Earticle

Reinforcement learning (RL) is fundamental to current artificial intelligence (AI). Since dynamic programming, which is based on Richard Bellman’s principle of optimality, is the basis of RL (and other AI disciplines such as A* search), it is important to apply that principle correctly and artfully. This tutorial uses examples, many from published articles, to examine both the artful and occasionally erroneous application of Bellman’s principle. Our contribution is twofold. First, we emphasize the necessity of a general thought experiment by asking what we call “consultant questions” to determine the proper state space for DP formulation to be established by Bellman’s optimality principle. Second, we convey the role of art based on common sense and effortful cleverness in designing efficient DP. Since Bellman’s principle is intuitive based on common sense, it is sometimes misunderstood. For illustrative purposes, given a directed acyclic graph (DAG), we consider four deterministic minimum-cost path network problems (found in the literature), each bringing up some validity issue concerning Bellman’s principle on a different criterion. We approach them consistently by the conventional wisdom of state augmentation (on top of the physical system state space). As a result, we show that our artful choice of state description not only renders the principle valid in each problem, but also makes each DP as efficient as the standard DP that solves a shortest-path problem in the same DAG, circumventing successfully the so-called curse of dimensionality, a price to be paid frequently by state enlargement. After that, we discuss the potential importance of the treated type of path network problems for practical applications (e.g., communications network).

This paper investigates how constraints are used in traditional generative phonology and Optimality Theory. In the traditional framework, constraints apply not only to surface representations, but also to the sequential procedure which derives the output from the input. Constraints are not allowed to be violated in this step-by-step derivation,. However, the status of constraints in Optimality Theory is fundamentally different from the way constraints have employed in earlier generative phonological frameworks. In terms of OCP and Coda Conditions, I show that the traditional notion of inviolability of constraints has been radically changed in OT. Higher ranked constraints end up surface-unviolated whereas lower-ranked constraints end up surface-violated. In this framework, it is possible that the attested form violates a number of constraints but is still the optimal candidate.

The purpose of the current study is to examine tonal interaction patterns of blended words in South Kyungsang(SK) Korean in the viewpoint of Optimality Theoretic(OT) approach. From previous research of SK Korean (loanword) compound tone patterns(Lee, 2011, 2013), we observed that the tonal interaction of SK compounds shows fundamental constraints such as the prohibition of long monotone, word initial L tone sequence, and two pitch falls in a prosodic domain. From the basis of the previous study on the tonal interactions of SK blended words(Lee, 2017), this paper investigates whether the OT constraints that were utilized in the tonal interaction analyses of SK compound words can be used in blended word analyses. The results of the present study indicate that the OT constraints of the SK tone system(Lee, 2009), e.g. no word initial L tone sequence, and the prohibition of long monotone (H) words, and of two accents (pitch falls) in a prosodic domain can be utilized in the OT analyses of the SK Korean blended words as well.

This article provides an Optimality-Theoretic analysis of the developmental process of vowel lengthening and shortening, especially focusing on HCL and PCS I and II. In doing so, we have proposed one faithfulness constraint and two markedness constraints. The diachronic changes could be accounted for through three stages. The constraint ranking of the first stage is Ident-IO(long) >> *V[-long] sonC[ αpl]-vdO[ αpl]]σ(V), *V[+long] CC (otherwise), that of the second stage is Ident-IO(long), *V[-long] sonC[ αpl]-vdO[ αpl]]σ(V), *V[+long] CC(otherwise), and that of the last stage is *V[-long] sonC[ αpl]-vdO [αpl]]σ(V), *V[+long] CC(otherwise) >> Ident-IO(long). Additionally, this study supports the direction of diachronic sound change agreed upon, which moves from faithfulness constraints >> markedness ones to markedness constraints >> faithfulness constraint ones. That is, our analysis reflects the promotion of markedness constraints *V[-long] sonC[ αpl]-vdO[ αpl]]σ(V) and *V[+long] CC(otherwise) and the demotion of a faithfulness constraint Ident-IO(long).

We have analyzed OE stress pattern by means of OT, which focuses on the surface representation and constraint ranking. We can explain OE stress from the view point of foot-based OT. The stress foot is the basic unit of OE stress. We propose a new constraint ranking, that is, FT-FORM, NF, ALIGN HEAD >> ALIGN ST, WSP >> PARSE-SYL to explain primary and secondary stress. ALIGN HEAD can handle the fact that OE has the main stress on the first syllable. In addition, FOOT FORM and WSP solve the problem that secondary stress falls on the penultimate syllable followed additional elements.

The purpose of this paper is to show that verb to noun conversion in Papiamentu can be accounted for straightforwardly within the framework of Optimality Theory (Prince & Smolensky, 1993; McCarthy & Prince, 1995). To this end, I assume after Pater (2000) that a single constraint can be multiply instantiated in a constraint hierarchy, and each constraint may be indexed to apply to a particular set of lexical items. This approach is shown to capture distinctions between generality and exceptionality of tone and stress patterns in Papiamentu conversion in terms of the interaction between markedness and faithfulness constraints. I specifically claim that the lexically indexed faithfulness constraint BASE-IDENTITYP outranks the markedness constraints NOUN STRESS and NOUN TONE, which in turn rank above the general faithfulness constraint BASE-IDENTITY. It is shown that with the constraint ranking, all the tone and stress patterns that occur in deverbal nouns can be given a unified, satisfactory account. The analysis argued for in this paper makes better predictions than a traditional approach

There are plenty of phonological processes in natural language that only occur in the context of particular morphemes or lexical categories. The question is how such morpheme (or lexicon) specific processes, sometimes called exceptionality, can be captured in the framework of Optimality Theory (Prince and Smolensky 1993, McCarthy and Prince 1995). Two different types of treatment have mostly been proposed: (i) structural prespecification like floating features, mora, underspecification, etc (Zoll 1996, 2001, Inkelas, Orgun and Zoll 1997, Inkelas 2000, Wolf 2007, Kager to appear, and many others), and (ii) diacritic approach, mostly employing morpheme-specific constraints (Hammond 1995, Golston and Wiese 1995, Golston 1996, Russel 1995, Yip 1998, Pater 2000, 2004, 2007, Ota 2004, Green 2005, Flack to appear, and many others). The purpose of this study is to compare the two different views and to show that the latter is superior to the former in accounting for exceptional plural formation implemented by umlaut in English and German.

This paper points out that the current methodology of Optimality Theory may date back to Pāṇini's rule of optionality by examining Kiparsky's work on Pāṇini's grammar. It is well known that the approach to phonology taken in Optimality Theory lies in its claim that constraints are minimally violable. This is contrasted with the implicit view of traditional generative grammar that an active constraint is inviolable. In this paper, two attempts are made to demonstrate that this theory may also extend to non-phonological areas of linguistics. The first one is an attempt to apply this theory to syntax and the second one is an attempt to explore any possibility of this theory to give an account of dialectal differences on equal terms.

Harmonic Serialism (HS), a derivational version of OT, hinges upon two main premises of gradualness and harmonic improvement. Optimality in HS is always evaluated in a gradual, local, and consistent mode. Revisiting and reanalyzing Korean vowel hiatus phenomena in serial OT, this paper claims that optionality reflects HS's unique property of local optimality. To avoid vowel hiatus at a boundary in Korean, Vowel Coalescence (VC) or Glide Formation (GF) arises. In fact, hiatus remains unresolved while it is fixed via VC or GF optionally sometimes and obligatorily some other times. In essence, multiple optima result from different harmony improvement paths, a local variant with a gradual harmony ascent and a global variant from no gradual path to harmonic improvement

Velar softening, partly historical and partly synchronic in nature, has been dealt with in many different frameworks. The g→ǰ change is predictable, but the k→s change is unexpected and defies an easy explanation. In this paper, we review two key issues in English velar softening. The first issue is the predictability of these changes. The review of Lee(2004) and Randomski(2010) shows that velar softening is unpredictable and that there should be multiple inputs for all words that undergo velar softening. It is proposed in this paper, however, that in most cases we can dispense with multiple inputs. The second issue is the trigger of the velar softening process. It is assumed that part of i-initial suffixes triggers the change. The analysis in this paper shows that contextual velar softening can be explained with a single input, and we limit the use of multiple inputs only for historical residues.

The purpose of this study is two-fold: (i) to review various approaches to ‘long-distance assimilation’ (namely, harmony) within the framework of Optimality Theory (Prince and Smolensky 1993), and (ii) to examine which approach is the most successful to the process. In autosegmental phonology (Goldsmith 1976), the process is mainly accounted for by feature spreading, formalized by association lines between features and segments. However, the Optimality Theoretic approach requires a different perspective: it needs to involve, not rules forcing the harmonic features to spread, but constraint interaction filtering wrong candidates that are under- or over-harmonized. Various approaches have been introduced in this connection. This study classifies them into two main streams: (i) the autosegmental spreading approach, and (ii) the agree-based approach. Through the examination, this study shows that the agree-based approach, especially no-disagreement, is more successful than others to account for harmonic process. The approach without autosegmental representation does not require any specific pro- spreading constraint or representational modification for harmony, and does not induce any problems others have, such as wrong typological prediction and ‘sour-grapes property’. In conclusion, this study verifies that harmonic process can be accounted for without autosegmental spreading.

In Optimality Theory (OT: McCarthy and Prince 1993, Prince and Smolensky 1993/2004), sound change is accounted for by constraint reranking. However, it does not properly show the ‘gradual’ aspect of a diachronic change, though the unranking process is introduced for variation e.g. [knou] > [knou, nou] > [nou] (Cho 1998). Following Oh (2002), this paper proposes an alternative approach to the representation of historical sound change within the OT framework, on the basis of the sound change of know ([knou] > [hnou] > [nou] > [nou]) in Early Modern English (Dobson 1968). To show the gradual changes of [k], MAX, *VL NAS, IDENT [cont], and *σ[kn- constraints are interacted with each other. To display the gradual change of frequency among variants, the meaning of dotted lines is reinterpreted as weak domination contrary to that of straight lines as strong domination, based on Oh (2002). This paper concludes that gradual diachronic changes can be fully represented within the framework of OT through a series of unranking, reranking and ranking tools.

This paper explores how Korean learners of English perceive stop-nasal sequences and how L2 perception can be accounted for within the Optimality Theoretic framework. Voiced stop-nasal sequences in English are possible at the phonetic level, while voiceless stop-nasal ones are permissible with an inserted glottal stop at the phonetic level. Those sequences in Korean are possible at the phonemic level, but they are not permissible at the phonetic level because of a phonological constraint in Korean. This research was designed to provide answers to the following questions: (a) How do native Korean listeners perceive English stop-nasal sequences? (b) Is their perception affected by Korean phonology? (c) What factors play a role in their perception? The results show that Korean listenersʼ perception is different from English listenersʼ perception. Overall, Korean listeners show the perception of epenthetic vowels, which is affected primarily by the release and/or voicing of stops. Their perception is affected by the phonological patterning of Korean in which the release of stops and voiced stops occur only in the syllable-initial position. The variations could be accounted for by some constraints proposed by Davis and Shin (1999) and Dep-IO, and by adopting a model of floating constraints which yield the variable rankings apparent in L2 perception.

In Korean, either nasalization or lateralization occurs to avoid nasal- liquid sequences. However, the choice depends on the lexical items. For example, some words like /shinlaŋ/ ‘husband’ undergo lateralization resulting in [shillaŋ], whereas others like /ɨmunlon/ ‘phonology’ undergo nasalization resulting in [ɨmunnon]. According to Lee (2006), nasalization occurs when nasal and liquid sounds appear across the morpheme boundary. Thus, /ɨmun+lon/ causes nasalization instead of lateralization. Then, how are English words including nasal-liquid sequence pronounced by Korean speakers? To answer this question, this study examines the pronunciation of 10 Korean female middle-school students about nasal- liquid sequence found in both Korean and English. The result of the experiment exhibits some aspects: (i) nasal-liquid sequences are actually pronounced as various ways according to speaker’s recognition of each lexical item, (ii) such optionality of Korean words is consistently transferred into English, (iii) ungrammatical processes in Korean such as deletion or insertion to avoid nasal-liquid sequence never occur in English pronunciation as well. This study provides explicit account for such aspects by adopting optional lexical indexation (Park 2008b) within the framework of Optimality Theory (Prince & Smolensky 1993, McCarthy & Prince 1995). In addition, this study shows that optional lexical indexation is superior to the previous approach of constraints re-ranking in accounting for phonological optionality and transfer.

The main purpose of this study is to provide an explicit analysis of speech errors in cross-linguistic transfer by using Recursive Constraint Demotion algorithm (Tesar & Smolensky 1998, 2000, Tesar 1998, Prince 2002) within the framework of Optimality theory (Prince & Smolensky 1993, McCarthy & Prince 1995). This study focuses on the examples that cause speech error where Korean phonological grammar, especially with regard to nasalization and lateralization, affects English pronunciation. Language learning simply means knowing the constraint hierarchy, since constraints are basically universal and language variation results from different rankings of constraints, according to the basic concept of OT. Therefore, this study pursues how Korean learners of English erroneously use native grammar (nasalization and lateralization), and confirms what happens when Korean grammar is applied to English pronunciation. In addition, variational speech error in Korean-English transfer (either mis-applied nasalization or laterlization to English) is also explained. This study shows that variational speech error results from speakers different usage of morphological structure for a given word.

There have been three kinds of arguments about the status of semi-vowel [j] in Korean within the Korean syllable structure. The three theories, Nucleus Hypothesis, Onset Hypothesis, and Tautosyllabic Hypothesis, consider the semi-vowel [j] in Korean as part of nucleus, onset, and both respectively. Each theory, however, has provided conflicting accounts in reference to analogous evidence. Thus, in this paper we argue that the permission of semi-vowel [j] of Seoul dialect in Korean can be given a functionally grounded treatment by means of *ONSET/X subhierarchy(Smith, 2003: 3) that maintains its universally fixed ranking based on the sonority scale, if the *ONSET/X is defined to be sensitive to the moracity of the segments. And then according to Optimality Theory(Prince & Smolensky, 1993) we will analyse semi-vowel [j] of Seoul and Chonnam dialect in Korean as nucleus. And then the result will be extended to the semi-vowel [j] in English.