Rule-Based Phonology

We discuss a set-theoretic treatment of segments as sets of valued features and of natural classes as intensionally defined sets of sets of valued features. In this system, the empty set { } corresponds to a completely underspecified segment, and the natural class [ ] corresponds to the set of all segments, making a feature ± Segment unnecessary. We use unification, a partial operation on sets, to implement feature-filling processes, and we combine unification with set subtraction to implement feature-changing processes. We show how unification creates the illusion of targeting only underspecified segments, and we explore the possibility that only unification rules whose structural changes involve a single feature are UG-compatible. We show that no such Singleton Set Restriction can work with rules based on set subtraction. The system is illustrated using toy vowel harmony systems and a treatment of compensatory lengthening as total assimilation.

Sažetak: Suvremeni istraživa ki interesi koji su usmjereni na razvoj itanja i prate e teško e pokušavaju odrediti razinu zna aja razli itih kognitivnih procesa kao prediktora teško a itanja. Brojna istraživanja izdvajaju upravo fonološke varijable kao najpresudnije za uspjeh u itanju. Usprkos injenici da kod djece s perinatalnim ošte enjem mozga kao i kod djece s posebnim jezi nim teško ama postoji rizik za pojavu teško a itanja, segment fonoloških vještina je kod ovih populacija nedostatno istraživan (posebno u hrvatskome jeziku). Cilj rada je usporedba triju razli itih skupina šestogodišnjaka (djece urednoga jezi noga razvoja, djece s perinatalnim ošte enjem mozga i djece s posebnim jezi nim teško ama) na zadacima fonoloških vještina i fonološkog pam enja. Osim raš lambe razlika, opisane su osobitosti skupina na navedenim zadacima. Rezultati pokazuju da se navedene skupine statisti ki zna ajno razlikuju na ve ini varijabli kojima se procjenjuju fonološke vještine i fonološko pam enje. Osim empirijske važnosti, istraživanje pridonosi i logopedskome radu jer opisuje obilježja fonološke sastavnice u djece s rizikom za teško e u enja. Tako er, ono omogu ava oblikovanje primjerenih terapijskih postupaka temeljnih na fonološkim vještinama kao prediktorima teško a itanja. Klju ne rije i: predvještine itanja, rizik za teško e itanja, fonološka svjesnost, fonološko imenovanje 1 Podaci su prikupljeni u okviru znanstvenoistraživa koga projekta "Kognitivni i jezi ni razvoj u djece s neurorazvojnim rizikom" odobrenoga od Ministarstva znanosti, obrazovanja i športa Republike Hrvatske.

Although important insights have been gained by comparing small amounts of data from disparate sources, linguistic generalizations are unlikely to prove valid when based on data whose synchronic status is not well-defined. As a case in point, the recent proposal by Koutsoudas, Sanders & Noll 1974 for universal principles of grammatical rule ordering is here shown to be insufficient, from their own data and on the basis of phonological rules derived from a thorough synchronic analysis of Old Icelandic. It is maintained that a proposal can have positive empirical consequences only if it imposes no conditions to eliminate extrinsic ordering from a grammar.* * We would like to thank Jay Keyser, Robin Cooper, and Emmon Bach for helpful comments arising from discussion concerning aspects of this paper. We alone, however, are responsible for any errors in fact or analysis.

A linguistic theory reaches explanatory adequacy if it arrives at a linguistically-appropriate grammar based on the kind of input available to children. In phonology, we assume that children can succeed even when the input consists of surface evidence alone, with no corrections or explicit paradigmatic information – that is, in learning from distributional evidence. We take the grammar to include both a lexicon of underlying representations and a mapping from the lexicon to surface forms. Moreover, this mapping should be able to express optionality and opacity, among other textbook patterns. This learning challenge has not yet been addressed in the literature. We argue that the principle of Minimum Description Length (MDL) offers the right kind of guidance to the learner – favoring generalizations that are neither overly general nor overly specific – and can help the learner overcome the learning challenge. We illustrate with an implemented MDL learner that succeeds in learning vari...

Different Alignment Based Learning (ABL) algorithms have been proposed for unsupervised grammar induction, e. g. Zaanen (2001) and Déjean (1998), in particular for the induction of syntactic rules. However, ABL seems to be better suited for the induction of morphological rules. In this paper we show how unsupervised hypothesis generation with ABL algorithms can be used to induce a lexicon and morphological rules for various types of languages, e. g. agglutinative or polysynthetic languages. The resulting morphological rules and structures are optimized with the use of conflicting constraints on the size and statistical properties of the grammars, i. e. Minimium Description Length and Minimum Relative Entropy together with Maximum Average Mutual Information. Further, we discuss how the resulting (optimal and regular) grammar can be used for lexical clustering/classification for the induction of syntactic (context free) rules.

A linguistic theory reaches explanatory adequacy if it arrives at a linguistically-appropriate grammar based on the kind of input available to children. In phonology, we assume that children can succeed even when the input consists of surface evidence alone, with no corrections or explicit paradigmatic information – that is, in learning from distributional evidence. We take the grammar to include both a lexicon of underlying representations and a mapping from the lexicon to surface forms. Moreover, this mapping should be able to express optionality and opacity, among other textbook patterns. This learning challenge has not yet been addressed in the literature. We argue that the principle of Minimum Description Length (MDL) offers the right kind of guidance to the learner – favoring generalizations that are neither overly general nor overly specific – and can help the learner overcome the learning challenge. We illustrate with an implemented MDL learner that succeeds in learning vari...

A linguistic theory reaches explanatory adequacy if it arrives at a linguistically-appropriate grammar based on the kind of input available to children. In phonology, we assume that children can succeed even when the input consists of surface evidence alone, with no corrections or explicit paradigmatic information – that is, in learning from distributional evidence. We take the grammar to include both a lexicon of underlying representations and a mapping from the lexicon to surface forms. Moreover, this mapping should be able to express optionality and opacity, among other textbook patterns. This learning challenge has not yet been addressed in the literature. We argue that the principle of Minimum Description Length (MDL) offers the right kind of guidance to the learner – favoring generalizations that are neither overly general nor overly specific – and can help the learner overcome the learning challenge. We illustrate with an implemented MDL learner that succeeds in learning vari...

A linguistic theory reaches explanatory adequacy if it arrives at a linguistically-appropriate grammar based on the kind of input available to children. In phonology, we assume that children can succeed even when the input consists of surface evidence alone, with no corrections or explicit paradigmatic information – that is, in learning from distributional evidence. We take the grammar to include both a lexicon of underlying representations and a mapping from the lexicon to surface forms. Moreover, this mapping should be able to express optionality and opacity, among other textbook patterns. This learning challenge has not yet been addressed in the literature. We argue that the principle of Minimum Description Length (MDL) offers the right kind of guidance to the learner – favoring generalizations that are neither overly general nor overly specific – and can help the learner overcome the learning challenge. We illustrate with an implemented MDL learner that succeeds in learning vari...

Different Alignment Based Learning (ABL) algorithms have been pro-posed for unsupervised grammar induction, eg Zaanen (2001) and Déjean (1998), in particular for the induction of syntactic rules. How-ever, ABL seems to be better suited for the induction of morphological rules. ...

Different Alignment Based Learning (ABL) algorithms have been pro-posed for unsupervised grammar induction, eg Zaanen (2001) and Déjean (1998), in particular for the induction of syntactic rules. How-ever, ABL seems to be better suited for the induction of morphological rules. ...

We develop an analysis of Back vowel harmony for the Finno-Ugric language Votic. Consistent with the tenets of Substance Free Phonology (Hale & Reiss 2008), our analysis makes no reference to markedness considerations related to ease of articulation, ease of perception, typological patterning or the notion of contrast. Our computational and representational toolkit for expressing feature-filling and feature-changing processes is based on the Logical Phonology model (Bale & Reiss 2018) which, in the simplest cases, treats segments as sets of valued features. Rules are modeled as functions mapping strings of segments to strings of segments via basic set theoretic operations like set subtraction and unification. A phonology is the composition of the rules of a language in a particular order. We address some of the complexities of defining long-distance segment interactions in Logical Phonology, and show how vowel transparency can be treated in purely formal terms, without relying on markedness.

Different Alignment Based Learning (ABL) algorithms have been proposed for unsupervised grammar induction, eg Zaanen (2001) and Déjean (1998), in particular for the induction of syntactic rules. However, ABL seems to be better suited for the induction of morphological rules. In this paper we show how unsupervised hypothesis generation with ABL algorithms can be used to induce a lexicon and morphological rules for various types of languages, eg agglutinative or polysynthetic languages. The resulting morphological rules and structures are optimized with the use of conflicting constraints on the size and statistical properties of the grammars, ie Minimium Description Length and Minimum Relative Entropy together with Maximum Average Mutual Information. Further, we discuss how the resulting (optimal and regular) grammar can be used for lexical clustering/classification for the induction of syntactic (context free) rules.

and the audiences at MIT and NELS 48. 1 Where L stands for 'liquid'. Here and below we use phonological rules, which offer a particularly direct handle on the representation of both optionality and opacity, for our examples and our simulations.

There is widespread agreement that both phonological knowledge and spoken linguistic signals  manifest putative universals, which, systematically used, enable infinite productivity. However, less agreement prevails on whether the phonological system is combinatorial  or blending or in nature. In a combinatorial system (DNA is an example), discrete units are combined recursively to yield infinite forms, while maintaining the individuality of each block. For linguistic forms, optimal transmission generally assumes that some contextual blending of phonemes is necessary. It has been variously claimed that transmissibility thus renders phonology merely a summary requirement imposed by the speech channel, while others argue that such purely blending systems cannot account for highly unnatural yet productive patterns that have been attested, and that the greedy reductionism of cognitive abilities to motor functions amounts to substance abuse. An alternative view holds that while phonology is indeed algebraically combinatorial, it is also  subject to functional pressures albeit only selectively. We here report a test of the latter alternative hypothesis.  Our data clearly illustrate that while phonological processes are, in fact, affected by channel concerns, these are attended to selectively in a manner that does not override grammatical requirements. In light of the distinct but converging findings from the two experiments, we conclude that phonology manifests algebraically combinatorial rules that are selectively subject to functional optimization in order to attain discrete infinity while maintaining transmissibility.

Speakers' knowledge of the sound pattern of their language -- their knowledge of morpho-phonology -- goes well beyond the plain phonetic forms of words. According to a long-standing model in linguistics, morpho-phonological knowledge is distributed between a lexicon with morphemes, usually referred to as Underlying Representations (URs), and context-sensitive rewrite rules that transform URs to surface forms. In this paper we provide what to our knowledge is the first unsupervised learner that acquires both URs and phonological rules, including both optionality and rule interaction (both transparent and opaque), from distributional cues alone. Our learner is based on the principle of Minimum Description Length (MDL) which -- like the closely related Bayesian approach -- aims at balancing the complexity of the grammar and its fit of the data.

We develop an evaluation metric for Optimality Theory that allows a learner to induce a lexicon and a phonological grammar from unanalyzed surface forms. We wish to model aspects of knowledge such as the English-speaking child's knowledge that the aspiration of the first segment of k h aet is predictable and the French-speaking child's knowledge that the final l of is optional and can be deleted while that of parle 'speak' cannot. We will try to show that the learner we present succeeds in obtaining this kind of knowledge and is better equipped to do so than other existing learners in the literature.

Stabler (1984) attacks generative models of language acquisition for relying on the psychologically implau- sible basis of simplicity for grammar selection. In this paper I argue that simplicity judgements can be of- fered a psychologically plausible basis. First, a ca- pacity for simplicity judgements is equated with the ability to model a distribution. This ability, in turn, is shown to be one of the capabilities of self-organising maps. These, I argue, are neurologically, and hence psychologically, plausible.