Analyzing free variation with harmony – A case study of verb-cluster serialization

In German, a verb selected by another verb normally precedes the selecting verb. Modal verbs in the perfect tense provide an exception to this generalization because they require the perfective auxiliary to occur in cluster-initial position according to prescriptive grammars. Bader and Schmid (2009b) have shown, however, that native speakers accept the auxiliary in all positions except the cluster-final one. Experimental results as well as corpus data indicate that verb cluster serialization is a case of free variation. I discuss how this variation can be accounted for, focusing on two mismatches between acceptability and frequency: First, slight acceptability advantages can turn into strong frequency advantages. Second, syntactic variants with basically zero frequency can still vary substantially in acceptability. These mismatches remain unaccounted for if acceptability is related to frequency on the level of whole sentence structures, as in Stochastic OT (Boersma and Hayes2001). However, when the acceptability-frequency relationship is modeled on the level of individual weighted constraints, using harmony as link (see Pater2009, for different harmony based frameworks), the two mismatches follow given appropriate linking assumptions.

Learning a gradient grammar of French liaison

In certain French words, an orthgraphically-final consonant is unpronounced except, in certain environments, when it precedes a vowel. This phenomenon, liaison, shows significant interactions with several other patterns in French (including h-aspiré, schwa deletion, and the presence of other morphemes in the liaison context). We present a learning algorithm that acquires a grammar that accounts for these patterns and their interactions. The learned grammar employs Gradient Symbolic Computation (GSC), incorporating weighted constraints and partially-activated symbolic representations. Grammatical analysis in the GSC framework includes the challenging determination of the numerical strength of symbolic constituent activations (as well as constraints). Here we present the first general algorithm for learning these quantities from empirical examples: the Error-Driven Gradient Activation Readjustment (EDGAR). Smolensky and Goldrick (2016) proposed a GSC analysis, with hand-determined numerical strengths, in which liaison derives from the coalescence of partially-activated input consonants. EDGAR allows us to extend this work to a wider range of liaison phenomena by automatically determining the more comprehensive set of numerical strengths required to generate the complex pattern of overall liaison behaviour.

Optimized Graph cut Color Image Segmentation using Genetic Algorithm with Weighted Constraints (OGcut)

This paper proposes a Novel color image segmentation using Graph cut method by minimizing the weighted energy function. This method is applying a pair of optimal constraints namely: color constraint and gradient constraint. In the state-of-the-art methods, the background and foreground details are manually initialized and used for verifying the smoothness of the region. But in this proposed method, they are dynamically calculated from the input image. This feature of the proposed method can be used in color image segmentation where more number of unique segments exists in a single image. The genetic algorithm is applied to the graph obtained from the graph cut method. The crossover and mutation operators are applied on various subgraphs to populate the different segments.

English adaptation in Mandarin A-not-A constructions

A-not-A refers to a Mandarin reduplication construction where the underlying form /RED-pu-A/ contains a reduplication of the first syllable in A. In this study I investigate the kinds of adaptations that occur when an English word serves as the base A in code-switching speech. Since the complex onsets and most codas allowed in English are illegal in Mandarin syllables, the reduplicated part is expected to adapt to Mandarin phonotactics to some degree. I ran a production experiment where 20 native Mandarin-speakers were asked to produce A-not-A constructions with 55 mono- and multi-syllabic English words. Results from the experiment showed varied adaptation methods in syllable structure and tones. To model the results, I used the Maximum Entropy Harmonic Grammar (MaxEnt) with weighted constraints on syllable structure markedness and base-reduplicant faithfulness.

Dative intervention is a gang effect

This paper addresses two restrictions regarding agreement with nominative arguments in Icelandic DAT-NOM constructions. The first is the reported asymmetry in intervention effects in mono-clausal versus bi-clausal environments. The second regards the well-known Person Restriction that prohibits agreement with non-3rd person arguments. It is argued that both of these phenomena can be viewed as instances of cumulative constraint interaction, where less important constraints in the grammar ‘gang up’ to block some higher constraint. In order to account for this, I adopt a model of syntax with both weighted constraints and serial optimization that is known as Serial Harmonic Grammar in the phonological literature. It will be demonstrated that such a system can offer a more principled analysis of the construction-specific nature of the aforementioned phenomena.

Predicting semi-regular patterns in morphologically complex words

We expect generative models of language to correctly predict surface forms from underlying forms, but morphologically complex words, especially compounds, can exhibit idiosyncratic outputs, which require an extra lexical listing. This results in (a) a poorer Minimum Description Length of our model and (b) failure of a grammar to capture patterning among exceptions. To solve an instance of this problem, we examine pitch-accent patterns of 2-mora-2-mora Japanese Yamato (native) noun-noun compounds, hitherto considered semi-predictable but which show gradient tendencies among constituents to trigger a particular accent pattern. In the framework of Gradient Symbolic Computation (Smolensky, Paul & Matthew Goldrick. 2015. Gradient symbolic computation. LSA Summer Institute Workshop. Chicago.), a type of harmonic grammar which allows partially activated feature values and weighted constraints, such gradient patterns can be captured through the additive combination of coalescing features on each conjunct, which results in a pitch accent when the summed activations surpass a threshold determined by the grammar. The ability of this framework to completely predict these semi-regular patterns holds promise that it can also explain similar kinds of patterns in other languages.

When accent preservation leads to clash

In English, some complex words can display exceptional accent preservation (EAP): they can preserve an accent from their base even when this would violate a general restriction against adjacent accents (e.g. retúrn → retùrnée). This article analyses EAP both empirically and theoretically. The analysis of a set of 291 derivatives from Wells (2008) shows that this phenomenon can be partially attributed to the relative frequency of the base and its derivative and partially also to syllable structure, and that these two factors have a cumulative effect. It is also shown that the existence of a more deeply embedded base (e.g. colléct→ colléctive → còllectívity ~ collèctívity) can increase the likelihood for a derivative to display EAP. A formal account of the phenomenon is proposed building on Collie's (2007, 2008) ‘fake cyclicity’ analysis, using weighted constraints (Pater 2009, 2016) and Max-Ent-OT (Goldwater & Johnson 2003). Finally, a model of lexical access building on Hay's (2001, 2003) model and integrating more deeply embedded bases is proposed.