Autor: Radek Ocelák (FF UK)


The paper investigates the interaction of topic-focus articulation (TFA), word order, and prosody in spoken Czech within the framework of Optimality Theory (OT). After an outline of TFA and of the ways of its expression in Czech, drawn on the backround of the Functional Generative Description approach, I propose an OT analysis of this complicated phenomenon for rudimentary Czech sentences.


1. Introduction

Topic-focus articulation is one of the traditional labels for the phenomenon that has been also studied under the name of background/fo­cus, topic/com­ment, given/new, theme/rheme, or most originally the psychological subject/psy­chologi­cal predicate opposition. Although there may be fine conceptual differences coming with these terminological options in different approaches, the core of the phenomenon is the following: often it is the case that a sentence does not bring a new information out of the blue, but anchors it to what is already known or otherwise established in the relevant discourse context. Some (namely, the topic) part of the sentence, so to say, picks an element of the already established discourse knowledge as something the whole sentence is about. Another (namely, the focus) part serves to deliver new information about this chosen element.[1] This distinction is especially interesting in so far as it does not always coincide with the usual grammatical subject/predicate distinction, and in natural languages it is realized by various means, including word order (dominant in Italian), prosody (dominant in English), and morphological marking (Japanese).

In Czech, topic-focus articulation (TFA)—conceived as a semantic division of the sentence—is expressed by means of word order and prosody. The present paper aims to explore within the Optimality Theory (OT) framework (e.g., Prince and Smolensky, 1997; Hendriks and de Hoop, 1999; Blutner and Strigin, 2008) the interaction of these three aspects in primitive Czech sentences (consisting only of a subject and an intransitive predicate) and to demonstrate that the OT perspective can provide us with a new insight at this field. I start by examples from Czech and by locating the undertaken task against the background of a particular formal approach to TFA, namely that of the Functional Generative Description framework, from which I also adopt some assumptions. I then provide a detailed optimality-theoretic analysis of a simple Czech example, under a proposed set and ranking of optimality-theoretic constraints and with use of bidirectional optimization. A discussion follows. I conclude by sketching what improvements seem to be needed in order for the proposed OT model to cover also cases of more complex topic-focus structure than those considered here.

2. An outline of topic-focus articulation (TFA) in Czech

Czech is sometimes inaccurately labeled as a language with free word order. Yet there are some strict word order rules, and more importantly, the remaining variation is largely TFA-driven. In general, different word order realizations are only acceptable as expressions of different semantic topic-focus divisions of the sentence, and thus diverge in the discourse conditions under which they are felicitous. Consider the sentence (1):

(1) Petr    včera        rozbil   vázu. 

      Peter yesterday broke   a vase.

While the word order instantiated in (1) is – in a sense yet to be specified – basic for Czech, there are 23 more word order permutations of this sentence, most of which are felicitous under some, in each case slightly varied, conditions.[2] (Here I assume the default position of the main accent on the last phonological word of the sentence in question.) For example,

(2) Včera          rozbil   vázu    Petr.

Yesterday  broke    a vase  Peter.

is acceptable in case that yesterday’s breaking of a vase by an agent not yet known is an established theme or question of the discourse; (2) then newly identifies the agent. So, (2) is an ordinary expression of the topic-focus structure T: včera, rozbil, vázu; F: Petr. This is, however, only a simplified example. Both topic-focus structures and possible sentence realizations are to be further refined. Thus there will be more than just 24 realizations to be linked to 16 possible topic-focus structures in this case (assuming for now that any subset of the sentence’s elements can form the focus and all remaining ones fall under the topic in each case).

Let us now refine the realization (expression) side by including aspects of prosody. Keeping the word order intact, the sentence (1) – just as any of its felicitous permutations – can appear in various prosodic forms. Apart from the main accent, as previously assumed on the last word of (1), I will further consider the option of emphatic accent and of secondary accent falling on a word.[3] According to their function they may also respectively be called corrective and contrastive (this will be elaborated later); phonetic-wise I assume their characterization as, respectively, rising-decreasing and rising, as opposed to the decreasing character of the main accent. Most purely combinatoric variants are clearly excluded (such as those putting more types of accent on one word or those laying several main accents on a single sentence; exact prosodic constraints will be proposed later). But there still remain many variants which are acceptable under specific discourse conditions, or, in other words, which are expressive of specific topic-focus constellations. Such are, e.g., (3) and (4):

(3) Petr [včera]EMPH rozbil  vázu (, [ne]EMPH dneska.) 

Peter yesterday      broke a vase (, not today.)

(4) [Petr]CONTR včera         rozbil  [vázu]MAIN (, ale. 

Peter                   yesterday   broke  a vase (, but

[Pavel]CONTR dneska  šlápl        [na kočku]MAIN.) 

Paul                  today     stepped on a cat.)

In large correspondence with these prosodic options, I will assume somewhat refined underlying topic-focus structures. The focus, whatever it consists of, can, apart from the basic case (-Corr), come as corrective (+Corr), meaning that instead of only resolving an open issue of the discourse context, the focus settles some issue anew, rectifying an already accepted solution. And as to the topic, one (and only one) of its items can be labeled contrastive (+Contr; put in contrast to some other element that is similarly expectable as part of the topic).

In view of the excessive variance both of sentence realizations when word order and prosody are taken into account and of underlying topic-focus structure with the proposed refinements, it seems reasonable to start the investigation with even more primitive structures than the one instantiated in (1). We will be primarily concerned with the structures consisting only of a subject and an intransitive predicate. Before that, I will briefly locate the task here undertaken in a theoretical context, particularly against the background of how TFA is conceived in one particular formal approach, that of Functional Generative Description.

3. The present task on the background of Functional Generative Description (FGD)

Given the sensitivity of the Czech word order (as opposed to the English) to the topic-focus articulation, TFA has for a long time been one of the central interests for Czech linguists. That is the case also in Functional Generative Description (the most formal wing of the Prague linguistic tradition), where TFA with respect to Czech is probably given the most elaborate treatment.[4] In this framework, TFA is taken as an aspect of the deepest layer of language structure that FGD works with, the layer of so-called tectogrammatical structure. Tectogrammatical structure is conceived as a semantically relevant syntactic structure, or the level of „language-specific meaning“, one of more inputs for semantic interpretation (which itself is taken to be language-unspecific). Many, but not all ambiguities of the sentence surface are disambiguated on the tectogrammatical level; e.g. quantifier scope ambiguities are not. On the other hand, even very fine variations of topic-focus structure find their representation on this level.

The tectogrammatical representations of sentences in FGD are dependency, valency-based, non-binary trees with the sentence’s main verb as the root and with other nodes representing other lexical words. TFA is represented in the following way. There is a linear order of so-called communicative dynamism (or “deep word order”) defined on the nodes of the tree.[5] This order of nodes is by a single cut divided into the topic and the focus part of the tree. So the division is exhaustive (each node belongs either to the topic, or to the focus). It is assumed that topic can, but focus cannot be empty. (So a fully focal sentence is possible, wheres a fully topical one, bringing no information at all, is ruled out.) Also, unlike in other approaches, neither topic nor focus is bound to be formed by a single constituent (or subtree, in dependency terms). The “deep word order” of the topic part can vary according to the context, depending on the level of discourse activation of its items – e.g. the most salient or activated item is realized as the leftmost, least dynamic node, called topic proper. The order of the focus part, on the contrary, is determined by the systemic ordering of verb’s arguments, which is specific for each particular language. (The Czech systemic ordering requires Actor to come before Time, that before Location, that before Manner, etc.; from the English ordering it differs especially by the late introduction of Addressee and Patient.[6] Since it is this systemic ordering which determines the order of the nodes in focus, unlike in topic, it seems rather misleading to call it “communicative dynamism order” nevertheless.)

The “deep word order” of a tectogrammatical representation more or less directly (with application of some “shallow” word order rules) translates to the surface order. As to the prosody, only the default prosodic form (main accent on the last phonological word) is taken into account. Therefore, in FGD there seems to be no substantial separate stage of TFA realization by means of word order and accent. Most of the important information is present already on the deepest syntactic level of analysis. We could formulate an optimality-theoretic task, different from ours and much more ambitious, which would comprise both semantics and a part of what I here call realization of TFA: given a description of a discourse situation (with specification of salient items, raised questions etc.), plus the message to be conveyed, as the input, find the optimal output – the tectogrammatical representation most fitting to that input. The task I undertake in this paper, on the other hand, leaves most of the semantic dimension of the problem aside and focuses on the realization problem in its entireness, not disregarding prosody.

So, I take for granted the set of topic-focus variations for a given syntactic structure, and regard them as possible inputs for production optimization (i.e., for the process of finding the optimal expression for what needs to be expressed). Many assumptions are adopted from FGD: the basic structure is dependency and valency-based;[7] the topic-focus division is exhaustive; focus is obligatory, topic is not; neither needs to be formed by a single subtree; if there are more elements in the topic, they come with an order of dynamism. On the other hand, my approach is theoretically neutral in that it does not assume TFA to be represented in the syntactic structure itself. It is rather an external specification which can be added at another stage. Another difference is that I assume no preliminary order on the elements of focus – it is only the employed OT constraints that will make the surface word order reflect the systemic ordering of the particular language. And, unlike FGD, I include the possibility of focus being corrective (see above).

For candidate outputs of the production optimization (from which the optimal expression is to be chosen) I take the realizations of the given sentence structure by means of varied word order and prosody, as outlined above. For optimization in the direction of interpretation (i.e., finding the optimal interpretant for a given expression), the inputs and the outputs are reversed.[8]

4. Optimality Theory analysis

I now get to the OT analysis of how TFA, word-order and prosody interact in the case of primitive Czech sentence structures. I will consider the structure of a subject (Actor) and an intransitive predicate instantiated in (5):

(5) Rodiče odjeli. 

The parents left.

The set of possible production inputs (topic-focus variations of the given sentence structure) consists of:

1. Focus: rodiče, odjeli (-Corrective); Topic: –

2. F: rodiče, odjeli (+Corrective); T: –

3. F: odjeli (-Corrective); T: rodiče (-Contrastive)

4. F: odjeli (-Corrective); T: rodiče (+Contrastive)

5. F: odjeli (+Corrective); T: rodiče (-Contrastive)

6. F: odjeli (+Corrective); T: rodiče (+Contrastive)

7. F: rodiče (-Corrective); T: odjeli (-Contrastive)

8. F: rodiče (-Corrective); T: odjeli (+Contrastive)

9. F: rodiče (+Corrective); T: odjeli (-Contrastive)

10. F: rodiče (+Corrective); T: odjeli (+Contrastive)

The set of candidate outputs can be quite large. Either we can ban the unacceptable prosodic variants already in the listing of candidate outputs, or we can formulate relevant prosodic OT constraints and rank them very high, so that they always make the undesired candidates suboptimal:

(6) MaxOneAccWord: On any word there is at most one accent.

(7) OneAccSent: There is exactly one strong (i. e., main or emphatic) accent per sentence (clause).

(8) *StrongAcc-SecAcc: Strong accent is not followed in one sentence (clause) by secondary accent.

(9) MainSecOneWord: Main and secondary accent can only fall on one word. (Emphatic accent can fall on a larger segment.)

Here I assume the latter option – non-restricted listing of candidate outputs and strong (“absolute”) prosodic constraints. For simplicity, though, I disregard in the following all candidates that violate any of these four constraints, and in turn, I leave these constraints implicit (for they do not make a difference with respect to the remaining candidates). We are left with the following candidate outputs:

a) Rodiče [odjeli]MAIN

b) [Rodiče]MAIN odjeli

c) [Rodiče]CONTR [odjeli]MAIN

d) Rodiče [odjeli]CORR

e) [Rodiče]CONTR [odjeli]CORR

f) [Rodiče]CORR odjeli

g) Odjeli [rodiče]MAIN

h) [Odjeli]MAIN rodiče

i) [Odjeli]CONTR [rodiče]MAIN

j) Odjeli [rodiče]CORR

k) [Odjeli]CONTR [rodiče]CORR

l) [Odjeli]CORR rodiče

m) [Rodiče odjeli]CORR

n) [Odjeli rodiče]CORR

I propose the following violable OT constraints:

(10) CorrFoc -> EmphAcc: If the focus is corrective, the whole of it bears an emphatic accent.

(11) ContrTop -> SecAcc: If there is a contrastive element in the topic, this element bears a secondary accent.

(12) AccLastSent: There is a main accent on the last phonological word of the sentence.

(13) AccLastFoc: There is a main accent on the last focus element of the sentence.

(14) *EmphAcc: There is no emphatic accent in the sentence.

(15) *SecAcc: There is no secondary accent in the sentence.

(16) Stay: Preserve the word order as required by the systemic ordering of the particular language. (This can be more specically implemented in terms of a block of simple constraints requiring Actor to come first, Location to come first, Verb to come first, etc., where the exact ranking is language-specific.)

Further, I propose the following ranking on the constraint set. (As the core assumption of Optimality Theory states, the optimal candidate need not (and often, will not) observe all of the constraints. It can violate some and still be optimal: a candidate X is optimal provided that each other candidate Y involves more violations than X does on the highest ranking constraint on which X and Y differ in their number of violations.)

(17) {CorrFoc ->EmphAcc, ContrTop  -> SecAcc} >{AccLastSent,

AccLastFoc} > {*EmphAcc, *SecAcc, Stay}

The same ranking on the same set of constraints will be used in both the production and the interpretation optimization. Since I am concerned with the whole OT system of 10 possible topic-focus structures and 14 surface realizations, I will skip drawing detailed optimality-theoretic tableaux and I will only report the optimal output(s) for each input in both directions of optimization.

As the constraints are more restrictive in the production direction, there is always exactly one winning candidate in this direction, violating the reported constraints:

  1. F: rodiče, odjeli (-Corrective); T: –

↦ a) Rodiče [odjeli]MAIN – no violation

  1. F: rodiče, odjeli (+Corrective); T: –

↦ m) [Rodiče odjeli]CORR – violation of AccLastSent, AccLastFoc, *EmphAcc

  1. F: odjeli (-Corrective); T: rodiče (-Contrastive)

↦ a) Rodiče [odjeli]MAIN – no violation

  1. F: odjeli (-Corrective); T: rodiče (+Contrastive)

↦ c) [Rodiče]CONTR [odjeli]MAIN – violation of *SecAcc

  1. F: odjeli (+Corrective); T: rodiče (-Contrastive)

↦ d) Rodiče [odjeli]CORR – violation of AccLastSent, AccLastFoc, *EmphAcc

  1. F: odjeli (+Corrective); T: rodiče (+Contrastive)

↦ e) [Rodiče]CONTR [odjeli]CORR– violation of AccLastSent, Acc­LastFoc, *EmphAcc, *SecAcc

  1. F: rodiče (-Corrective); T: odjeli (-Contrastive)

↦ g) Odjeli [rodiče]MAIN– violation of Stay

  1. F: rodiče (-Corrective); T: odjeli (+Contrastive)

↦ i) [Odjeli]CONTR [rodiče]MAIN – violation of *SecAcc, Stay

  1. F: rodiče (+Corrective); T: odjeli (-Contrastive)

↦ f) [Rodiče]CORR odjeli – violation of AccLastSent, AccLastFoc, *EmphAcc

  1. F: rodiče (+Corrective); T: odjeli (+Contrastive)

↦ k) [Odjeli]CONTR [rodiče]CORR – violation of AccLastSent, Acc­LastFoc, *EmphAcc, *SecAcc, Stay

These outcomes seem intuitive for Czech (but see the discussion later). In the direction of interpretation, the constraints are often too weak to pick one optimal candidate[9] and among those they pick as optimal, not all are intuitively acceptable as possible interpretations. As opposed to the direction of production, unidirectional optimization is insufficient in the case of interpretation. On the other hand, bidirectional optimization[10] gives intuitive results:

a)  Rodiče [odjeli]MAIN

↦ optimal interpretation outputs: 1, 3. For both 1 and 3, a) is an optimal realization, so both 1 and 3 are bidirectionally optimal outputs.

b)  [Rodiče]MAIN odjeli

↦ optimal interpretation output: 7. But b) is not an optimal realization for 7, so b) has no bidirectionally optimal output.

c)  [Rodiče]CONTR [odjeli]MAIN

↦ 1, 3, 4. Only 4 is bidirectionally optimal output for c).

d)  Rodiče [odjeli]CORR

↦ 1, 3, 5, 7. Only 5 is bidirectionally optimal output for d).

e)  [Rodiče]CONTR [odjeli]CORR

↦ 1, 3, 4, 5, 6, 7. Only 6 is bidirectionally optimal output for e).

f)  [Rodiče]CORR odjeli

↦ 1, 3, 7, 9. Only 9 is bidirectionally optimal output for f).

g)  Odjeli [rodiče]MAIN

↦ 1, 7. Only 7 is bidirectionally optimal output for g).

h)  [Odjeli]MAIN rodiče

↦ 3, for which h) is not an optimal realization. h) has no bidirectionally optimal output.

i)   [Odjeli]CONTR [rodiče]MAIN

↦ 1, 7, 8. Only 8 is bidirectionally optimal output for i).

j)   Odjeli [rodiče]CORR

↦ 1, 3, 7, 9. No bidirectionally optimal output.

k)  [Odjeli]CONTR [rodiče]CORR

↦ 1, 3, 7, 8, 9, 10. Only 10 is bidirectionally optimal output for k).

l)   [Odjeli]CORR rodiče

↦ 1, 3, 5, 7. No bidirectionally optimal output.

m)[Rodiče odjeli]CORR

↦ 1, 2, 3, 5, 7, 9. Only 2 is bidirectionally optimal output for m).

n)  [Odjeli rodiče]CORR

↦ 1, 2, 3, 5, 7, 9. No bidirectionally optimal output.

So, the following optimal pairings of topic-focus structure on one hand and surface realization on the other are predicted: 1a, 2m, 3a, 4c, 5d, 6e, 7g, 8i, 9f, 10k (as already determined in the unidirectional optimization of production). The realizations b), h), j), l), n) appear in no optimal pairing, which seems rather plausible (the cases b) and h) are further addressed in the subsequent discussion).[11] The realization a), Rodiče [odjeli]MAIN, is predicted to be ambiguous for the topic-focus structures 1 and 3; this is in accord with the observation from Czech (but also English) that sentences with the default word order and main stress on the last word are ambiguous for foci of various sizes. As for the remaining 8 realizations, they form a 1:1 relationship with the remaining 8 topic-focus structures. Among these 8 realizations, 5 are of the word order Rodiče odjeli, which corresponds to the preference for the basic word order. (As realization of the structure 2, m) wins over n) solely due to n)’s violation of the weak condition Stay; the same holds for f) over j) as realization of the structure 9.)

5. Discussion

There are a number of issues and possible objections to this OT analysis.

First, the analysis has a speculative flavour, because the sets of possible inputs (or outputs) of both directions are defined only intuitively and not independently of each other. Especially the possible topic-focus structures are not directly empirically accessible, let alone observable in co-occurence with possible surface realizations. Admittedly, the option of contrastive topic and of corrective focus is not accepted independently, but is rather a consequence of intuitions about the semantic function of secondary and emphatic accent, which are empirically observable. The close correspondence between contrastive topic and secondary accent on one hand and between corrective focus and emphatic accent on the other, captured by the two conditional constraints of our analysis, is then no big surprise.

Second, the resultant asymmetry of production and interpretation (the fact that only the latter is in need of bidirectional optimization) is clearly a consequence of a production bias in formulating the employed OT constraints. For instance, were the two strong constraints concerning contrastive topic and corrective focus formulated as biconditionals instead of conditionals, the number of unidirectionally optimal candidate outputs in interpretation would be severely reduced. On the other hand, bidirectionality is a way of getting stronger results (the adequate, hopefully) without making the employed grammatical constraints “heavier”. Aloni, Butler and Hindsill (2007) quote some experimental findings from language acquisition that are arguably in favor of asymmetry between TFA production and interpretation, the latter lagging behind the former. That corresponds to the analysis proposed here, where topic-focus interpretation is predicted to be cognitively more involved than topic-focus production.

Third, it is an essential feature of the present account that the notion of strong sentence accent (the sentence’s unique intonation center; commonly marked by capitals in the literature on TFA) is split between main and emphatic accent, the latter corresponding to the option of corrective focus. This might be seen as trivialization of the problem of interaction between topic-focus structure and surface realization. The analysis implies that the main accent can only fall on the last phonological word of Czech sentence, and that a strong sentence accent is emphatic whenever it occurs elsewhere in the sentence. That goes against the usual claim that marked prosody and marked word-order are alternative means of expressing the same topic-focus structure in Czech. The question is ultimately empirical. My analysis is inspired by the observation that pre-final sentence accent in Czech tends to be stronger than the default final accent, the former involving a phonetically rising component, and that on sentence’s last word both these forms of accent (main and emphatic) are possible. Also, the pairs of realizations allegedly expressing the same topic-focus structure, such as “Odjeli RODIČE” and “RODIČE odjeli“, do not seem to be completely synonymous and interchangeable. The latter is intuitively more emphatic and unlikely, for instance, in a continuous talk. That is what I try to account for with the notion of corrective focus.[12]

6. Conclusion

Compared to the treatment of Czech topic-focus articulation in FGD, the present analysis is innovative in concentrating on the prosodic dimension of TFA as well. Due to complexity of the resulting interaction of three aspects (topic-focus division on the semantic side, word order as well as prosody on the expression side), I was only able to investigate on the level of the most primitive sentence structures. Here, I believe, OT analysis with its assumption of hierarchy of violable constraints proved fruitful, allowing elegant generalizations about Czech TFA that are uniform for both production and interpretation perspective, independent of FGD’s assumption of the tectogrammatical layer and descriptively adequate to a considerable extent. Undoubtedly there is much to be added to the present toy model if it is to be extended to the case of more complicated sentence structures. In this, many of the generalizations arrived at in FGD could arguably be just reformulated in OT terms. (I assume that extending the considered sentence structure will not bring many further prosodic complications; the rest has been studied extensively in FGD.) In particular, further constraints are needed which would govern the order of elements in topic with respect to their communicative dynamism; word order where it is fixed and irrespective of topic-focus structure, such as that of nominal modifiers; word order of particles, notably the complicating effects of focalizers. By a further strong constraint, splitting of the focus by elements of the topic should be banned.


[1] A major conceptual difference, explicitly made in some approaches, lies in defining topic (background, theme…) either in terms of looser aboutness, or in terms of givenness, where presence of a relevant expression in the previous text is required.

[2] The only ungrammatical orders are all those beginning with the predicate rozbil, possibly with the exception of Rozbil Petr včera vázu expressing the topic-focus structure F: Petr, včera, vázu; T: rozbil.

[3] In the subsequent analysis the emphatic accent will also be allowed to lay on larger than one-word segments.

[4] Hajičová (1991), Hajičová, Partee and Sgall (1998)

[5] More precisely, communicative dynamism is defined as a partial order on the tree, by which the nodes of each local subtree are linearly ordered. The overall linear order of dynamism is then defined in these terms.

[6] Zikánová (2006)

[7] Since I am only working with very simple structures, this rudimentary syntactic analysis seems sufficient and may be cross-theoretically acceptable. Cf. Aloni, Butler, Hindsill (2007) and Samek-Lodovici (2005), who in an optimization task similar to mine use minimalist syntactic structures and refer to them in their sets of OT constraints as well.

[8] As to the semantic dimension of TFA, which I left largely aside, there are various approaches. The FGD authors tend to see the problem in psychological terms and the discourse representation perspective, working with the notion of “degree of activation in memory” or “stock of shared knowledge” (Hajičová, 1991). Other accounts of TFA semantics are in the dynamic predicate logic style (Aloni, Beaver, Clark and van Rooij, 2007) or in terms of tripartite structures (B. Partee’s contribution to Hajičová, Partee and Sgall, 1998). Peregrin, 1996, summarizes the main issues to be solved in an adequate formal semantics of TFA.

[9] Consider for example d), Rodiče [odjeli]CORR: from the constraints, AccLastSent and AccLastFoc are trivially violated with any candidate output (because d) has no main accent), so is *EmphAcc (because there is an emphatic accent in d)), whereas *SecAcc and Stay are trivially satisfied (again simply by d) itself). The only constraints that make a difference with respect to the candidate outputs are the conditional ones, CorrFocŠEmphAcc and ContrTopŠSecAcc, and these are survived by the candidates 1, 3, 5, 7 (for in 1, 3, 7 there is no contrastive topic and no corrective focus, and in 5 there is a corrective focus on odjeli which is reflected by the emphatic accent in d)).

[10] A candidate production output is bidirectionally optimal if and only if it is an optimal candidate for the production input and, at the same time, the latter constitutes an optimal interpretation output for the former taken as an interpretation input; cf. Blutner and Strigin (2011).

[11] I do not have clear intuitions about the excluded realization j). If it is plausible, contrary to the present analysis, it might suggest a different formulation of AccLastSent, one in favor of any strong (not just main) accent on sentence’s final word. That would lead to a substantial change in the proposed OT system.

[12] H. Zeevat instead suggests that the strong sentence accent becomes pre-final in Czech in order for the sentence to repeat the word order of a previous sentence while keeping the intended focus marked. That is often the case and indeed, these sentences are called “second instance sentences” in FGD. But in my opinion, presence of a previous sentence with the same word order is not a necessary condition for their felicity, and my solution in terms of corrective focus seems to cover these as well as other cases.


Aloni, M.; Beaver, D.; Clark, B.; van Rooij, R.: 2007. The Dynamics of Topic and Focus. In: Aloni, M.; Butler, A.; Dekker, P. (eds.): Questions in Dynamic Semantics. Elsevier, Amsterdam. Pp. 123–145.

Aloni, M.; Butler, A.; Hindsill, D.: 2007. Nuclear Accent, Focus, and Bidirectional OT. In: Aloni, M.; Butler, A.; Dekker, P. (eds.): Questions in Dynamic Semantics. Elsevier, Amsterdam. Pp. 253–268.

Blutner, R. and Strigin, A.: 2011. Bidirectional Grammar and Bidirectional Optimization. In: Benz, A. and Mattausch, J. (Eds.): Bidirectional Optimality Theory, Amsterdam: Benjamins. Pp. 221–248.

Hajičová, E.; Partee, B. H.; Sgall, P.: 1998. Topic-Focus Articulation, Tripartite Structures, and Semantic Content. Kluwer Academic Publishers, Dordrecht.

Hajičová, E.: 1991. Topic-focus articulation and coreference in models of discourse production. Journal of Pragmatics, 16, pp. 157–166.

Hendriks, P. and de Hoop, H.: 2001. Optimality Theoretic Semantics. Linguistics and Philosophy, 24, 1, pp. 1–32.

Peregrin, J.: 1996. Topic and focus in a formal framework. In: Partee, B.; Sgall, P. (eds.): Discourse and Meaning. Benjamins, Amsterdam. Pp. 235–254.

Prince, A. and Smolensky, P.: 1993/2002. Optimality: Constraint interaction in generative grammar. Technical Report, Rutgers University Center for Cognitive Science and Computer Science Department, University of Colorado.

Samek-Lodovici, V.: 2005. Prosody-Syntax Interactions in the Expression of Focus. Natural Language and Linguistic Theory, 23 (3), pp. 687–755.

Zikánová, Š.: 2006. What do the data in Prague Dependency Treebank say about systemic ordering in Czech? The Prague Bulletin of Mathematical Linguistics, 86, pp. 39–46.