View in EDS HTML Full Text PDF Full Text

Noise, Economy, and the Emergence of Information Structure in a Laboratory Language

Saved in:
Bibliographic Details
Title: Noise, Economy, and the Emergence of Information Structure in a Laboratory Language
Language: English
Authors: Stevens, Jon S., Roberts, Gareth
Source: Cognitive Science. Feb 2019 43(2).
Availability: Wiley-Blackwell. 350 Main Street, Malden, MA 02148. Tel: 800-835-6770; Tel: 781-388-8598; Fax: 781-388-8232; e-mail: cs-journals@wiley.com; Web site: http://www.wiley.com/WileyCDA
Peer Reviewed: Y
Page Count: 33
Publication Date: 2019
Document Type: Journal Articles
Reports - Research
Descriptors: Communication (Thought Transfer), Time, Difficulty Level, Information Theory, Behavior Patterns
DOI: 10.1111/cogs.12717
ISSN: 0364-0213
Abstract: The acceptability of sentences in natural language is constrained not only grammaticality, but also by the relationship between what is being conveyed and such factors as context and the beliefs of interlocutors. In many languages the critical element in a sentence (its focus) must be given grammatical prominence. There are different accounts of the nature of focus marking. Some researchers treat it as the grammatical realization of a potentially arbitrary feature of universal grammar and do not provide an explicit account of its origins; others have argued, however, that focus marking is a (grammaticalized) functional solution to the problem of efficiently transmitting information via a noisy channel. By adding redundancy to highlight critical elements in particular, focus protects key parts of the message from noise. If this information-theoretic account is true, then we should expect focus-like behavior to emerge even in non-linguistic communication systems given sufficient noise and pressures for efficiency. We tested this in an experiment in which participants played a simple communication game in which they had to click cells on a grid to communicate one of two line figures drawn across the grid. We manipulated the noise, available time, and required effort, and measured patterns of redundancy. Because the lines in many cases overlapped, meaning that only some parts of each line could be used to distinguish it from the other, we were able to compare the extent to which effort was expended on adding redundancy to critical (non-overlapping) and non-critical (overlapping) parts of the message. The results supported the information-theoretic account of focus and shed light on the emergence of information structure in language.
Abstractor: As Provided
Entry Date: 2019
Accession Number: EJ1206726
Database: ERIC
Full text is not displayed to guests.
FullText Links:
  – Type: pdflink
    Url: https://content.ebscohost.com/cds/retrieve?content=AQICAHj0k_4E0hTGH8RJwT4gCJyBsGNe_WN95AvKlDbXJGqwxwESn9FKL0RVPTHgEgKvbhnDAAAA4zCB4AYJKoZIhvcNAQcGoIHSMIHPAgEAMIHJBgkqhkiG9w0BBwEwHgYJYIZIAWUDBAEuMBEEDA_1DOIhZFQDwCSllwIBEICBm15lrV3sAtVpsrk_d4fAy0AcGT9RC4PDu0qo_EBs5OaXTeJu6xeN8p3JW9cbhRa0CxO1Hzqm3B51RtCHheU5HPuf4vwv24niGMECUZPPniDxOM6F_jzhGTljzkU_elj-V_0V5caJMJz8nJ5TmhLHB9nOasGjRJUgO7YBwVPOn8kU1k8T0KdUT0WEqsw4aCMhCNJ2ZSYt4HO3YgtO
Text:
  Availability: 1
  Value: <anid>AN0134910149;cgn01feb.19;2019Feb27.06:21;v2.2.500</anid> <title id="AN0134910149-1">Noise, Economy, and the Emergence of Information Structure in a Laboratory Language </title> <p>The acceptability of sentences in natural language is constrained not only grammaticality, but also by the relationship between what is being conveyed and such factors as context and the beliefs of interlocutors. In many languages the critical element in a sentence (its focus) must be given grammatical prominence. There are different accounts of the nature of focus marking. Some researchers treat it as the grammatical realization of a potentially arbitrary feature of universal grammar and do not provide an explicit account of its origins; others have argued, however, that focus marking is a (grammaticalized) functional solution to the problem of efficiently transmitting information via a noisy channel. By adding redundancy to highlight critical elements in particular, focus protects key parts of the message from noise. If this information‐theoretic account is true, then we should expect focus‐like behavior to emerge even in non‐linguistic communication systems given sufficient noise and pressures for efficiency. We tested this in an experiment in which participants played a simple communication game in which they had to click cells on a grid to communicate one of two line figures drawn across the grid. We manipulated the noise, available time, and required effort, and measured patterns of redundancy. Because the lines in many cases overlapped, meaning that only some parts of each line could be used to distinguish it from the other, we were able to compare the extent to which effort was expended on adding redundancy to critical (non‐overlapping) and non‐critical (overlapping) parts of the message. The results supported the information‐theoretic account of focus and shed light on the emergence of information structure in language.</p> <p>Keywords: Linguistic focus; Laboratory languages; Cultural evolution; Information theory; Communication; Pragmatics; Information structure; Experimental semiotics</p> <p>In any language, utterances are constrained in what form they can take. The grammatical rules of the language are an obvious example (e.g., "Love you I" is an impossible sentence in English). In some cases, however, the acceptability of a sentence depends not only on its grammaticality, but also on the relationship between the information being conveyed and factors such as discourse context and the beliefs of the interlocutors (Bolinger, [<reflink idref="bib5" id="ref1">5</reflink>]; Roberts, [<reflink idref="bib41" id="ref2">41</reflink>]; Rooth, [<reflink idref="bib48" id="ref3">48</reflink>]). For example, the question Who invented the printing press? can be answered in English in several ways, each conveying the same basic message:</p> <p></p> <p> <ephtml> <table><tbody><tr><td align="left">(1)</td><td align="left">Q:</td><td align="left">Who invented the printing press?</td></tr><tr><td align="left">A:</td><td align="left">Gutenberg invented the printing press. / Gutenberg did. / Gutenberg.</td></tr></tbody></table> </ephtml> </p> <p>Crucially, though, there are a number of similar grammatical sentences that cannot be answers to this question. For example, this particular question cannot be answered with ''#The printing press was invented", where '#' denotes a response that is <emph>infelicitous</emph>, that is, unacceptable in context. Similarly, any answer in which some element were given greater prosodic emphasis than ''Gutenberg" (as in Example 2, where small caps denote prosodic emphasis) would also be unacceptable:</p> <p></p> <p> <ephtml> <table><tbody><tr><td align="left">(2)</td><td align="left">Q:</td><td align="left">Who invented the printing press?</td></tr><tr><td align="left">A:</td><td align="left"><sc>Gutenberg</sc> invented the printing press. / #Gutenberg invented the <sc>printing press</sc>. / #Gutenberg <sc>invented</sc> the printing press.</td></tr></tbody></table> </ephtml> </p> <p>These infelicitous sentences are all in principle grammatical in English; they are unacceptable only as answers to this particular question. However, their unacceptability is not due to the <emph>quality</emph> of the information provided. All of them are factually accurate; but they are still unacceptable as answers to the question. By contrast, an answer like "thomas edison invented the printing press" is factually incorrect but entirely acceptable from a linguistic point of view. The unacceptability of the various infelicitous sentences given is due to constraints imposed by information structure: An acceptable and complete answer to the question, however long or short it is, must contain an element identifying the inventor of the printing press. It must not, moreover, emphasize other elements at the expense of that one. This unelidable, or (following Schmitz, [<reflink idref="bib49" id="ref4">49</reflink>]) "critical," element is the <emph>focus</emph> of the answer. Several theories of focus have been offered (for an overview, see Stevens, [<reflink idref="bib57" id="ref5">57</reflink>]), but a dominant role is played by the notion of <emph>discourse alternatives</emph> (see in particular Rooth, [<reflink idref="bib48" id="ref6">48</reflink>]; Roberts, [<reflink idref="bib41" id="ref7">41</reflink>]). In the example, the question sets up a space of potential alternatives ("Gutenberg invented the printing press," Edison invented the printing press, and so on) from which the answerer is expected to select as accurate and relevant a response as they are able (Grice, [<reflink idref="bib18" id="ref8">18</reflink>]; Wilson & Sperber, [<reflink idref="bib62" id="ref9">62</reflink>]). The information that is not shared with all the alternatives (''Gutenberg") is the critical information that allows the correct response to be selected from its competitors, and thus constitutes the focus of the sentence. The grammar of English and many other languages (e.g., German, Portuguese, Mandarin) requires that focused elements bear a higher degree of phonological prominence than other elements, though the precise realization of this prominence varies between languages. In English, it takes the form of a pitch accent within the focused element, coupled with an increase in duration and intensity (Katz & Selkirk, [<reflink idref="bib23" id="ref10">23</reflink>]; Ladd, [<reflink idref="bib29" id="ref11">29</reflink>]). While a prosodic component to focus marking is very common cross‐linguistically, other mechanisms exist in many languages, with focus being marked either by syntactic position (as in, e.g., Welsh or Hungarian; Borsley, Tallerman, & Willis, [<reflink idref="bib6" id="ref12">6</reflink>]; Horvath, [<reflink idref="bib20" id="ref13">20</reflink>]) or by the inclusion of extra morphemes (as in, e.g., Middle Egyptian or Guruntum; Zimmermann & Onea, [<reflink idref="bib64" id="ref14">64</reflink>]; Loprieno, [<reflink idref="bib33" id="ref15">33</reflink>]), which is also an available option in English (e.g., ''It was Gutenberg who invented the printing press"). It should be noted at this point that it is something of a simplification to group all of these phenomena together as if they were identical. The semantic implications of what we call focus marking in these languages vary. In Hungarian or Welsh, for example, the so‐called focus position requires an exhaustive interpretation, similarly to English clefts. That is, the equivalent of ''John played guitar for the Beatles" in Hungarian with <emph>John</emph> in the focus position would entail (falsely) that the Beatles had no other guitar players (Horvath, [<reflink idref="bib20" id="ref16">20</reflink>]). Also, Guruntum and other West Chadic languages have focus marking that further requires that the utterance be interpreted emphatically (Zimmermann, [<reflink idref="bib63" id="ref17">63</reflink>]). Nevertheless, we consider that these phenomena have enough in common that–for present purposes—it is useful and reasonable to treat them as parts of the same general story.</p> <p>However it is marked, the alignment between informational focus and its linguistic realization is robust: If prosodic prominence is applied to something other than the critical element (as in Example 2) the sentence is treated as unacceptable. (Such mismatches have been found to give rise to a number of observable neurological correlates; Hruska & Alter, [<reflink idref="bib21" id="ref18">21</reflink>]). The robustness of this alignment, taken together with the existence of straightforwardly morphosyntactic mechanisms for marking it has led many linguists to treat focus as a purely syntactic feature. Rochemont ([<reflink idref="bib46" id="ref19">46</reflink>]), for instance, defended the thesis ''that focus has a uniform grammatical definition only as a syntactic element" (p. 1). In this view, a speaker of English is forced by the grammar to mark focus on a particular element for the same kind of reason that a speaker of a gender‐marking language might have to use a different form of an adjective with a masculine noun than a feminine one. This treatment of focus as entirely grammar‐internal has been typical of generative accounts such as the cartographic approach (defined by Rizzi & Cinque, [<reflink idref="bib40" id="ref20">40</reflink>] as an attempt to ''to draw maps as precise and detailed as possible of syntactic configurations"; see Rizzi, [<reflink idref="bib39" id="ref21">39</reflink>] for such an account of focus). As to the origins of focus, such accounts are essentially silent. A consistent interpretation would be that focus is simply a language‐specific realization of an underlying (and potentially arbitrary) feature of universal grammar, with grammar‐ and (a fortiori) language‐external factors more or less excluded on axiomatic grounds; Rizzi and Cinque ([<reflink idref="bib40" id="ref22">40</reflink>], p. 45), for instance, explicitly articulated the assumption that the functional projections involved in syntactic accounts of focus are universal to all languages or, at the very least, belong to a universal inventory of projections from which individual languages can select as necessary. While this kind of account does not entirely exclude the possibility of ultimately functional language‐external explanations for the emergence of focus systems, it does imply that focus is essentially baked into universal grammar, so such an explanation would have to involve biological evolution.</p> <p>More recently, however, a cultural‐evolutionary account has been developed that seeks to explain the origins of focus systems in terms of adaptation to language‐external constraints. While focus marking is indeed grammaticalized in many languages, in the sense that the particular rules about how to mark it are conventions built into the grammar, Stevens ([<reflink idref="bib55" id="ref23">55</reflink>],[<reflink idref="bib56" id="ref24">56</reflink>]) argued that this is simply a fact about how focus is realized in those languages. Explaining focus in a more general sense requires reference to the pressures acting on communication. Focus systems are, by this account, neither drawn from some universal inventory of syntactic features (like singular vs. plural)[<reflink idref="bib1" id="ref25">1</reflink>] nor simply arbitrary grammatical quirks of particular languages (like the famously numerous noun classes in Bantu languages, whose relationships to concrete semantic categories are loose as best; see Braver & Bennett, [<reflink idref="bib7" id="ref26">7</reflink>]); rather, they emerge through language use as a highly functional solution to the problem of efficiently transmitting information via a noisy channel (Shannon & Weaver, [<reflink idref="bib53" id="ref27">53</reflink>]).</p> <p>If this is correct, then focus should not be treated as a special property of natural language (or even as an innate communicative adaptation); rather, we should expect focus‐like behavior to emerge through use in any communication system in which information must be efficiently protected from noise. The purpose of the work presented here is to test this in the laboratory. In particular, we present a simple laboratory communication game in which participants exchanged messages using a non‐linguistic medium. By manipulating noise, effort, and time pressures, we were able to investigate how different information‐theoretic pressures influence the form of the messages, shedding light on how focus systems might emerge in real‐world communication systems, including natural language. The next section outlines in more detail the information‐theoretic view of focus. Section 3 motivates our experimental approach, Section 4 lays out our hypothesis and predictions, and the method and results are presented in Sections 5 and 6.</p> <hd id="AN0134910149-2">Focus and information transmission</hd> <p>Stevens ([<reflink idref="bib55" id="ref28">55</reflink>],[<reflink idref="bib56" id="ref29">56</reflink>]), following Schmitz ([<reflink idref="bib49" id="ref30">49</reflink>]) and Bergen and Goodman ([<reflink idref="bib2" id="ref31">2</reflink>]), analyzed focus as a solution to the problem of how to transmit linguistic meaning through a noisy channel. The vast majority of communication is subject to noise, which in information‐theoretic terms refers to the random alteration or deletion of parts of the signal (see Levy, [<reflink idref="bib31" id="ref32">31</reflink>] and Gibson, Bergen, & Piantadosi, [<reflink idref="bib17" id="ref33">17</reflink>] on the ubiquity of noise in communication and the necessity of taking it into account in models of language processing). In real‐world terms, this could mean literal noise obscuring the phonetic signal, mishearing or inattention on the part of the hearer, or anything else leading to the random loss or alteration of elements in the message. Noise can be compensated for by adding redundancy to the signal (Shannon & Weaver, [<reflink idref="bib53" id="ref34">53</reflink>]). Spelling alphabets, in which the word cat might be spelled out as "Charlie Alpha Tango," are an example of explicitly designed systems for adding redundancy to signals: A full word is substituted for each letter, making the letters easier to distinguish from each other at the expense of added time and effort. Because the additional time and effort is considerable, such systems are used only for channels in which noise is perceived to be high or accurate communication is particularly crucial. In ordinary language, redundancy may be added using prosodic emphasis (e.g., lengthening a word or increasing its acoustic intensity), by including extra morphemes to highlight the element (e.g., "it was Guten‐berg who..."), or applying a syntactic operation to move the element to a different (more salient) position in the sentence (as in Welsh or Hungarian).[<reflink idref="bib2" id="ref35">2</reflink>]</p> <p>Noise, however, is not the only pressure acting on communication. If that were so, we might expect redundancy to be maximized—applied, for instance, to every element in the sentence. As should be clear from the examples given above, this is not what we find; rather, we find a system for applying redundancy in a minimal fashion such that the critical element is better protected from noise than those elements in the sentence that overlap with the potential alternatives. Natural‐language focus seems therefore to result from a tension between a drive to maximize redundancy, on the one hand, and a drive to reduce effort and time costs, on the other (on the remarkably short time available to speakers in linguistic interactions, see Levinson, [<reflink idref="bib30" id="ref36">30</reflink>]; for a summary of the role of effort in pragmatics, see Horn, [<reflink idref="bib19" id="ref37">19</reflink>]). Because adding redundancy to the signal is inherently effortful, and often adds time, it is more efficient to apply redundancy selectively, distributing redundancy only to the parts of the signal that benefit the most from it—that is, those parts of the signal that are most threatened by noise. If parts of the signal are recoverable by the hearer (i.e., can be inferred from context), then it is inefficient to add redundancy to them. Critical elements are those that are non‐recoverable, and this is where redundancy will most help combat noise. With respect to phonological prominence in particular, Stevens ([<reflink idref="bib55" id="ref38">55</reflink>],[<reflink idref="bib56" id="ref39">56</reflink>]) argued that prosodic focus systems distribute peaks in prominence so as to approximate an ideal mapping between redundancy and criticality of information, although for various reasons (such as independent phonological constraints) the ideal is not actually met. Fig. 1 illustrates this ideal, where the material that survives the effects of noise is exactly the material that cannot be recovered from "context." In what follows, we expand on what we mean by context, which relies on the notion of alternatives.</p> <p>Graph: From Stevens ([<reflink idref="bib55" id="ref40">55</reflink>]), an ideal relationship between prominence, noise, and information criticality vs. recoverability.</p> <p>In many prominent theories of focus (e.g., Roberts, [<reflink idref="bib41" id="ref41">41</reflink>]; Rooth, [<reflink idref="bib48" id="ref42">48</reflink>]), the discourse context is taken to delineate a set of alternative utterances—things the speaker could have said. The delineation of such a restricted set makes the information in what was said more easily recoverable by listeners. This is most clear in the case of questions and answers (as in Example 3).</p> <p></p> <p> <ephtml> <table><tbody><tr><td align="left">(3)</td><td align="left">Q:</td><td align="left">Who invented the printing press?</td></tr><tr><td align="left">A:</td><td align="left"><sc>Gutenberg</sc> invented the printing press.</td></tr></tbody></table> </ephtml> </p> <p>If we assume, following Roberts ([<reflink idref="bib41" id="ref43">41</reflink>]), that any discourse is structured into <emph>questions under discussion</emph> (QUDs), and that assertions are meant to address those QUDs, then the set of alternative assertions to ''Gutenberg invented the printing press" is simply the set of assertions that would address the (in this case explicitly stated) QUD, ''Who invented the printing press?" (e.g., "Gutenberg invented the printing press," "Edison invented the printing press," etc.).</p> <p>We can make this clearer by representing assertions as structured objects consisting of two parts (Krifka, [<reflink idref="bib28" id="ref44">28</reflink>]): the qud, represented as an assertion with an open variable (e.g., "<emph>x</emph> invented the printing press" for "who invented the printing press?"), and the answer, which supplies the value of the open variable (e.g., 'Gutenberg'). This leads to a set of alternative assertions as in Fig. 2. Since the context constrains assertions to those that address the same QUD, the only locus of variation is in the element that supplies the open variable—"Gutenberg," "Edison," and so on. That is, everything else is predictable, and therefore recoverable by the hearer. The only critical element is the identity of <emph>x</emph> in "<emph>x</emph> invented the printing press".</p> <p>Graph: A structured set of alternative assertions.</p> <p>The same logic applies to utterances that are not responses to explicit questions. For example, focus can be used to draw a mutual contrast between two elements of an utterance:</p> <p></p> <p> <ephtml> <table><tbody><tr><td align="left">(4)</td><td align="left">It was a <sc>German</sc> inventor, not an <sc>American</sc> one.</td></tr></tbody></table> </ephtml> </p> <p>In this case, the type of element being contrasted is different. Rather than selecting assertions from alternatives, the task is better envisioned as one of resolving the meaning of the constituents "German inventor" and "American one" (which stands for "American inventor"), each of which refers to some entity. As in Fig. 2, we can conceive of these entity meanings as having some internal structure in the form of attributes like nationality and occupation whose values are predicates that hold for that entity. Fig. 3 illustrates. Here, the same mechanics can be brought in to predict the optimal focus pattern. The occupation is shared between the alternatives, and thus does not constitute critical information. Thus, any redundancy in the signal aimed at combating noise should be added only to the element that conveys the critical nationality information. In this alternative‐based framework, we may refer to the redundancy added to the critical information as <emph>non‐overlapping redundancy</emph>, since the critical element is precisely that element whose meaning does not overlap with all of the alternatives. Accordingly, we may refer to redundancy on the recoverable information (like occupation) as <emph>overlapping redundancy</emph>.</p> <p>Graph: A structured set of alternative entities.</p> <p>We can consider precisely the same phenomenon in a different modality. Fig. 4 shows two line figures, displayed against grids. The two figures are not identical, but overlap to an extent. That overlapping part is shared between the two lines just as "x invented the printing press" and occupation are shared between the entities in Figs. 2 and 3. If a person were tasked with singling out one of the two lines (perhaps by drawing it), they could of course (given sufficient time) choose to represent the whole line. A more efficient response, however, would be to focus the most effort on reproducing only a part of the line that does not overlap with the other line the critical element in this case.</p> <p>Graph: Two partially overlapping line figures.</p> <p>On this view, focus is a natural and non‐language‐specific response to the pressures of noise and efficiency on communication. Noise adds pressure to make the signal redundant, while efficiency adds pressure to economize and distribute the redundancy selectively, minimizing the relative degree of overlapping redundancy. Efficiency can further be broken down into the following: (a) minimization of effort to produce a signal, and (b) minimization of the time taken to do so. These are clearly related, although not identical, and may be valued differently in different contexts.</p> <p>The emergence and (cultural) evolution of focus systems can, by this account, be understood as a cultural evolutionary process operating over multiple timescales (Fig. 5). First, during a single interaction, speakers develop ad hoc strategies to respond to noise, effort, and time constraints (cf. Brennan & Clark, [<reflink idref="bib8" id="ref45">8</reflink>]; Clark, [<reflink idref="bib9" id="ref46">9</reflink>]; Krauss & Weinheimer, [<reflink idref="bib27" id="ref47">27</reflink>]). Second, over multiple interactions, successful strategies are maintained and become habitual behaviors (Krauss & Weinheimer, [<reflink idref="bib27" id="ref48">27</reflink>]; Neal, Wood, & Quinn, [<reflink idref="bib38" id="ref49">38</reflink>]). Third, children acquire strategies from observing adults (Romaine, [<reflink idref="bib47" id="ref50">47</reflink>]).[<reflink idref="bib3" id="ref51">3</reflink>] Finally, the strategies that one generation acquires are themselves transmitted to the next generation, and so on; this process of iterated learning over generations leads originally more ad hoc, discourse‐level strategies to be systematized and incorporated into the grammar as rules (Kirby, Griffiths, & Smith, [<reflink idref="bib25" id="ref52">25</reflink>]; Tamariz & Kirby, [<reflink idref="bib58" id="ref53">58</reflink>]; Traugott & Dasher, [<reflink idref="bib59" id="ref54">59</reflink>]).</p> <p>Graph: Four overlapping timescales for the emergence of linguistic structure.</p> <p>In summary, the information‐theoretic account claims that focus systems are a set of grammaticalized mechanisms for providing selective redundancy to an utterance. These mechanisms include prosodic emphasis on the critical element (as in "Gutenberg invented the printing press"), its movement to a more salient position (as in Welsh or Hungarian), or the inclusion of extra morphosyntax to mark it out (as in an English cleft sentence like "it was Gutenberg who invented the printing press"). Some of these mechanisms are inherently uneven; for example, the existence of salient positions in the sentence means necessarily that some parts of the sentence cannot be as salient as others. Certain forms of prosodic emphasis, however, could be applied equally to all parts of the utterance. This is what we might expect if noise were a problem, but effort and time were no object. Given that they are, redundancy must be applied to the parts of the utterance that need it most—the critical elements that do not overlap with potential alternative utterances. A natural corollary of this account is that any communication system subject to noise, effort, and/or time pressures is likely to develop an analog to focus, at least to the extent expected over the second timescale described above. Furthermore, work on iterated learning suggests that any such system that is transmitted over generations would likely develop rule‐like focus systems (see e.g., Kirby et al., [<reflink idref="bib25" id="ref55">25</reflink>]; Kirby, Tamariz, Cornish, & Smith, [<reflink idref="bib26" id="ref56">26</reflink>]; while such work has yet to be conducted on focus systems in particular, Macuch Silva & Roberts, [<reflink idref="bib34" id="ref57">34</reflink>] used an iterated‐learning paradigm to investigate the emergence of reduplication as a response to noise.)</p> <p>In the study presented here, we did not investigate generational transmission, but investigated whether and when ad hoc strategies for the efficient distribution of redundancy would emerge in a non‐linguistic signaling system. To this end, we had participants play a visual communication game in which we manipulated noise, effort, and time pressures. As predicted, participants responded by selectively distributing redundancy to critical parts of the signal as pressures were increased. In the following section, we briefly describe the background to our experimental approach; in Section 4, we present our hypothesis and specific predictions in more detail; and in Section 5.1, we present our method.</p> <hd id="AN0134910149-3">Experimental semiotics</hd> <p>To test our hypothesis, we required that participants play a game in which they had to communicate but could not use natural language, and in which we could manipulate the constraints acting on the system. Over the last decade and a half, an experimental approach has been developed that answers these requirements. This approach, termed <emph>Experimental Semiotics</emph> by Galantucci ([<reflink idref="bib15" id="ref58">15</reflink>]), involves participants playing games in which they either learn a novel artificial language (e.g., Kirby, Cornish, & Smith, [<reflink idref="bib24" id="ref59">24</reflink>]; Sneller & Roberts, [<reflink idref="bib54" id="ref60">54</reflink>]) or collaboratively construct a novel communication system in the laboratory (e.g., Fay, Garrod, Roberts, & Swoboda, [<reflink idref="bib12" id="ref61">12</reflink>]; Galantucci, [<reflink idref="bib14" id="ref62">14</reflink>]; Roberts, Lewandowski, & Galantucci, [<reflink idref="bib45" id="ref63">45</reflink>]). The approach was devised primarily to investigate the emergence of language and of linguistic structure and differs from classic artificial language learning approaches (e.g., Culbertson, Smolensky, & Legendre, [<reflink idref="bib10" id="ref64">10</reflink>]; Fedzechkina, Jaeger, & Newport, [<reflink idref="bib13" id="ref65">13</reflink>]; Kam & Newport, [<reflink idref="bib22" id="ref66">22</reflink>]) in the inclusion of a necessary social component. That is, participants are exposed to each other's communicative output, either directly through interaction (e.g., Galantucci, [<reflink idref="bib14" id="ref67">14</reflink>]; Sneller & Roberts, [<reflink idref="bib54" id="ref68">54</reflink>]), or – in iterated learning experiments—indirectly through exposure in training to the output of previous participants (e.g., Kirby et al., [<reflink idref="bib24" id="ref69">24</reflink>]; Roberts & Fedzechkina, [<reflink idref="bib44" id="ref70">44</reflink>]); in some experiments (e.g., Kirby et al., [<reflink idref="bib26" id="ref71">26</reflink>]), both occur (for a review of the approach see Galantucci, Garrod, & Roberts, [<reflink idref="bib16" id="ref72">16</reflink>]; for arguments in favor of its wider application to testing historical and sociolinguistic hypotheses, see Roberts, [<reflink idref="bib43" id="ref73">43</reflink>]). For those experiments involving communication, the task ranges from taking turns to naming items from a closed set of referents (e.g., Kirby et al., [<reflink idref="bib26" id="ref74">26</reflink>]) to relatively open‐ended coordination and trading games (e.g., Galantucci, [<reflink idref="bib14" id="ref75">14</reflink>]; Roberts, [<reflink idref="bib42" id="ref76">42</reflink>]). The communication media employed have been similarly broad, ranging from graphical or gestural communication (e.g., Bergmann, Dale, & Lupyan, [<reflink idref="bib3" id="ref77">3</reflink>]; Dale & Lupyan, [<reflink idref="bib11" id="ref78">11</reflink>]; Fay et al., [<reflink idref="bib12" id="ref79">12</reflink>]; Galantucci, [<reflink idref="bib14" id="ref80">14</reflink>]; Micklos, [<reflink idref="bib36" id="ref81">36</reflink>]) to auditory (Verhoef, Kirby, & de Boer, [<reflink idref="bib61" id="ref82">61</reflink>]) or even tactile (Trendafilov, Lemmelä, & Murray‐Smith, [<reflink idref="bib60" id="ref83">60</reflink>]) signaling.</p> <p>The goal of most Experimental Semiotic studies is to observe the emergence of conventions from nothing (or at least from some highly unstructured starting point). As might be imagined, these studies often pose a significant challenge to participants, with many failing to communicate at all (Galantucci, [<reflink idref="bib14" id="ref84">14</reflink>]; Scott‐Phillips, Kirby, & Ritchie, [<reflink idref="bib51" id="ref85">51</reflink>]). For our purposes, however, posing a challenge of this nature would have been counterproductive. The point of our experiment was not to observe whether our participants could communicate successfully, but rather to see how they would distribute effort and redundancy in their signals. In contrast to the majority of Experimental Semiotic studies, therefore, we set up the task such that communication was trivial, but could be made more or less efficient. In particular, we presented participants with paired line figures drawn over grids as in Fig. 4, one of which was highlighted for one of the players (Fig. 6). In each turn, the player with the highlighted line figure would signal to the other player which figure that was by selecting cells in a blank grid to be transmitted. The basic task is trivial, but it is also easy to manipulate time constraints, the number of clicks (and thus the amount of effort) necessary to select cells, and the likelihood that a selected cell will in fact be transmitted, thus allowing the impact of different information‐theoretic pressures to be investigated.</p> <p>Graph: Sender's screen.</p> <hd id="AN0134910149-4">Hypothesis and predictions</hd> <p>Our participants played a simple non‐linguistic communication game; we manipulated noise, time, and effort pressures. With respect to the emergence of focus‐like behavior in such a system, there are three possibilities as to what might occur, each of which has a different theoretical consequence:</p> <p></p> <ulist> <item> The "walled‐off hypothesis": No analogs to focus arise in our non‐linguistic communication task → focus is purely linguistic, and language is not recruited for this task.</item> <p></p> <item> The "language‐recruitment hypothesis" : Analogs to focus arise in all conditions, regardless of levels of noise, time, and effort constraints → either focus is the reflex of domain‐general processes for which communication pressures do not matter, or language is always recruited for our non‐linguistic communication task</item> <p></p> <item> The "communicative need hypothesis": Analogs to focus arise only when time/effort/noise pressures are added → either we should consider linguistic focus as having likely arisen in response to such communicative pressures, or we should consider that language is recruited for non‐linguistic communication only when those pressures apply</item> </ulist> <p>We hypothesize that the last of these, the communicative need hypothesis, is correct. By "analogs to focus" it is important to understand that we do not mean behavior that is identical to focus systems found in natural language. Such systems, where focus is an integral part of the grammar of a language, and where the application or non‐application of focus marking is typically governed by categorical rules, should be expected to arise over generations, not over the course of repeated interactions, as in our experiment (see Fig. 5). Rather, what we mean by "analogs to focus" is behavior in which such features as message length and the distribution of redundancy vary in specific ways in response to noise, time, and effort pressures. Crucially, if the communicative need hypothesis is correct, these features should be sensitive to such pressures to a degree that is not the case in natural‐language focus systems, which are now grammaticalized systems rather than emergent strategies for responding to information‐theoretic constraints.</p> <p>In particular, we make predictions about the following features.</p> <hd id="AN0134910149-5">Message length</hd> <p>If the communicative need hypothesis is correct, overall message length should vary according to time and effort costs (compare the different answers in Example 1 above). In particular, messages should be longer if effort costs are low and time is not pressing, and should get shorter as these pressures increase. This may seem a trivial prediction—as available time decreases, for instance, it seems obvious that messages should get shorter. However, the key prediction is that messages will not simply be minimally short in all cases. There are two reasons for this. First, any message that is longer than it needs to be contains redundancy, which helps protect against noise, so there are advantages to keeping messages longer. Second, identifying the most efficient message may itself take effort and time.[<reflink idref="bib4" id="ref86">4</reflink>] The prediction of the walled‐off hypothesis is somewhat unclear here, and the two hypotheses are hard to distinguish with regard to this feature alone, though we might expect a bimodal distribution of message lengths, such that some participants prefer to save time and realize they can do so by sending minimally short messages, while others simply send the longest messages they can. The language‐recruitment hypothesis, by contrast, predicts that messages will be similarly short regardless of time and effort cost.</p> <hd id="AN0134910149-6">Distribution of redundancy</hd> <p>According to the communicative need hypothesis, longer messages should differ from shorter messages not only with respect to length, but also with respect to the proportion of overlapping redundancy (degree of redundancy added to elements that do not serve to distinguish the intended meaning from its alternatives). In particular, there should be a positive relationship between message length and the proportion of overlapping redundancy (or, to put it another way, a positive relationship between message <emph>concision</emph> and the proportion of <emph>non</emph>‐overlapping redundancy). This is because overlapping redundancy is a greater luxury than non‐overlapping redundancy, and should be dispensed with more readily. The shortest possible message should consist of only non‐overlapping material. The other two hypotheses make quite different predictions from this, and from each other. The walled‐off hypothesis predicts that redundancy should be distributed rather evenly, with little regard for overlap. The language‐recruitment hypothesis predicts that we should see very little overlapping redundancy at all, even in longer messages. That is, if language is recruited in this task, we should not see equivalents of "<emph>Gutenberg</emph> invented the <emph>printing press</emph>".</p> <hd id="AN0134910149-7">Distribution of effort</hd> <p>The distribution of effort in a message should take noise into account, just as a speaker whose answer "Gutenberg" was initially misheard might reply with the more effortful sentence, "I said it was Gutenberg". In particular, more effort should be expended on adding redundancy to non‐overlapping items when noise is higher. This should be true both in an absolute sense (more effort expended on adding redundancy overall) and in a relative sense (more effort expended on non‐overlapping redundancy than on overlapping redundancy). This differs from the previous prediction, which concerns <emph>where</emph> redundancy is added (e.g., which elements are given prominence). The <emph>Distribution of effort</emph> prediction concerns the amount of redundancy applied to a given element – in speech this might correspond to the degree of prosodic prominence, for instance. The question can be asked only for systems that allow variable degrees of redundancy. Here, the prediction for the walled‐off hypothesis is similar to that for distribution of redundancy: an even distribution of effort. The language‐recruitment hypothesis, by contrast, agrees to some extent with the communicative need hypothesis. That is, to the extent that higher noise involves more effortful focus marking in natural language, we should expect this to be recruited in the experiment, such that more effort is applied when noise is greater. The language‐recruitment hypothesis differs here with respect to the <emph>obligatory</emph> nature of grammatical focus marking in most natural languages. It is not felicitous in English, for instance, to apply no prosodic marking to the critical element at all, even if noise is very low. Based on the language‐recruitment hypothesis, then, we should not expect the proportion of overlapping redundancy to vary between conditions; but such variation is entirely consistent with the communicative need hypothesis.</p> <hd id="AN0134910149-8">Communicative accuracy</hd> <p>Unless noise and time pressures become so great that they simply prevent accurate communication, we should expect communicative accuracy to remain relatively constant, regardless of the variation between messages predicted above. This is because that variation should be designed to help maintain accuracy under different conditions. Here, the language‐recruitment hypothesis makes a similar prediction; based on the walled‐off hypothesis, however, we should expect communication to suffer as a result of noise, time, and effort pressures.</p> <p>In summary, the walled‐off hypothesis differs from the communicative need hypothesis in predicting that focus‐like responses to noise will not occur, while the language‐recruitment hypothesis differs by predicting that focus‐like behavior will occur regardless of the presence of noise.</p> <hd id="AN0134910149-9">Experiment 1</hd> <p>In Experiment 1, participants played a simple communicative game in which levels of noise were manipulated along with the time limit. In order to test the predictions given above, we measured message length, distribution of redundancy, distribution of effort, and communicative accuracy, including how these features changed over time.</p> <hd id="AN0134910149-10">Method</hd> <p></p> <hd id="AN0134910149-11">Participants</hd> <p>Sixty University of Pennsylvania students participated in pairs for course credit or $5. Pairs who did well at the task received $2 extra each.</p> <hd id="AN0134910149-12">Procedure</hd> <p>In each trial a pair of participants played a simple cooperative signaling game, with a space theme (see Appendix for instructions). Each sat in a separate cubicle with a computer; neither participant could see or hear the other. The game consisted of a series of turns in each of which one player was nominated as <emph>Sender</emph> and the other as <emph>Receiver</emph>; players alternated roles, with the Sender in the first turn being selected at random. At the start of a turn the Receiver saw a white screen with the message "You are waiting on a message transmission from the other player." The Sender saw a screen as in Fig. 6. On the left were two 7 × 7 grids, over each of which a different line figure was drawn. Every line figure consisted of a continuous line drawn between the center points of eleven contiguous grid cells. One of the two line figures was selected in green, while the other was white.[<reflink idref="bib5" id="ref87">5</reflink>]</p> <p>On the right of the screen was an empty 7 × 7 grid, slightly larger than the other two. Beneath the two leftmost grids there was also a button marked <emph>Send</emph>. The Sender's task was to communicate to the Receiver which of the two grids was selected in green by clicking on cells in the rightmost grid. At the moment of the Sender's first click, a timer would start. Once the timer stopped (after either 5 or 30 s, depending on the condition; see Section <emph>Experimental conditions</emph>) the grids would disappear and be replaced by the message "You are waiting on a guess from the other player." Clicking the Send button would have the same result. Once the Sender's turn had come to an end in one of those two ways, the Receiver's screen would change to display the two line figures that had been displayed to the Sender (in a random order, but each in the same orientation as for the Sender) as well as a third 7 × 7 grid in which certain cells might be colored black, and a button marked <emph>OK</emph> (Fig. 7). The black cells would always be cells that the Sender had clicked; however, not all cells that the Sender had clicked would necessarily be sent, with the likelihood that a given cell would be sent depending on how many times it had been clicked on (see Section 5.1.3). The Receiver's task was to select which of the two line figures they thought the Sender was trying to communicate. Both players were then told whether the Receiver chose correctly. Then a new turn began. There were 48 turns in total, which were preceded by two practice turns. In half the turns (the <emph>Overlap turns</emph>), the two line figures overlapped by five squares (as in Figs. 6 and 7). This is analogous to competing natural‐language sentences such as "German inventor Gutenberg invented the printing press" and "American inventor Edison invented the printing press," which overlap with respect to five words. In the other 24 turns (the <emph>Filler turns</emph>), there was no such overlap, such that any cell through which a line figure passed would serve to distinguish it from its competitor.</p> <p>Graph: Receiver's screen.</p> <p>Graph: image_n/cogs12717-fig-0007.png</p> <p>The only differences between the practice turns and the other turns were that the players' success in the practice turns did not count toward their final score, and that, following the practice turns, they were told to ask questions if they had any. As a motivator, players who scored over 80% in the non‐practice turns were rewarded with $2 each. Only two pairs—both in the Low noise (5 s) condition—did not qualify for this reward.</p> <hd id="AN0134910149-13">Experimental conditions</hd> <p>There were three between‐subjects conditions in total (Table 1). Conditions differed from each other with respect to the time available to the sender (either 5 or 30 s) and with respect to the level of noise affecting communication, operationalized as the probability with which a given cell that the Sender had clicked on would actually be sent and thus be seen by the Receiver. Any cell that had been clicked on at least once had a chance of being sent, and each additional click that the Sender made on a cell would increase that chance. Each click would also make the cell appear darker to the Sender, with the darkness increasing proportionally to the probability that the cell would be sent. (Though those cells that did arrive would appear uniformly black to the Receiver.) The probability of a given cell being sent was calculated as 1 – (1–<emph>d</emph>)<sups><emph>n</emph></sups>, where <emph>d</emph> is a decay parameter between 0 and 1, and <emph>n</emph> equals the number of times the Sender clicked on the cell in question. Two values were used for the decay parameter. In the <emph>High noise condition</emph>, it was set at 0.1. In the <emph>Low noise conditions</emph>, it was set at 0.4. This meant that it would take many more clicks in the High noise condition than in the Low noise conditions to feel confident that the cell would be sent. For instance, four clicks in the latter condition would result in an 87% chance of the cell being sent, but a 34% chance in the former. In all three conditions, the number of cells clicked on can be taken to correspond roughly to utterance length, while cell darkness can be taken to be analogous to greater effort in marking prominence.</p> <p>Conditions of Experiment 1</p> <p> <ephtml> <table><thead valign="top"><tr><th align="left">Condition Name</th><th align="left">Time Limit</th><th align="left">Constraint Type</th><th align="left">Constraint Specification</th></tr></thead><tbody><tr><td align="left">High noise</td><td align="left">30 s</td><td align="left">Noise</td><td align="left">Probability of cell deletion decays with <italic>d </italic>=<italic> </italic>0.1</td></tr><tr><td align="left">Low noise</td><td align="left">30 s</td><td align="left">Noise</td><td align="left">Probability of cell deletion decays with <italic>d </italic>=<italic> </italic>0.4</td></tr><tr><td align="left">Low noise (5 s)</td><td align="left">5 s</td><td align="left">Noise</td><td align="left">Probability of cell deletion decays with <italic>d </italic>=<italic> </italic>0.4</td></tr></tbody></table> </ephtml> </p> <p>The length of time available for the Sender to click on cells was varied, being set at either 5 or 30 s. The purpose of this was to be able to observe the way in which time pressures modulate the response to noise. The 5 s time limit was found, in piloting, to be enough time to send a message in all conditions, while still imposing a rather substantial time pressure. Participants in pilots, however, rarely took more than 30 s to send their messages. For Low noise, there was both a 5 s condition and a 30 s condition. However, we ran only a 30 s version of the High noise condition. This is because 5 s was not enough time for participants to send more than one cell in the High effort condition, or to have a good chance of doing so in the High noise condition, meaning that the results would be rather trivial and most of our measures could not be calculated.</p> <p>Participants were given explicit information about the condition they were in. That is, they were told in advance what the time limit was (and could see it count down on the screen); they also knew that the darkness of the cell corresponded to the likelihood that it would be sent (though the precise equation was not shared with them), and that all cells would appear black to the Receiver (see Appendix).</p> <hd id="AN0134910149-14">Measures</hd> <p>We measured several variables. In what follows, a <emph>message cell</emph> refers to a cell for which there could be some expectation that it would be part of the message sent to the Receiver. In Experiment 1, this means any cell clicked on.</p> <hd id="AN0134910149-15">Message length</hd> <p>This was calculated by counting the number of message cells. Since one cell would be sufficient to distinguish between any two line figures (assuming it arrived), any message consisting of more than one cell can be considered to contain redundancy.</p> <hd id="AN0134910149-16">Overlapping redundancy</hd> <p>This was calculated by counting how many message cells overlapped with both line figures and would therefore not differentiate the target figure from its competitor. (Note that this could be measured only for the Overlap turns, and not the Filler turns.)</p> <hd id="AN0134910149-17">Click effort</hd> <p>This was calculated by counting the total number of clicks made by the Sender.</p> <hd id="AN0134910149-18">Communicative accuracy</hd> <p>This was calculated by counting the number of turns in which the Receiver chose the correct line figure.</p> <hd id="AN0134910149-19">Results</hd> <p>Data reported in the following sections come from Overlap turns only, for the sake of consistency between the measures. Unless otherwise stated, all models reported are mixed effects linear regression models with random intercepts for both subject and item, using the Satterthwaite approximation of degrees of freedom to obtain a <emph>p</emph>‐value from a t‐value.</p> <hd id="AN0134910149-20">Message length</hd> <p>We predicted that overall message length should vary according to time and effort costs, with messages being longer than strictly necessary if time and effort constraints allowed. This was supported by the data. One cell would have been sufficient in any condition to signal which line figure to choose; however, messages were longer than this in every condition (Fig. 8). In the Low noise conditions, message length also remained relatively constant throughout a game, though it was considerably lower in the Low noise (5 s) than in the Low noise condition (<emph>β </emph>= −4.83, <emph>S E </emph>=<emph> </emph>0.75, <emph>t </emph>= −6.48, <emph>p </emph><<emph> </emph>.001), suggesting that the shorter message lengths in the 5 s condition was due to time constraints. In the High noise condition, message length began high and fell over the course of the game in a rather linear fashion, though it did not quite converge with message length in the Low noise (5 s) condition. If the two 30 s conditions are compared, there is a significant interaction between turn number and condition (<emph>β </emph>= −1.18, <emph>S E </emph>=<emph> </emph>0.74, <emph>t </emph>= −1.6, <emph>p </emph><<emph> </emph>.001), but no overall effect of turn number (<emph>β </emph>= −0.008, <emph>SE </emph>=<emph> </emph>0.01, <emph>t </emph>= −0.59, <emph>p </emph>=<emph> </emph>.56); this suggests that message length decreased as the game progressed only when noise was high. A likely interpretation of this result is that noise (and the effort required to compensate for it) came to exercise increasing pressure as time went on. It may also partly reflect participants' growing familiarity with the game. Indeed, the time it took Senders to start clicking on cells decreased significantly over the course of the game (<emph>β </emph>= −7.278<emph>e</emph> − 02, <emph>SE </emph>=<emph> </emph>9.885<emph>e</emph> − 03, <emph>t </emph>= −7.363, <emph>p </emph><<emph> </emph>.001; there was no difference between conditions). Familiarity alone is unlikely to be the sole explanation for the trend, however; to the extent that familiarity with the game leads to shorter messages, we should expect this to manifest itself in a realization that tracing the whole (or the majority) of the line figure is unnecessary. The expected consequence of such a realization is a relatively abrupt fall in message length rather than a steady linear decline.</p> <p>Graph: Message length in Experiment 1.</p> <p>Graph: image_n/cogs12717-fig-0008.png</p> <hd id="AN0134910149-21">Distribution of redundancy</hd> <p>We predicted that shorter messages should differ from longer messages with respect to the proportion of overlapping redundancy. Consistent with this, there was a significant positive relationship between message length and the proportion of overlapping redundancy: <emph>β </emph>= 1.374<emph>e</emph> − 02, <emph>SE </emph>=<emph> </emph>3.275<emph>e</emph> − 03, <emph>t </emph>=<emph> </emph>4.2, <emph>p </emph><<emph> </emph>.001 (Fig. 9).</p> <p>Graph: Correlation between message length and overlapping redundancy.</p> <p>Graph: image_n/cogs12717-fig-0009.png</p> <hd id="AN0134910149-22">Distribution of effort</hd> <p>We predicted that the distribution of effort would take noise into account and that non‐overlapping redundancy would be higher when noise is higher, both in an absolute sense and relative to overlapping redundancy. There was a clear difference between conditions, suggesting that noise was indeed taken into account (Fig. 10). As might be expected, click effort was lowest in the Low noise (5 s) condition. However, it was higher in the High noise condition than in the Low noise condition (<emph>β </emph>= 58.57, <emph>S E </emph>=<emph> </emph>7.56, <emph>t </emph>=<emph> </emph>7.75, <emph>p </emph><<emph> </emph>.001), even though rounds lasted 30 s in both these conditions. If all conditions are taken together, there was no significant relationship between click effort and overlapping redundancy (<emph>β </emph>= 4.561<emph>e</emph> − 04, <emph>S E </emph>=<emph> </emph>2.553<emph>e</emph> − 04, <emph>t </emph>=<emph> </emph>1.786 <emph>p </emph>= .075; Fig. 11A). Interestingly, this was also true of the High effort condition alone (<emph>β </emph>= 5.847<emph>e</emph> − 04, <emph>SE </emph>=<emph> </emph>3.620<emph>e</emph> − 04, <emph>t </emph>=<emph> </emph>1.62, <emph>p </emph>=<emph> </emph>.11; Fig. 11B). However, if the first three turns are ignored, there was a relationship (<emph>β </emph>= 1.361<emph>e</emph> − 03, <emph>SE </emph>=<emph> </emph>4.055<emph>e</emph> − 04, <emph>t </emph>=<emph> </emph>3.36, <emph>p </emph>=<emph> </emph>.001). While one must be wary of arbitrarily excluding data, this is consistent with the pattern seen in the message length data, suggesting that in the first few turns, participants were getting to grips with the game, and that their strategy improved over time. Notably, the same is not true of either of the other conditions (where there is no relationship even if all data from the first half of the game are excluded), suggesting that the pattern is a response to noise. Indeed, with respect to the proportion of overlapping redundancy, there was a significant interaction of turn and condition for the High noise and Low noise conditions (<emph>β </emph>= 0.006, <emph>SE </emph>=<emph> </emph>0.002, <emph>t </emph>=<emph> </emph>3.07, <emph>p </emph><<emph> </emph>.01). Although all the noise conditions began at roughly the same place (Fig. 12), the proportion of overlapping redundant cells clicked remained relatively constant in the Low noise condition, but decreased in the High noise condition. There is also a main effect of condition if the interaction terms are excluded from the model (<emph>β </emph>= 0.11, <emph>SE </emph>=<emph> </emph>0.04, <emph>t </emph>=<emph> </emph>2.81, <emph>p </emph><<emph> </emph>.01). The proportion of clicks on overlapping redundant cells was significantly lower in the Low noise (5 s) than in the High noise condition (<emph>β </emph>= −0.10, <emph>SE </emph>=<emph> </emph>0.04, <emph>t </emph>= −2.44, <emph>p </emph><<emph> </emph>.05).</p> <p>Graph: Click effort in experiment 1.</p> <p>Graph: image_n/cogs12717-fig-0010.png</p> <p>Graph: Proportion of clicks on overlapping redundant cells in Experiment 1.</p> <p>Graph: image_n/cogs12717-fig-0011.png</p> <p>Graph: Proportion of clicks on overlapping redundant cells in Experiment 1 over time.</p> <p>Graph: image_n/cogs12717-fig-0012.png</p> <hd id="AN0134910149-23">Accuracy rates</hd> <p>We predicted that accuracy would remain similar between conditions, regardless of differences in message length, click effort, and redundancy. This was broadly true. Overall mean accuracy was very high (96%; Fig. 13) and was significantly lower only in the Low noise (5 s) condition (<emph>β </emph>= −0.07, <emph>S E </emph>=<emph> </emph>0.02, <emph>t </emph>= −2.81, <emph>p </emph><<emph> </emph>.01), where it was 90%. This likely represents an underestimation of noise by participants; this was the only condition in which senders had little time to send a signal but could not be sure before sending it how much the receiver would see.</p> <p>Graph: Accuracy rates by condition (Experiment 1). Error bars show 95% confidence intervals.</p> <p>Graph: image_n/cogs12717-fig-0013.png</p> <hd id="AN0134910149-24">Discussion of Experiment 1</hd> <p>The results of Experiment 1 were consistent with the communicative‐need hypothesis. Message length varied according to time and effort costs, and was longer than necessary in all conditions (contrary to the language‐recruitment hypothesis). Shorter messages involved a lower proportion of overlapping redundancy (contrary to the walled‐off hypothesis), but redundancy was applied to overlapping segments more than would be expected based on the language‐recruitment hypothesis. More effort was applied when noise was higher, contrary to the walled‐off hypothesis. Finally, communicative accuracy remained consistent across conditions.</p> <p>These results are encouraging. A remaining question concerns effort. In Experiment 1, noise posed a threat to signaling, but could be mitigated by applying effort. The effect of a pressure for greater effort, in other words, cannot be distinguished in this experiment from the effect of noise alone. Within this paradigm, manipulating noise is hard without also manipulating effort requirements, at least if we want participants to be able to respond to noise pressures (which, for our measures, we do). The reverse, however, is not so difficult. In Experiment 2, we manipulated effort requirements in a way parallel to Experiment 1, but in an absolute, deterministic way that eliminated noise.</p> <hd id="AN0134910149-25">Experiment 2</hd> <p>In Experiment 2, participants played the same simple communicative game as in Experiment 1, except that whether or not a cell would be sent depended entirely on the Sender's clicks exceeding a specific threshold. We also manipulated time pressures as in Experiment 1.</p> <hd id="AN0134910149-26">Method</hd> <p></p> <hd id="AN0134910149-27">Participants</hd> <p>Sixty University of Pennsylvania students participated in pairs for course credit or $5. Pairs who did well at the task received $2 extra each.</p> <hd id="AN0134910149-28">Procedure</hd> <p>The basic task was identical to the task in Experiment 1, except that the Sender could be certain whether or not a given cell would be sent. In particular, it would be sent if it had been clicked on ≥<emph>x</emph> times, where x was a specific number that was stated in the instructions and varied according to condition (Section 6.1.3). Once a cell had been clicked on enough times to be sent, it would turn black. Otherwise it would stay white.</p> <hd id="AN0134910149-29">Experimental conditions</hd> <p>There were three between‐subjects conditions in total (Table 2). Conditions differed from each other with respect to the time available to the sender (either 5 or 30 s) and with respect to how many times a Sender had to click on a cell for it to be sent. The Sender had to click fifteen times on each cell in the <emph>High effort condition</emph> and five times in the <emph>Low effort conditions</emph>. This would make the cell turn black and guarantee that it would be sent; if the Sender exceeded this number, the cell would still be sent; if they fell short of this number, it would not be.</p> <p>Conditions of Experiment 2</p> <p> <ephtml> <table><thead valign="top"><tr><th align="left">Condition Name</th><th align="left">Time Limit</th><th align="left">Constraint Type</th><th align="left">Constraint Specification</th></tr></thead><tbody><tr><td align="left">High effort</td><td align="left">30 s</td><td align="left">Effort</td><td align="left">15 clicks required to send a cell</td></tr><tr><td align="left">Low effort</td><td align="left">30 s</td><td align="left">Effort</td><td align="left">5 clicks required to send a cell</td></tr><tr><td align="left">Low effort (5 s)</td><td align="left">5 s</td><td align="left">Effort</td><td align="left">5 clicks required to send a cell</td></tr></tbody></table> </ephtml> </p> <p>As in Experiment 1, participants were given explicit information about the condition they were in, including how many clicks would be sufficient for a cell to be sent.</p> <hd id="AN0134910149-30">Measures</hd> <p>The same variables were measured in the same way as in Experiment 1, except that a message cell in Experiment 2 consisted of any cell that had turned black. This is, however, consistent with Experiment 1 inasmuch as a message cell is defined as any cell that the Sender can have some expectation will be sent.</p> <hd id="AN0134910149-31">Results</hd> <p>Data reported in the following sections come from Overlap turns only, for the sake of consistency between the measures. Unless otherwise stated, all models reported are mixed effects linear regression models with random intercepts for both subject and item, using the Satterthwaite approximation of degrees of freedom to obtain a <emph>p</emph>‐value from a <emph>t</emph>‐value.</p> <hd id="AN0134910149-32">Message length</hd> <p>Given the lack of probabilistic cell deletion, it would have been even easier in Experiment 2 than in Experiment 1 to signal any line figure using one cell only. However, messages were again longer than this in every condition (Fig. 14). Consistent with Experiment 1, it was, however, considerably lower in the Low effort (5 s) condition than in the Low effort condition (<emph>β </emph>= −7.64, <emph>S E </emph>=<emph> </emph>0.65, <emph>t </emph>= −11.83, <emph>p </emph><<emph> </emph>.001). In the High effort condition, message length began high and fell over the course of the game in a rather linear fashion, converging with the Low effort (5 s) condition, and thus falling lower than message length in the High noise condition. Indeed, there is a significant interaction between turn number and condition if the High noise and the High effort conditions are compared (<emph>β </emph>= −3.09, <emph>SE </emph>=<emph> </emph>0.37, <emph>t </emph>= −8.29, <emph>p </emph><<emph> </emph>.001), likely due to the lack of a probabilistic threat to the signal in the High effort condition. There was also a significant interaction between turn number and condition with respect to the two 30 s effort conditions (<emph>β </emph>= −2.40, <emph>SE </emph>=<emph> </emph>0.77, <emph>t </emph>= −3.13, <emph>p </emph><<emph> </emph>.01) but no overall effect of turn number (<emph>β </emph>= −0.001, <emph>S E </emph>=<emph> </emph>0.02, <emph>t </emph>= −0.08, <emph>p </emph>=<emph> </emph>.93), suggesting that message length decreased as the game progressed only when effort was high.</p> <p>Graph: Message length in Experiment 2.</p> <p>Graph: image_n/cogs12717-fig-0014.png</p> <hd id="AN0134910149-33">Distribution of redundancy and effort</hd> <p>We predicted that shorter messages should differ from longer messages with respect to the proportion of overlapping redundancy. The pattern was similar to that seen in Experiment 1: <emph>β </emph>= 3.487<emph>e</emph> − 02, <emph>SE </emph>=<emph> </emph>2.343<emph>e</emph> − 03, <emph>t </emph>=<emph> </emph>14.88, <emph>p </emph><<emph> </emph>.001 (Fig. 15).</p> <p>Graph: Correlation between message length and overlapping redundancy.</p> <p>Graph: image_n/cogs12717-fig-0015.png</p> <p>While the relationship between click effort and condition in this experiment was similar to that in Experiment 1 (<emph>β </emph>= −37.867, <emph>SE </emph>=<emph> </emph>8.473, <emph>t </emph>= −4.469, <emph>p </emph><<emph> </emph>.001), this is rather trivial given the nature of the task. More notably, the proportion of clicks devoted to overlapping redundancy (as opposed to non‐overlapping redundancy) was positively related to click effort, as in Experiment 1: <emph>β </emph>= 2.589<emph>e</emph> − 03, <emph>SE </emph>=<emph> </emph>2.367<emph>e</emph> − 04, <emph>t </emph>=<emph> </emph>10.94, <emph>p </emph><<emph> </emph>.001 (Fig. 16A) and the relationship was even clearer in the High effort condition: <emph>β </emph>= 0.003, <emph>SE </emph>=<emph> </emph>0.0002, <emph>t </emph>=<emph> </emph>18.43, <emph>p </emph><<emph> </emph>.001 (Fig. 16B).</p> <p>Graph: Proportion of clicks on overlapping redundant cells in Experiment 2.</p> <p>Graph: image_n/cogs12717-fig-0016.png</p> <hd id="AN0134910149-34">Accuracy rates</hd> <p>We predicted that accuracy would remain similar between conditions, regardless of differences in message length, click effort, and redundancy. This was true of the Effort conditions, for which overall mean accuracy (97%) did not differ significantly between conditions (Fig. 17).</p> <p>Graph: Accuracy rates by condition. Error bars show 95% confidence intervals.</p> <p>Graph: image_n/cogs12717-fig-0017.png</p> <hd id="AN0134910149-35">Discussion of Experiment 2</hd> <p>The results of Experiment 2 are very similar to those of Experiment 1, suggesting that the effect of noise on signaling operates via effort pressures. The pattern seen for distribution of effort and redundancy cannot be related to noise in Experiment 2, but it must be due to variation in required effort: The more work one is required to do, the more it must be concentrated where it is most needed. Noise, however, may still have an independent role to play. The fact that message length in the High effort condition converged with message length in the Low effort condition, while such a convergence did not quite occur in Experiment 1, is likely due to the uncertainty engendered by noise in Experiment 1.</p> <hd id="AN0134910149-36">Overall discussion</hd> <p>We set out to test an information‐theoretic account of linguistic focus, the phenomenon whereby phonological or morphosyntactic mechanisms are employed to highlight the most important information in context. Many existing accounts of focus (such as the so‐called cartographic approach; see e.g. Rizzi, [<reflink idref="bib39" id="ref88">39</reflink>]) conceive of it as a grammar‐internal feature, similar to grammatical gender or number and do not attempt to explain why such a feature would exist in grammar. The information‐theoretic account laid out by Stevens ([<reflink idref="bib55" id="ref89">55</reflink>]), Schmitz ([<reflink idref="bib49" id="ref90">49</reflink>]), and Bergen and Goodman ([<reflink idref="bib2" id="ref91">2</reflink>]) aims to fill this gap, and holds that focus is best explained as a means of adding minimal redundancy to utterances to combat noise. Focus is thus an approximate solution to an optimization problem, namely, the problem of simultaneously maximizing, on the one hand, communicative success in the face of noise and, on the other hand, economy of time and effort. According to this account, time, effort and noise are crucial components in the development of focus‐marking strategies. Because there is nothing language‐specific about these constraints, the information‐theoretic account is domain‐general and predicts that, in any communication system that faces the same constraints, we should expect to see effort and redundancy distributed to those parts of the message that are most critical to communicative success.</p> <p>This entails a number of specific predictions regarding communicative messages, namely that (a) message length should become shorter when time or effort pressures are increased but may be longer than apparently necessary when pressures are low; (b) redundancy in the message should be distributed disproportionately to the elements of the message that cannot be recovered from context (the critical elements), especially in shorter messages and in the presence of greater noise; (c) effort should be similarly distributed and should increase when noise is increased; and (d) none of these behaviors should sacrifice overall communicative accuracy. Focus in established natural languages can be seen as a fossilized system exhibiting these predicted properties in particular ways.</p> <p>First, while it is not typically thought of as connected with focus, message length varies according to perceive communicative uncertainty; in fact, experimental work on natural language has shown that referring expressions vary substantially in length depending on perceived common ground, without loss of communicative accuracy (Clark, [<reflink idref="bib9" id="ref92">9</reflink>]; Krauss & Weinheimer, [<reflink idref="bib27" id="ref93">27</reflink>]). While there is a potential communicative advantage to providing long detailed messages regardless of common ground, it has long been noted that speakers try not to make their messages substantially longer than necessary (Grice, [<reflink idref="bib18" id="ref94">18</reflink>]). In fact, it should likely be assumed that natural‐language communication operates under quite strong time pressures. Responses to conversational turns in natural language occur with remarkable rapidity, and there is a social pressure against long pauses (Levinson, [<reflink idref="bib30" id="ref95">30</reflink>]). To some extent this may have influenced behavior in our experiments. That is, we explicitly allowed unlimited thinking time before sending messages, but it is notable that our participants still did not take very long. Senders took on average 3.8 s to plan their messages, and a mean of under 9 s on the very first turn, when one might have anticipated that they would take full advantage of their unlimited thinking time. (Receivers took on average 5.9 s to guess, with a mean of 11.9 s on their first turns.) At no point did any Sender take over 46 s to plan her message, and only in 5% of turns across all conditions of both experiments did any Sender take over 10 s.</p> <p>Second, natural‐language focus as laid out in the introduction to this paper involves the application of redundant emphasis to those parts of the sentence that do not overlap with potential alternatives in order to preserve communicative accuracy. Our experiment, with partially overlapping line figures to which redundancy could be applied selectively, provided a clear analogue to this.</p> <p>Third, natural‐language focus often involves increased effort, such as increased prosodic intensity or the addition of morphosyntactic material to the message (which may as a result take longer to produce). This amount of extra effort expended varies according to perceived noise. The construction of more detailed referring expressions can be thought of in the same way (Clark, [<reflink idref="bib9" id="ref96">9</reflink>]; Krauss & Weinheimer, [<reflink idref="bib27" id="ref97">27</reflink>]).</p> <p>Where natural‐language focus might be expected to differ from the behavior of our participants (and from ad hoc responses to noise in communication more generally) is in the flexibility of the response. That is, while we should expect our participants to vary their behavior in response to varying levels of noise, we do not expect the <emph>use</emph> of natural‐language focus systems to vary according to perceived noise (though we might expect the <emph>degree</emph> of redundancy to be varied). We also do not expect our participants, based on the communicative need hypothesis, to restrict focus‐like behavior only to critical elements. These elements should be privileged, but there is no particular reason to expect redundancy to be strictly excluded elsewhere, any more than we should expect the absence of differential redundancy in cases of low noise to be problematic. Such obligatoriness is, rather, expected of grammaticalized systems—products of cultural evolution over generations.</p> <p>The observed behavior in our experiments bear out all of our predictions, including the prediction that focus‐like properties should be observed only under pressure. When no pressures were applied—that is, when communication was easy, with a relaxed time limit and no noise—participants were content to simply trace the entire line figure, making message lengths rather high. This would seem to rule out the possibility that participants were automatically recruiting linguistic resources and mapping the grammatical category of focus onto segments of the line drawing. When the time limit was shortened, or the effort required to fill in a grid cell increased, we saw significant economization of the signal, and not because longer signals were now impossible. As can clearly be seen in Figs. 14 and 8, message length in the High noise and High effort conditions began high and decreased (cf. Krauss & Weinheimer, [<reflink idref="bib27" id="ref98">27</reflink>]). The distribution of redundancy in participants' messages also became focus‐like, but—again—only when there was pressure to economize. In the High effort and High noise conditions, more than in the other conditions, longer messages permitted redundancy in the non‐critical, recoverable grid cells. Also as predicted, increasing noise increased overall effort, as participants aimed to protect their messages against noise by making cells darker and thus less likely to be deleted. Finally, communicative accuracy was broadly constant across conditions; economizing does not sacrifice the ability to communicate.</p> <p>Our results support the notion that the information‐theoretic factors of noise and economy are crucial in the development of focus systems. Participants in our visual communication game quickly developed ad hoc strategies that mirrored the function of linguistic focus. Although different languages will, through cultural evolution, develop different grammatical mechanisms to mark focus, the fact that focus is nearly universally marked across languages suggests that it requires a deeper explanation that appeals not to a universal inventory of syntactic features (though such an inventory may exist), but rather to more general notions of information transmission. In this respect focus may simply be a particular fossilized grammatical consequence of a rather general set of responses to noise and uncertainty in communication (see e.g., Clark, [<reflink idref="bib9" id="ref99">9</reflink>]; Gibson et al., [<reflink idref="bib17" id="ref100">17</reflink>]; Krauss & Weinheimer, [<reflink idref="bib27" id="ref101">27</reflink>]; Levy, [<reflink idref="bib31" id="ref102">31</reflink>]; Levy, Bicknell, Slattery, & Rayner, [<reflink idref="bib32" id="ref103">32</reflink>]).</p> <p>While our results seem most consistent with the communicative need hypothesis and allow us to feel confident in rejecting the walled‐off hypothesis, they remain potentially compatible with the language‐recruitment hypothesis, provided we assume that linguistic resources are recruited only when certain communication pressures are applied. To put it differently, our results are in principle compatible with either of the following explanations of focus:</p> <p></p> <ulist> <item> Focus systems develop naturally as a consequence of the competing pressures of noise amelioration and economization. Therefore, it is only when these pressures are applied that focus‐like systems develop in non‐linguistic communication.</item> <p></p> <item> Focus systems are purely reflexes of an underlying syntactic category which is part of the universal and innate inventory of syntactic categories. However, participants are capable of conceiving of our task as a linguistic one, mapping the category of focus on to line segments in their messages. But crucially, this only occurs when certain communication pressures operate.</item> </ulist> <p>We argue that the latter explanation should be eliminated on the basis of Occam's Razor. While the two explanations are, practically speaking, indistinguishable, it is not at all obvious in the latter case why linguistic resources should fail to be recruited when communication pressures are relaxed. In fact, such a story runs counter to a notion sometimes assumed in such accounts, namely that the human grammatical system evolved for internal thought, and not for communication (see Berwick & Chomsky, [<reflink idref="bib4" id="ref104">4</reflink>]). If we take seriously the notion that syntactic features largely reflect primitives of internal symbolic thought, then our results lead us to reject the notion that focus is one of these primitives, instead arguing that focus is, at its core, part of a set of strategies for improving communication with others.</p> <p>Several questions remain. First, it is important to stress again that the experiment reported here models the emergence of strategies that, over generations, could be expected to develop into grammaticalized focus systems. At that point, we would expect the application of focus to become more rule‐like, and more detached from actual time, effort, and noise pressures. Happily, there are now well–developed experimental tools for investigating such processes (Kirby et al., [<reflink idref="bib25" id="ref105">25</reflink>]), providing a clear path for further experimental work on this question (cf. also Macuch Silva & Roberts, [<reflink idref="bib34" id="ref106">34</reflink>]). A second limitation is theoretical: It is that focus is not the only information‐structural category. For example, Stevens ([<reflink idref="bib55" id="ref107">55</reflink>]), following Selkirk ([<reflink idref="bib52" id="ref108">52</reflink>]), argued for a separate category of givenness (Schwarzschild, [<reflink idref="bib50" id="ref109">50</reflink>]), distinct from focus, which encodes textual salience in the absence of well‐defined alternatives. Stevens ([<reflink idref="bib55" id="ref110">55</reflink>]) also suggests that perhaps givenness, unlike focus, does indeed fulfill the criteria of a syntactic feature in the Chomskyan sense. Future work within this experimental paradigm can assess whether such an argument can find empirical support in non‐linguistic communicative behavior.</p> <hd id="AN0134910149-37">Acknowledgments</hd> <p>The first author's contributions were partly funded by a grant from the German Research Foundation (DFG). Special thanks to Lisa Raithel, Anton Benz, and Robin Clark.</p> <hd id="AN0134910149-38">Appendix Instructions for participants</hd> <p>In what follows [T] and [X] are variables that would be displayed to participants as either "30" or "5" (for [T]) and "15" or "5" (for [X]) depending on condition. All participants in the same condition saw precisely the same instructions, regardless of whether they would be Sender or Receiver on the first round.</p> <hd id="AN0134910149-39">Noise conditions</hd> <p>You live in a galaxy far far away. Along with your partner, located on a separate planet, you are responsible for coordinating traffic through your solar system. Part of this work involves making sure that the correct route‐maps are provided to cargo‐ships. Unfortunately, there has been a system failure! Some of your route‐maps have got confused with each other, and the same has happened to your partner. Luckily, the failure has affected different sets of maps in each case. You have the information to put your partner right, and vice versa. But your communication system is also failing and you don't have much time. You will have to communicate quickly to each other to help choose the correct route maps.</p> <p>This game consists of a series of rounds. In each round, one of you will take the role of sender and the other will take the role of receiver.</p> <p>At the start of a round the sender will see two route‐maps, each of which looks like a line figure drawn over a grid of squares. One of the two (the correct map) will be highlighted in green. Next to these two maps will be an empty grid. The sender's job is to communicate to the receiver which of the two maps is highlighted by filling in squares in the empty grid. It's up to the sender which squares to fill and how many, but there are two important things to keep in mind:</p> <p></p> <ulist> <item> The system failure has left you with a very limited communication channel. Sometimes information will be lost on its way from sender to receiver! The more the sender clicks on a square, the darker it will become, and the more likely it is that it will get through. Note that the receiver will only see black and white squares. If a light‐gray square gets through, it will appear black to the receiver.</item> <p></p> <item> You don't have much time! From the moment the sender clicks inside a square on the grid, they will have only [T] seconds before their turn is up. ONCE THE TIME IS UP NO MORE SQUARES CAN BE FILLED. The message has to be sent in time, or spacecraft will collide.</item> </ulist> <p>Once the sender's turn is over, the receiver will be presented with the sender's completed grid along with the same two maps (though not necessarily in the same order). The receiver's task is to choose which of the two maps the sender was trying to convey.</p> <p>If the receiver chooses correctly, you both score one point (representing a cargo ship that got to the right place). You will then swap roles and start a new round.</p> <p>At the end of the game, if you were successful on at least 80% of rounds, you will win a prize in recognition of your good work under trying circumstances. If you score below 80%, you'll lose your prestigious jobs and have to go back to cleaning cargo holds.</p> <p>Please make sure that you've read all instructions carefully before proceeding. Let the researcher know if you have any questions.</p> <hd id="AN0134910149-40">Effort conditions</hd> <p>You live in a galaxy far far away. Along with your partner, located on a separate planet, you are responsible for coordinating traffic through your solar system. Part of this work involves making sure that the correct route maps are provided to cargo ships. Unfortunately, there has been a system failure! Some of your route maps have got confused with each other, and the same has happened to your partner. Luckily, the failure has affected different sets of maps in each case. You have the information to put your partner right, and vice versa. But your communication system is also failing and you do not have much time. You will have to communicate quickly to each other to help choose the correct route maps.</p> <p>This game consists of a series of rounds. In each round, one of you will take the role of sender and the other will take the role of receiver.</p> <p>At the start of a round, the sender will see two route maps, each of which looks like a line figure drawn over a grid of squares. One of the two (the correct map) will be highlighted in green. Next to these two maps will be an empty grid. The sender's job is to communicate to the receiver which of the two maps is highlighted by filling in squares in the empty grid. It is up to the sender which squares to fill and how many, but there are two important things to keep in mind:</p> <p></p> <ulist> <item> The system failure has left you with a very limited communication channel. To have a square of the grid count as filled you have to click on that square [X] times. Then it will turn black. THE SENDER MUST CLICK ON IT ENOUGH TIMES TO FILL IT. There is no such thing as partially filling a square in the grid.</item> <p></p> <item> You don't have much time! From the moment the sender clicks inside a square on the grid, they will have only [T] seconds before their turn is up. ONCE THE TIME IS UP NO MORE SQUARES CAN BE FILLED. The message has to be sent in time, or spacecraft will collide.</item> </ulist> <p>Once the sender's turn is over, the receiver will be presented with the sender's completed grid along with the same two maps (though not necessarily in the same order). The receiver's task is to choose which of the two maps the sender was trying to convey.</p> <p>If the receiver chooses correctly, you both score one point (representing a cargo ship that got to the right place). You will then swap roles and start a new round.</p> <p>At the end of the game, if you were successful on at least 80% of rounds, you will win a prize in recognition of your good work under trying circumstances. If you score below 80%, you'll lose your prestigious jobs and have to go back to cleaning cargo holds.</p> <p>Please make sure that you've read all instructions carefully before proceeding. Let the researcher know if you have any questions.</p> <ref id="AN0134910149-41"> <title> Footnotes </title> <blist> <bibl id="bib1" idref="ref25" type="bt">1</bibl> <bibtext> See Barner ([1]) for evidence that the linguistic number categories of singular, dual, trial, and plural reflect basic innate categories which are distinct from learned mechanisms involved in more complex number cognition‐like counting and arithmetic.</bibtext> </blist> <blist> <bibl id="bib2" idref="ref31" type="bt">2</bibl> <bibtext> Whether or not fronting requires an extra syntactic operation is, of course, theory dependent, but this is not crucial. The point is that for approaches that require it (including, most obviously, minimalism), the extra operation can be seen as involving greater effort; for approaches (such as construction grammar) that do not, fronting still involves putting the critical element in a position of salience at the expense of other elements. (For an introduction to and comparison of different theoretical approaches to grammar, including minimalism and construction grammar, see Müller, [37].)</bibtext> </blist> <blist> <bibl id="bib3" idref="ref51" type="bt">3</bibl> <bibtext> Observation and mimicry of the prosodic parameters of one's native language in fact begins in the womb (Mampe, Friederici, Christophe, & Wermke, [35]).</bibtext> </blist> <blist> <bibl id="bib4" idref="ref86" type="bt">4</bibl> <bibtext> A famous example of this phenomenon comes from Blaise Pascal, who apologized for the length of a letter, explaining that he had no time to shorten it.</bibtext> </blist> <blist> <bibl id="bib5" idref="ref1" type="bt">5</bibl> <bibtext> Given that the orientation of the two figures was never varied and that the two line figures were never identical, it is important to note that there was always at least one cell that would, on its own, allow the two line figures to be distinguished.</bibtext> </blist> </ref> <ref id="AN0134910149-42"> <title> References </title> <blist> <bibtext> Barner, D. (2017). Language, procedures, and the non‐perceptual origin of number word meanings. Journal of Child Language, 44, 553 – 590.</bibtext> </blist> <blist> <bibtext> Bergen, L., & Goodman, N. D. (2015). The strategic use of noise in pragmatic reasoning. Topics in Cognitive Science, 7, 336 – 350.</bibtext> </blist> <blist> <bibtext> Bergmann, T., Dale, R., & Lupyan, G. (2013) The impact of communicative constraints on the emergence of a graphical communication system. In M. Knau, M. Pauen, N. Sebanz & I. Wachsmuth (Eds.), Proceedings of the Annual Meeting of the Cognitive Science Society (vol. 35, pp. 1887 – 1892).</bibtext> </blist> <blist> <bibtext> Berwick, R. C., & Chomsky, N. (2015). Why only us: Language and evolution. Cambridge, MA : MIT Press.</bibtext> </blist> <blist> <bibtext> Bolinger, D. (1972). Accent is predictable (if you're a mind‐reader). Language, 48, 633 – 644.</bibtext> </blist> <blist> <bibl id="bib6" idref="ref12" type="bt">6</bibl> <bibtext> Borsley, R. D., Tallerman, M., & Willis, D. (2007). The syntax of Welsh. Cambridge, UK : Cambridge University Press.</bibtext> </blist> <blist> <bibl id="bib7" idref="ref26" type="bt">7</bibl> <bibtext> Braver, A., & Bennett, W. G. (2016). Phonotactic c(l)ues to Bantu noun class disambiguation. Linguistics Vanguard, 2, 11 pages.</bibtext> </blist> <blist> <bibl id="bib8" idref="ref45" type="bt">8</bibl> <bibtext> Brennan, S. E., & Clark, H. H. (1996). Conceptual pacts and lexical choice in conversation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22, 1482 – 1493.</bibtext> </blist> <blist> <bibl id="bib9" idref="ref46" type="bt">9</bibl> <bibtext> Clark, H. H. (1996). Using language. Cambridge, UK : Cambridge University Press.</bibtext> </blist> <blist> <bibtext> Culbertson, J., Smolensky, P., & Legendre, G. (2012). Learning biases predict a word order universal. Cognition, 122, 306 – 329.</bibtext> </blist> <blist> <bibtext> Dale, R., & Lupyan, G. (2010). Squiggle: Large‐scale social emergence of simple symbols. In K. Smith, A. Smith, M. Schouwstra, & B. de Boer (Eds.), Proceedings of the 8th Evolution of Language Conference (391 – 392). London : World Scientific.</bibtext> </blist> <blist> <bibtext> Fay, N., Garrod, S., Roberts, L., & Swoboda, N. (2010). The interactive evolution of human communication systems. Cognitive Science, 34, 351 – 386.</bibtext> </blist> <blist> <bibtext> Fedzechkina, M., Jaeger, T. F., & Newport, E. (2016). Balancing effort and information during language acquisition: Evidence from word order and case marking. Cognitive Science, 41, 416 – 446.</bibtext> </blist> <blist> <bibtext> Galantucci, B. (2005). An experimental study of the emergence of human communication systems. Cognitive Science, 29, 737 – 767.</bibtext> </blist> <blist> <bibtext> Galantucci, B. (2009). Experimental semiotics: A new approach for studying communication as a form of joint action. Topics in Cognitive Science, 1, 393 – 410.</bibtext> </blist> <blist> <bibtext> Galantucci, B., Garrod, S., & Roberts, G. (2012). Experimental semiotics. Language and Linguistics Compass, 6, 477 – 493.</bibtext> </blist> <blist> <bibtext> Gibson, E., Bergen, L., & Piantadosi, S. T. (2013). Rational integration of noisy evidence and prior semantic expectations in sentence interpretation. Proceedings of the National Academy of Sciences, 110, 8051 – 8056.</bibtext> </blist> <blist> <bibtext> Grice, H. P. (1975). Logic and conversation. In P. Cole & J. L. Morgan (Eds.), Syntax and semantics. Vol. 3 (pp. 41 – 58). Speech Acts. New York : Academic Press.</bibtext> </blist> <blist> <bibtext> Horn, L. R. (2005). Current issues in neo‐Gricean pragmatics. Intercultural Pragmatics, 2, 191 – 204.</bibtext> </blist> <blist> <bibtext> Horvath, J. (2010). "Discourse‐features," syntactic displacement, and the status of contrast. Lingua, 120, 1346 – 1369.</bibtext> </blist> <blist> <bibtext> Hruska, C., & Alter, K. (2004). Prosody in dialogues and single sentences: How prosody can influence speech perception. In A. Steube (Ed.), Information structure: Theoretical and empirical aspects (pp. 221 – 226). Berlin : Walter de Gruyter.</bibtext> </blist> <blist> <bibtext> Kam, C. L. H., & Newport, E. (2009). Getting it right by getting it wrong: When learners change languages. Cognitive Psychology, 59, 30 – 66.</bibtext> </blist> <blist> <bibtext> Katz, J., & Selkirk, E. (2011). Contrastive focus vs. discourse‐new: Evidence from phonetic prominence in english. Language, 87, 771 – 816.</bibtext> </blist> <blist> <bibtext> Kirby, S., Cornish, H., & Smith, K. (2008). Cumulative cultural evolution in the laboratory: An experimental approach to the origins of structure in human language. Proceedings of the National Academy of Sciences, 105, 10681 – 10686.</bibtext> </blist> <blist> <bibtext> Kirby, S., Griffiths, T., & Smith, K. (2014). Iterated learning and the evolution of language. Current Opinion in Neurobiology, 28, 108 – 114.</bibtext> </blist> <blist> <bibtext> Kirby, S., Tamariz, M., Cornish, H., & Smith, K. (2015). Compression and communication in the cultural evolution of linguistic structure. Cognition, 141, 87 – 102.</bibtext> </blist> <blist> <bibtext> Krauss, R. M., & Weinheimer, S. (1964). Changes in reference phrases as a function of frequency of usage in social interaction a preliminary study. Psychonomic Science, 1, 113 – 114.</bibtext> </blist> <blist> <bibtext> Krifka, M. (2008). Basic notions of information structure. Acta Linguistica Hungarica, 55, 243 – 276.</bibtext> </blist> <blist> <bibtext> Ladd, D. R. (1996). Intonational phonology. New York : Cambridge University Press.</bibtext> </blist> <blist> <bibtext> Levinson, S. (2016). Turn‐taking in human communication–origins and implications for language processing. Trends in Cognitive Sciences, 20, 6 – 14.</bibtext> </blist> <blist> <bibtext> Levy, R. (2008). A noisy‐channel model of rational human sentence comprehension under uncertain input. In M. Lapata & H. T. Ng (Eds.), Proceedings of the 13th Conference on Empirical Methods in Natural Language Processing (pp. 234 – 243). Waikiki, Honolulu : Association for Computational Linguistics.</bibtext> </blist> <blist> <bibtext> Levy, R., Bicknell, K., Slattery, T., & Rayner, K. (2009). Eye movement evidence that readers maintain and act on uncertainty about past linguistic input. Proceedings of the National Academy of Sciences, 106, 21086 – 21090.</bibtext> </blist> <blist> <bibtext> Loprieno, A. (1995). Ancient Egyptian: A linguistic introduction. Cambridge : Cambridge University Press.</bibtext> </blist> <blist> <bibtext> Macuch Silva, V., & Roberts, S. (2016). Language adapts to signal disruption in interaction. In S. Roberts, C. Cuskley, L. McCrohon, L. Barceló‐Coblijn, O. Fehér & T. Verhoef (Eds.), The Evolution of Language: Proceedings of the 11th International Conference (EVOLANG11). <ulink href="http://evolang.org/neworleans/papers/20.html">http://evolang.org/neworleans/papers/20.html</ulink>.</bibtext> </blist> <blist> <bibtext> Mampe, B., Friederici, A. D., Christophe, A., & Wermke, K. (2009). Newborns' cry melody is shaped by their native language. Current biology, 19, 1994 – 1997.</bibtext> </blist> <blist> <bibtext> Micklos, A. (2016) Interaction for facilitating conventionalization: Negotiating the silent gesture communication of noun verb pairs. In S. G. Roberts, C. Cuskley, L. McCrohon, L. Barceló‐Coblijn, O. Fehér, & T. Verhoef (Eds.), The evolution of language: Proceedings of the 11th International Conference (EVOLANG11). <ulink href="http://evolang.org/neworleans/papers/143.html">http://evolang.org/neworleans/papers/143.html</ulink>.</bibtext> </blist> <blist> <bibtext> Müller, S. (2016). Grammatical theory: From transformational grammar to constraint‐based approaches. Berlin : Language Science Press.</bibtext> </blist> <blist> <bibtext> Neal, D. T., Wood, W., & Quinn, J. M. (2006). Habits: A repeat performance. Current Directions in Psychological Science, 15 (198), 202.</bibtext> </blist> <blist> <bibtext> Rizzi, L. (1997) The fine structure of the left periphery. In L. Haegeman (Ed.), Elements of grammar (pp. 281 – 337). Dordrecht, the Netherlands : Springer.</bibtext> </blist> <blist> <bibtext> Rizzi, L., & Cinque, G. (2008). The cartography of syntactic structures. In V. Moscati (Ed.), CISCL working papers on language and cognition (vol. 2, pp. 43 – 59). Siena : Centro Interdipartimentale di Studi Cognitivi sul Linguaggio.</bibtext> </blist> <blist> <bibtext> Roberts, C. (1996). Information Structure: Towards an integrated formal theory of pragmatics. In J. Yoon & A. Kathol (Eds.), OSU WPL volume 49: Papers in semantics (pp. 91 – 136). Columbus : The Ohio State University Department of Linguistics.</bibtext> </blist> <blist> <bibtext> Roberts, G. (2010). An experimental study of the role of social selection and frequency of interaction in linguistic diversity. Interaction Studies, 11, 138 – 159.</bibtext> </blist> <blist> <bibtext> Roberts, G. (2017). The linguist's Drosophila: Experiments in language change. Linguistics Vanguard, 3, 20160086.</bibtext> </blist> <blist> <bibtext> Roberts, G., & Fedzechkina, M. (2018). Social biases modulate the loss of redundant forms in the cultural evolution of language. Cognition, 171C, 194 – 201.</bibtext> </blist> <blist> <bibtext> Roberts, G., Lewandowski, J., & Galantucci, B. (2015). How communication changes when we cannot mime the world: Experimental evidence for the effect of iconicity on combinatoriality. Cognition, 141, 52 – 66.</bibtext> </blist> <blist> <bibtext> Rochemont, M. S. (1986) Focus in generative grammar. Amsterdam : John Benjamins.</bibtext> </blist> <blist> <bibtext> Romaine, S. (1984). The language of children and adolescents: The acquisition of communicative competence. Oxford : Basil Blackwell.</bibtext> </blist> <blist> <bibtext> Rooth, M. (1992). A theory of focus interpretation. Natural Language Semantics, 1, 75 – 116.</bibtext> </blist> <blist> <bibtext> Schmitz, H.‐C. (2008). Accentuation and interpretation. New York : Palgrave MacMillan.</bibtext> </blist> <blist> <bibtext> Schwarzschild, R. (1999). GIVENness, AvoidF and other constraints on the placement of accent. Natural Language Semantics, 7, 141 – 177.</bibtext> </blist> <blist> <bibtext> Scott‐Phillips, T. C., Kirby, S., & Ritchie, G. R. S. (2009). Signalling signalhood and the emergence of communication. Cognition, 113, 226 – 233.</bibtext> </blist> <blist> <bibtext> Selkirk, E. (2008). Contrastive focus, givenness and the unmarked status of "discourse‐new." Acta Linguistica Hungarica, 55, 331 – 346.</bibtext> </blist> <blist> <bibtext> Shannon, C. E., & Weaver, W. (1949). The mathematical theory of communication. Champaign : University of Illinois Press.</bibtext> </blist> <blist> <bibtext> Sneller, B., & Roberts, G. (2018). Why some behaviors spread while others don't: A laboratory simulation of dialect contact. Cognition, 170C, 298 – 311.</bibtext> </blist> <blist> <bibtext> Stevens, J. (2016a). Focus games. Linguistics and Philosophy, 39, 395 – 441.</bibtext> </blist> <blist> <bibtext> Stevens, J. (2016b) A signaling account of contrastive focus. In N. Bade, P. Berezovskaya & A. Schöller (Eds.), Proceedings of Sinn und Bedeutung 20 (pp. 694 – 711). Düsseldorf : Gesellschaft für Semantik.</bibtext> </blist> <blist> <bibtext> Stevens, J. (2017). Pragmatics of focus. In M. Aronoff (Ed.), Oxford research encyclopedia of linguistics. Oxford, UK : Oxford University Press.</bibtext> </blist> <blist> <bibtext> Tamariz, M., & Kirby, S. (2016). The cultural evolution of language. Current Opinion in Psychology, 8, 37 – 43.</bibtext> </blist> <blist> <bibtext> Traugott, E. C., & Dasher, R. B. (2001). Regularity in semantic change, vol. 97. Cambridge, UK : Cambridge University Press.</bibtext> </blist> <blist> <bibtext> Trendafilov, D., Lemmelä, S., & Murray‐Smith, R. (2010). Negotiation models for mobile tactile interaction. In Roderick Murray‐Smith (Ed.), International workshop on mobile social signal processing. Berlin : Springer‐Verlag.</bibtext> </blist> <blist> <bibtext> Verhoef, T., Kirby, S., & de Boer, B. (2014). Emergence of combinatorial structure and economy through iterated learning with continuous acoustic signals. Journal of Phonetics, 43C, 57 – 68.</bibtext> </blist> <blist> <bibtext> Wilson, D., & Sperber, D. (2004). Relevance theory. In L. R. Horn, & G. Ward (Eds.), Handbook of pragmatics (pp. 607 – 632). Oxford, UK : Wiley Blackwell.</bibtext> </blist> <blist> <bibtext> Zimmermann, M. (2008). Contrastive Focus and emphasis. Acta Linguistica Hungarica, 55, 347 – 360.</bibtext> </blist> <blist> <bibtext> Zimmermann, M., & Onea, E. (2011). Focus marking and focus interpretation. Lingua, 121, 1651 – 1670.</bibtext> </blist> </ref> <aug> <p>By Jon S. Stevens and Gareth Roberts</p> <p>Reported by Author; Author</p> </aug> <nolink nlid="nl1" bibid="bib41" firstref="ref2"></nolink> <nolink nlid="nl2" bibid="bib48" firstref="ref3"></nolink> <nolink nlid="nl3" bibid="bib49" firstref="ref4"></nolink> <nolink nlid="nl4" bibid="bib57" firstref="ref5"></nolink> <nolink nlid="nl5" bibid="bib18" firstref="ref8"></nolink> <nolink nlid="nl6" bibid="bib62" firstref="ref9"></nolink> <nolink nlid="nl7" bibid="bib23" firstref="ref10"></nolink> <nolink nlid="nl8" bibid="bib29" firstref="ref11"></nolink> <nolink nlid="nl9" bibid="bib20" firstref="ref13"></nolink> <nolink nlid="nl10" bibid="bib64" firstref="ref14"></nolink> <nolink nlid="nl11" bibid="bib33" firstref="ref15"></nolink> <nolink nlid="nl12" bibid="bib63" firstref="ref17"></nolink> <nolink nlid="nl13" bibid="bib21" firstref="ref18"></nolink> <nolink nlid="nl14" bibid="bib46" firstref="ref19"></nolink> <nolink nlid="nl15" bibid="bib40" firstref="ref20"></nolink> <nolink nlid="nl16" bibid="bib39" firstref="ref21"></nolink> <nolink nlid="nl17" bibid="bib55" firstref="ref23"></nolink> <nolink nlid="nl18" bibid="bib56" firstref="ref24"></nolink> <nolink nlid="nl19" bibid="bib53" firstref="ref27"></nolink> <nolink nlid="nl20" bibid="bib31" firstref="ref32"></nolink> <nolink nlid="nl21" bibid="bib17" firstref="ref33"></nolink> <nolink nlid="nl22" bibid="bib30" firstref="ref36"></nolink> <nolink nlid="nl23" bibid="bib19" firstref="ref37"></nolink> <nolink nlid="nl24" bibid="bib28" firstref="ref44"></nolink> <nolink nlid="nl25" bibid="bib27" firstref="ref47"></nolink> <nolink nlid="nl26" bibid="bib38" firstref="ref49"></nolink> <nolink nlid="nl27" bibid="bib47" firstref="ref50"></nolink> <nolink nlid="nl28" bibid="bib25" firstref="ref52"></nolink> <nolink nlid="nl29" bibid="bib58" firstref="ref53"></nolink> <nolink nlid="nl30" bibid="bib59" firstref="ref54"></nolink> <nolink nlid="nl31" bibid="bib26" firstref="ref56"></nolink> <nolink nlid="nl32" bibid="bib34" firstref="ref57"></nolink> <nolink nlid="nl33" bibid="bib15" firstref="ref58"></nolink> <nolink nlid="nl34" bibid="bib24" firstref="ref59"></nolink> <nolink nlid="nl35" bibid="bib54" firstref="ref60"></nolink> <nolink nlid="nl36" bibid="bib12" firstref="ref61"></nolink> <nolink nlid="nl37" bibid="bib14" firstref="ref62"></nolink> <nolink nlid="nl38" bibid="bib45" firstref="ref63"></nolink> <nolink nlid="nl39" bibid="bib10" firstref="ref64"></nolink> <nolink nlid="nl40" bibid="bib13" firstref="ref65"></nolink> <nolink nlid="nl41" bibid="bib22" firstref="ref66"></nolink> <nolink nlid="nl42" bibid="bib44" firstref="ref70"></nolink> <nolink nlid="nl43" bibid="bib16" firstref="ref72"></nolink> <nolink nlid="nl44" bibid="bib43" firstref="ref73"></nolink> <nolink nlid="nl45" bibid="bib42" firstref="ref76"></nolink> <nolink nlid="nl46" bibid="bib11" firstref="ref78"></nolink> <nolink nlid="nl47" bibid="bib36" firstref="ref81"></nolink> <nolink nlid="nl48" bibid="bib61" firstref="ref82"></nolink> <nolink nlid="nl49" bibid="bib60" firstref="ref83"></nolink> <nolink nlid="nl50" bibid="bib51" firstref="ref85"></nolink> <nolink nlid="nl51" bibid="bib32" firstref="ref103"></nolink> <nolink nlid="nl52" bibid="bib52" firstref="ref108"></nolink> <nolink nlid="nl53" bibid="bib50" firstref="ref109"></nolink>
Header DbId: eric
DbLabel: ERIC
An: EJ1206726
AccessLevel: 3
PubType: Academic Journal
PubTypeId: academicJournal
PreciseRelevancyScore: 0
IllustrationInfo
Items – Name: Title
  Label: Title
  Group: Ti
  Data: Noise, Economy, and the Emergence of Information Structure in a Laboratory Language
– Name: Language
  Label: Language
  Group: Lang
  Data: English
– Name: Author
  Label: Authors
  Group: Au
  Data: <searchLink fieldCode="AR" term="%22Stevens%2C+Jon+S%2E%22">Stevens, Jon S.</searchLink><br /><searchLink fieldCode="AR" term="%22Roberts%2C+Gareth%22">Roberts, Gareth</searchLink>
– Name: TitleSource
  Label: Source
  Group: Src
  Data: <searchLink fieldCode="SO" term="%22Cognitive+Science%22"><i>Cognitive Science</i></searchLink>. Feb 2019 43(2).
– Name: Avail
  Label: Availability
  Group: Avail
  Data: Wiley-Blackwell. 350 Main Street, Malden, MA 02148. Tel: 800-835-6770; Tel: 781-388-8598; Fax: 781-388-8232; e-mail: cs-journals@wiley.com; Web site: http://www.wiley.com/WileyCDA
– Name: PeerReviewed
  Label: Peer Reviewed
  Group: SrcInfo
  Data: Y
– Name: Pages
  Label: Page Count
  Group: Src
  Data: 33
– Name: DatePubCY
  Label: Publication Date
  Group: Date
  Data: 2019
– Name: TypeDocument
  Label: Document Type
  Group: TypDoc
  Data: Journal Articles<br />Reports - Research
– Name: Subject
  Label: Descriptors
  Group: Su
  Data: <searchLink fieldCode="DE" term="%22Communication+%28Thought+Transfer%29%22">Communication (Thought Transfer)</searchLink><br /><searchLink fieldCode="DE" term="%22Time%22">Time</searchLink><br /><searchLink fieldCode="DE" term="%22Difficulty+Level%22">Difficulty Level</searchLink><br /><searchLink fieldCode="DE" term="%22Information+Theory%22">Information Theory</searchLink><br /><searchLink fieldCode="DE" term="%22Behavior+Patterns%22">Behavior Patterns</searchLink>
– Name: DOI
  Label: DOI
  Group: ID
  Data: 10.1111/cogs.12717
– Name: ISSN
  Label: ISSN
  Group: ISSN
  Data: 0364-0213
– Name: Abstract
  Label: Abstract
  Group: Ab
  Data: The acceptability of sentences in natural language is constrained not only grammaticality, but also by the relationship between what is being conveyed and such factors as context and the beliefs of interlocutors. In many languages the critical element in a sentence (its focus) must be given grammatical prominence. There are different accounts of the nature of focus marking. Some researchers treat it as the grammatical realization of a potentially arbitrary feature of universal grammar and do not provide an explicit account of its origins; others have argued, however, that focus marking is a (grammaticalized) functional solution to the problem of efficiently transmitting information via a noisy channel. By adding redundancy to highlight critical elements in particular, focus protects key parts of the message from noise. If this information-theoretic account is true, then we should expect focus-like behavior to emerge even in non-linguistic communication systems given sufficient noise and pressures for efficiency. We tested this in an experiment in which participants played a simple communication game in which they had to click cells on a grid to communicate one of two line figures drawn across the grid. We manipulated the noise, available time, and required effort, and measured patterns of redundancy. Because the lines in many cases overlapped, meaning that only some parts of each line could be used to distinguish it from the other, we were able to compare the extent to which effort was expended on adding redundancy to critical (non-overlapping) and non-critical (overlapping) parts of the message. The results supported the information-theoretic account of focus and shed light on the emergence of information structure in language.
– Name: AbstractInfo
  Label: Abstractor
  Group: Ab
  Data: As Provided
– Name: DateEntry
  Label: Entry Date
  Group: Date
  Data: 2019
– Name: AN
  Label: Accession Number
  Group: ID
  Data: EJ1206726
PLink https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=eric&AN=EJ1206726
RecordInfo BibRecord:
  BibEntity:
    Identifiers:
      – Type: doi
        Value: 10.1111/cogs.12717
    Languages:
      – Text: English
    PhysicalDescription:
      Pagination:
        PageCount: 33
    Subjects:
      – SubjectFull: Communication (Thought Transfer)
        Type: general
      – SubjectFull: Time
        Type: general
      – SubjectFull: Difficulty Level
        Type: general
      – SubjectFull: Information Theory
        Type: general
      – SubjectFull: Behavior Patterns
        Type: general
    Titles:
      – TitleFull: Noise, Economy, and the Emergence of Information Structure in a Laboratory Language
        Type: main
  BibRelationships:
    HasContributorRelationships:
      – PersonEntity:
          Name:
            NameFull: Stevens, Jon S.
      – PersonEntity:
          Name:
            NameFull: Roberts, Gareth
    IsPartOfRelationships:
      – BibEntity:
          Dates:
            – D: 01
              M: 02
              Type: published
              Y: 2019
          Identifiers:
            – Type: issn-print
              Value: 0364-0213
          Numbering:
            – Type: volume
              Value: 43
            – Type: issue
              Value: 2
          Titles:
            – TitleFull: Cognitive Science
              Type: main
ResultId 1