in Harmonic Grammar: the road to restrictive learning (above)] (draft)
In the Optimality Theoretic learnability and acquisition literature it has been frequently argued that certain ranking biases are necessary – specifically, Markedness >> Faithfulness, Specific Faithfulness >> General Faithfulness, OO-Faithfulness >> Markedness (e.g., Demuth 1995, Gnanadesikan 2004, Hayes 2004, McCarthy 1998, Prince & Tesar 2004, Smith 2000, Smolensky 1996, Tessier 2007). In this paper we show that altering the mode of constraint interaction from strict ranking as in Optimality Theory to additive weighting as in Harmonic Grammar substantially reduces the number of biases needed in order to ensure restrictive learning. Using weighted constraints and a simple GLA-style learning algorithm, a single bias such that wOutput-based constraints > wInput-Output-based constraints is sufficient. This bias is incorporated into the initial state by assigning substantially greater importance (i.e., higher values) to Markedness and Output-Output Faithfulness constraints than to Input-Output Faithfulness constraints, and by requiring that the weights of Output-based constraints be adjusted more quickly than those of Input-Output-based constraints.
The locus of variation in weighted constraint grammars. [2007] (handout)
This poster contrasts the predictions of two systems for selecting optima and incorporating variation in weighted constraint grammars – “Noisy HG” and “MaxEnt/OT”. Noisy HG constraint interaction is linear and additive as in standard Harmonic Grammar (Legendre, Miyata & Smolensky 1990, Smolensky & Legendre 2006), and variation is implemented by adding noise to the evaluation (per Boersma 1998; in HG, see Boersma & Pater 2007, Pater, Bhatt & Potts 2007). This system generally prevents harmonically-bounded candidates from emerging, and predicts that the likelihoods of variable processes will be interdependent if they share relevant constraints. MaxEnt/OT constraint interaction is log-linear, with candidate probabilities being computed directly by the grammar based on a single set of constraint values (Goldwater & Johnson 2003, Jäger to appear, Jäger & Rosenbach 2006). This system allows harmonically-bounded candidates to emerge, and predicts that the likelihoods of variable processes will always be independent of one another, even when they share relevant constraints. There is a simple Excel-based MaxEnt calculator for syllable structure typology available here.
Phonological acquisition as weighted constraint interaction. [2007, with Joe Pater & Anne-Michelle Tessier] (proceedings | erratum)
A theory of learnability that accounts for the human acquisition process should both converge on the correct final grammar, and model the path that learners take to get there. The Gradual Learning Algorithm (GLA; Boersma 1998, Boersma & Hayes 2001) can model some, but not all aspects of the learning path, and as we will show, is non-convergent. The Constraint Demotion Algorithm (Tesar & Smolensky 1998) is convergent, but non-gradual. In this paper we argue that the search for a gradual convergent learner may be aided by replacing Optimality Theory’s constraint ranking with numerical weighting, returning in this respect to OT's predecessor Harmonic Grammar (HG; Legendre et al. 1990, Smolensky & Legendre 2006).We demonstrate the advantages of weighting by using a minimally modified version of the GLA implemented by Boersma & Weenink (2006) that learns Harmonic Grammars, rather than OT grammars.
Child chain shifts as faithfulness to input prominence. [2007] (proceedings | 2006 conference handout)
Chain Shift in Phonological Acquisition. [2005] (thesis)
Chain shift scenarios spontaneously arise and then subside in the developing phonological systems of both first and second language learners. I propose that these patterns are a consequence of specific Ident constraints which reference prominent input feature combinations (where prominence is defined in terms of perceptual reinforcement). These specific faithfulness constraints are biased toward high ranking, as has been proposed for positional faithfulness constraints more generally (e.g., Hayes 2004, Tessier 2007, Smith 2000). In order for chain shifts to arise and eventually dissipate, then, the constraints responsible for the non-target-like processes involved in the chain shifts must simply be (gradually) reranked relative to the specific and general faithfulness constraints. Developmental chain shifts are thus restrictive intermediate stages that emerge in acquisition without the learner needing to postulate constraints or representations for which the target language lacks positive evidence. This enables predictions to be made about the types developmental paths that language acquirers can follow and places limits on the range of possible chain shift scenarios. My MA thesis considers L2 as well as L1 chain shifts, and provides a critique of previous approaches. (This work was also presented at the 2006 UMass-Brown Workshop on Phonological Acquisition.)
Resolving hierarchy conflict: local obviation in Blackfoot. [2005, with Heather BIiss] (download here)
This paper suggests a novel analysis of the apparent failure of Blackfoot (Algonquian: Western North America) to follow the Universal Animacy Hierarchy in some aspects of its grammar. In particular, we argue that the apparent [2>1] patterns of Blackfoot are a consequence of the interaction of person and obviation marking and propose extensions to Harley and Ritter's (2002) morphosyntactic feature geometry to include obviation features. If you are interested in knowing more about Blackfoot, check out the following links: University of Calgary Blackfoot Language Page, Donald Frantz's Blackfoot Language Page, Ethnologue.
Stridency and differential substitution in two dialects of French. [2005] (handout)
Dialect, stridency and differential substitution. [2004]
The question of what causes speakers of different L1s to select particular segments as substitutes for novel L2 segments has intrigued me for some time. This was one version of my thinking of the topic, taking the [s] for [theta] European French pattern vs. [t] for [theta] Québécois pattern as a case study. The proposal builds on the idea that speakers of different L1 dialects may have different sets of active features available for ready transfer to the L2. Preliminary pilot study results are included in the linked handout. (This work was also presented at the 2004 Alberta Conference on Linguistics.)
The Use of Global Foreign Accent Rating in Studies of L2 Acquisition. [2004] (download here)
This report reviews a range of studies that have employed global foreign accent rating as a means of gaining insight into the process of second language acquisition. Sections of the report focus upon methodology (rating scales, tokens, speaker and listener profiles), experimental findings (particularly with respect to speaker background and phonetic factors), and the details of individual studies. There has been recent related discussion of this issue on LinguistList.
A Review of the Literature on Second Language Learning. [2004, with John Archibald, Sylvie Roy & Sandra Harmel] (download here)
This report was prepared for Alberta Learning. Topics include: 1) effects of second language acquisition on the first language, 2) the role of content instruction in the teaching of a second language (content-based language teaching), 3) the effect of L2 instruction on students with special needs, 4) the effects of learning a third language on students for whom English is a second language.