The analysis of group decisions is an active field of research whose importance increases due to higher connectivity in society. This paper analyzes deliberative and aggregative models of group decision making. Weaknesses and advantages of network epistemology and judgement aggregation are presented in detail. Furthermore combinations of both approaches are discussed since both models for themselves cannot account for real world group decision making accurately. In sum the paper tries to contribute to better understanding group communication, which in turn can be applied to various fields of research. Entscheidungsfindung in Gruppen ist ein viel diskutiertes Forschungsgebiet, das durch die fortschreitende Vernetzung der Welt immer mehr an Bedeutung gewinnt. In dieser Arbeit werden deliberative und aggregative Entscheidungsfindungsmodelle analysiert. Vorteile und Schwächen der Theorien "network epistemology" und "judgement aggregation" werden im Detail vorgestellt. Zudem werden Kombinationen beider Modelle diskutiert, da jede Theorien für sich genommen nicht in der Lage ist, Gruppenentscheidungen akkurat zu beschreiben. Zusammengefasst ist diese Arbeit ein Beitrag zum Verständnis von Gruppenkommunikation, die ihrerseits in den verschiedenen Forschungsgebieten Anwendung findet.

This index covers all technical items—papers, correspondence, reviews, etc.—that appeared in this periodical during 2016, and items from previous years that were commented upon or corrected in 2016. Departments and other items may also be covered if they have been judged to have archival value. The Author Index contains the primary entry for each item, listed under the first author’s name. The primary entry includes the coauthors’ names, the title of the paper or other item, and its location, specified by the publication abbreviation, year, month, and inclusive pagination. The Subject Index contains entries describing the item under all appropriate subject headings, plus the first author’s name, the publication abbreviation, month, and year, and inclusive pages. Note that the item title is found only under the primary entry in the Author Index.

Out-of-vocabulary (OOV) keywords present a challenge for keyword search (KWS) systems especially in the low-resource setting. Previous research has centered around approaches that use a variety of subword units to recover OOV words. This paper systematically investigates morphology-based subword modeling approaches on seven low-resource languages. We show that using morphological subword units (morphs) in speech recognition decoding is substantially better than expanding word-decoded lattices into subword units including phones, syllables and morphs. As alternatives to grapheme-based morphs, we apply unsupervised morphology learning to sequences of phonemes, graphones, and syllables. Using one of these phone-based morphs is almost always better than using the grapheme-based morphs, but the particular choice varies with the language. By combining the different methods, a substantial gain is obtained over the best single case for all languages, especially for OOV performance.

Prior studies on online sentence processing have shown that the parser can resolve non-local dependencies rapidly and accurately. This study investigates the interaction between the processing of two such non-local dependencies: wh-filler-gap dependencies (WhFGD) and reflexive-antecedent dependencies. We show that reflexive-antecedent dependency resolution is sensitive to the presence of a WhFGD, and argue that the filler-gap dependency established by WhFGD resolution is selected online as the antecedent of a reflexive dependency. We investigate the processing of constructions like (1), where two NPs might be possible antecedents for the reflexive, namely which cowgirl and Mary. Even though Mary is linearly closer to the reflexive, the only grammatically licit antecedent for the reflexive is the more distant wh-NP, which cowgirl. (1). Which cowgirl did Mary expect to have injured herself due to negligence? Four eye-tracking text-reading experiments were conducted on examples like (1), differing in whether the embedded clause was non-finite (1 and 3) or finite (2 and 4), and in whether the tail of the wh-dependency intervened between the reflexive and its closest overt antecedent (1 and 2) or the wh-dependency was associated with a position earlier in the sentence (3 and 4). The results of Experiments 1 and 2 indicate the parser accesses the result of WhFGD formation during reflexive antecedent search. The resolution of a wh-dependency alters the representation that reflexive antecedent search operates over, allowing the grammatical but linearly distant antecedent to be accessed rapidly. In the absence of a long-distance WhFGD (Experiments 3 and 4), wh-NPs were not found to impact reading times of the reflexive, indicating that the parser's ability to select distant wh-NPs as reflexive antecedents crucially involves syntactic structure.

In current approaches to pragmatic reasoning the comprehension and production of referring expressions is modeled as a result of the interlocutors’ mutual perspective-taking. While such models of pragmatic reasoning have been empirically validated in referential language games experiments, empirical (and computational) work on the generation of referring expressions has shown that speakers do not always take the listener’s perspective into account, but instead produce referring expressions according to their own preferences. One particularly well studied example is color: speakers often include color terms in their referring expressions even if they do not help identify the intended referent. We show that like speakers, listeners treat color differently from other properties like e.g. size. Our results suggest that listeners do not seem to perform much pragmatic reasoning when the referring expression only expresses color, but instead follow a simple saliencebased heuristic.

For languages with fast vocabulary growth and limited resources, data sparsity leads to challenges in training a language model. One strategy for addressing this problem is to leverage morphological structure as features in the model. This paper explores different uses of unsupervised morphological features in both the history and prediction space for three word-based exponential models (maximum entropy, logbilinear, and recurrent neural net (RNN)). Multi-task training is introduced as a regularizing mechanism to improve performance in the continuous-space approaches. The models are compared to non-parametric baselines. From using the RNN with morphological features and multi-task learning, experiments with conversational speech from four languages show we can obtain consistent gains of 7-11% in perplexity reduction in a limited-resource scenario (10 hrs speech), and 12-18% when the training size is increased (80 hrs). Results are mixed for all other approaches, compared to a modified Kneser-Ney baseline, but morphology is useful in continuous-space models compared to their word-only baseline. Multi-task learning improves both continuous-space models.

Parser evaluation traditionally relies on evaluation metrics which deliver a single aggregate score over all sentences in the parser output, such as PARSEVAL. However, for the evaluation of parser performance concerning a particular phenomenon, a test suite of sentences is needed in which this phenomenon has been identified. In recent years, the parsing of discontinuous structures has received a rising interest. Therefore, in this paper, we present a test suite for testing the performance of dependency and constituency parsers on non-projective dependencies and discontinuous constituents for German. The test suite is based on the newly released TIGER treebank version 2.2. It provides a unique possibility of benchmarking parsers on non-local syntactic relationships in German, for constituents and dependencies. We include a linguistic analysis of the phenomena that cause discontinuity in the TIGER annotation, thereby closing gaps in previous literature. The linguistic phenomena we investigate include extraposition, a placeholder/repeated element construction, topicalization, scrambling, local movement, parentheticals, and fronting of pronouns.

Overspecification and the Cost of Pragmatic Reasoning about Referring Expressions Peter Baumann (baumann@u.northwestern.edu) Department of Linguistics Northwestern University, Evanston, IL, USA Brady Clark (bzack@northwestern.edu) Department of Linguistics Northwestern University, Evanston, IL, USA Stefan Kaufmann (stefan.kaufmann@uconn.edu) Department of Linguistics University of Connecticut, Storrs, CT, USA Abstract In current approaches to pragmatic reasoning the comprehen- sion and production of referring expressions is modeled as a result of the interlocutors’ mutual perspective-taking under the additional assumption that speakers try to minimize their ar- ticulatory effort or production cost. The latter assumption is usually not tested and instead built into the experimental tasks of referential language games by artificially restricting the set of possible referring expressions available to identify a refer- ent. We present two language game experiments: a produc- tion experiment, in which the speakers were allowed to freely choose a referring expression, and a comprehension experi- ment to replicate earlier findings with our stimuli. Our results show that while listeners easily perform pragmatic reasoning, speakers resort to overspecification when the effort of prag- matic reasoning becomes too high. Keywords: Pragmatics; Language games; Referring Expres- sions; Language Production; Language Comprehension Introduction In a complex situation, like a room full of people, one may be able to identify a single person upon hearing a rather short de- scription, like “the man with a hat”, even in cases in which the man in question is not the only one with a hat and despite the fact that he certainly has many other characteristics besides wearing a hat. This amazing efficiency of human communi- cation has made reference and the use of referring expressions a central topic in linguistic pragmatics. The most influen- tial descriptive account for the efficiency of human commu- nication was formulated by Grice (1975) in the form of his Maxims of Conversation, and since then, a number of propos- als have been made to provide more quantitative models of pragmatic reasoning based on the Gricean maxims or more general principles of human cognition and interaction. The most prominent examples are game-theoretic models based on strategic reasoning (e.g. Benz & Van Rooij, 2007; Franke, 2011; J¨ager, 2011) and Bayesian models grounded in so- cial cognition (e.g. Frank & Goodman, 2012; Goodman & Stuhlm¨uller, 2013). In both classes of models it is assumed that the speaker and hearer reason about each other’s perspec- tives: the hearer is assumed to interpret a speaker’s expression as referring to the referent for which the expression is ‘opti- mal’ under the perspective of the speaker, who in turn chooses the referring expression to be ‘optimal’ under the hearer’s per- spective, etc. A trivial solution to this recursive reasoning process is for the speaker to choose a referring expression that explicitly mentions all features of the intended referent and is thus absolutely unambiguous in the given context. Since such an expression can hardly qualify as efficient, however, the above models make the crucial additional assumption that the speaker incurs a cost for producing an utterance, thus all things being equal, speakers have a preference for the most economic (i.e. shortest and least effortful) expression. While theoretically appealing, production costs are noto- riously hard to quantify, as the articulatory effort of speech production 1 is negligible (Moon & Lindblom, 2003; Locke, 2008). And anecdotally one may even be tempted to reject the notion of production costs altogether: people talk a lot. However, we do not argue against a possible role for pro- duction costs in pragmatic reasoning. Instead, we show that speakers’ behavior cannot be explained in terms of this fac- tor alone: Under certain conditions, speakers do use costlier forms than would be required to identify the intended refer- ent, which suggests that the process of pragmatic reasoning (often termed implicature) is itself effortful (like any other reasoning process) and thus incurs a cost for the speaker. In language comprehension, it is well established that prag- matic reasoning can be effortful: in reading, sentences involv- ing implicatures take longer to process than sentences with- out implicatures (Hamblin & Gibbs, 2003), and in referential language games, target identification is less accurate when the message involves an implicature than when it does not (e.g. Degen & Franke, 2012). In language production, on the other hand, the speaker has the option of reducing the need for pragmatic inference by overspecification, i.e. saying more than is strictly necessary. And indeed, there is plenty of empirical evidence that speak- ers do make use of overspecification. In particular, research on the production of referring expressions (e.g. Koolen, Typological analyses of language efficiency (e.g. Piantadosi, Tily, & Gibson, 2012) are typically based on arguments involving processing restrictions in comprehension rather than production.

Dependencies and Hierarchical Structure in Sentence Processing Peter Baumann (baumann@u.northwestern.edu) Department of Linguistics Northwestern University, Evanston, IL, USA Abstract 1997), i.e. the length of the path in the phrase-structure tree to ‘travel’ from one element to the other. In this paper, we evaluate these three measures of distance between linguistic dependents. We will assume that the num- ber of words intervening between two dependent elements is a good approximation of the time of memory decay so that the three distance measures can be summarizes as follows: • L INEAR D ISTANCE : number of words between two depen- dents • DLT D ISTANCE : number of discourse referents 1 between two dependents • S TRUCTURAL D ISTANCE : number of syntactic nodes crossed in the syntactic phrase-structure tree between two dependents Following a recent trend in psycholinguistics (e.g. Pynte, New, & Kennedy, 2008; Demberg & Keller, 2008), we do not focus on specific constructions or sentence types to evaluate the three distance measures, but instead employ a regression analysis on an ‘eye-tracking corpus’, i.e. eye-tracking data of people reading naturally occurring texts, to determine if dependency integration cost, when determined by one of the three distance measures is a significant predictor of reading times. We will show that only structural distance between two de- pendent elements is a significant predictor of reading times, and that this effect can only be found on verbs, but not on nouns. More importantly, our analyses show that a higher dis- tance between two dependent elements leads to lower reading times. This result is the exact opposite of what most memory- based models predict, but it is not unprecedented: in a self- paced reading experiment conducted in German, Konieczny (2000) observed that verbs were read faster when the number of intervening words between them and their arguments was higher. Similar anti-locality effects have since been observed in controlled experiments in Japanese (Nakatani & Gibson, 2008) and Hindi (Vasishth & Lewis, 2006). And in an analy- sis very similar to ours, Demberg and Keller (2008) found a similar negative effect of DLT distance, i.e. higher distance led to lower reading times. Our results can be interpreted in different ways: one may argue that that dependency integration can lead to integra- tion costs, but that dependencies may also help to predict and thus facilitate the upcoming head, and that the latter process is more common than the former in everyday language com- prehension. A somewhat weaker conclusion is that current memory-based models, such as DLT, may be too narrow in In memory-based models of human sentence processing it is assumed that the completion of a dependency between a syn- tactic head and its dependents is a major source of processing difficulty in non-ambiguous sentences, and that this integration cost is a function of the distance between the two elements. However, it remains open how to measure the distance between two dependent elements. While many current models employ a linear distance measure, we instead propose to measure the distance between a head and its dependents as the path in the phrase structure tree connecting the two elements. We evalu- ate this structural distance as a measure of dependency inte- gration and show that it is a better predictor of human reading times than other measures. Moreover, we find that evaluated on reading data from naturally occurring texts, dependency inte- gration is not actually a cost, as higher dependency integration distances led to lower reading times. Keywords: sentence processing; syntax; reading; eye- tracking; modeling; memory Introduction How do we understand language and what are the cogni- tive mechanisms reflected in measures of human sentence comprehension? Many attempts to answer this big ques- tion can be subsumed under two broad categories: memory- based models and experience-based models. One particular aspect, in which these two modeling approaches differ, is the role they assign to hierarchical syntactic structure: while experience-based models of syntactic expectations are of- ten built on phrase-structure grammars modeling hierarchical sentence structure, memory-based models tend to be agnos- tic about it, as they are usually formulated in terms of non- hierarchical dependency structures. Within the latter group of models it is a common assump- tion that the completion of a dependency between a syntactic head and its dependents (such as e.g. a verb and its argument) is a major source of processing difficulty in non-ambiguous sentences (Gibson, 1998), and that this integration cost is – at least in part – a function of the distance between the two elements. However, specific models differ with regard to the ques- tion of how to measure the distance between two dependent linguistic units: some models emphasize the role of work- ing memory decay and thus assume the distance to be mea- sured in time (Wanner & Maratsos, 1978; King & Just, 1991), others, like the Dependency Locality Theory (DLT; Gib- son, 1998, 2000) emphasize the general capacity limitations of working memory and measure the distance in terms of intervening discourse referents, which are supposed to oc- cupy memory slots. A third possible alternative, which has not been implemented in memory-based models, is struc- tural distance within the phrase structure (see e.g. O’Grady, 1 We follow (Gibson, 1998) and assume that nouns and verbs in- troduce discourse referents.

Like many other processes in language comprehension, anaphora resolution is determined by what is said. But is this all? Or to what extent is anaphora resolution also influenced by what is not said but could have been said? We present a questionnaire, a self-paced reading study and a corpus analysis, suggesting that the existence of possible alternative constructions and referring expressions helps to constitute preferences for anaphora in referentially ambiguous sentences and also affects online sentence processing. These disambiguating effects may be understood as conversational implicatures licensed by pragmatic principles.

The world-wide proliferation of digital communications has created the need for language and speech processing systems for under-resourced languages. Developing such systems is challenging if only small data sets are available, and the problem is exacerbated for languages with highly productive morphology. However, many under-resourced languages are spoken in multi-lingual environments together with at least one resource-rich language and thus have numerous borrowings from resource-rich languages. Based on this insight, we argue that readily available resources from resource-rich languages can be used to bootstrap the morphological analyses of under-resourced languages with complex and productive morphological systems. In a case study of two such languages, Tagalog and Zulu, we show that an easily obtainable English wordlist can be deployed to seed a morphological analysis algorithm from a small training set of conversational transcripts. Our method achieves a precision of 100% and identifies 28 and 66 of the most productive affixes in Tagalog and Zulu, respectively.

This work compares ASR decoding at different subword levels crossed with alternative keyword search strategies to handle the OOV issue for keyword spotting in the low-resource setting. We show that a morpheme-based subword modeling approach is effective in recovering OOV keywords within a Turkish low-resource keyword spotting task, where mixed word and morpheme decoding approach outperforms the traditional subword-based search from word-decoded lattices that are broken down to subword lattices. Furthermore, unsupervised learning of morphology works almost as well as a rule-based system designed for the language despite the low-resource condition. A staged keyword search strategy benefits from both methods of morphological analysis.

Syntactic category disambiguation within an architecture of human language processing Peter Baumann (baumann@u.northwestern.edu) Northwestern University Department of Linguistics, 2016 Sheridan Road Evanston, IL 60208, USA Abstract for constraint-based models can also be accounted for un- der a modular architecture with a module for bottom-up syntactic category assignment. In this paper, we follow Corley and Crocker’s proposal and provide further evi- dence for the existence of a syntactic category module by showing that Corley and Crocker’s model of syntac- tic category disambiguation is a significant predictor of reading times in naturally occurring texts. In addition, we provide evidence that syntactic category disambigua- tion may be independent of syntactic top-down expec- tations, emphasizing the critical role of bottom-up pro- cesses within a modular architecture of human language processing. Syntactic category ambiguities are very frequent in nat- ural languages, and all architectures of language process- ing need a mechanism for disambiguating syntactic cat- egory ambiguities. Corley and Crocker (2000) suggested that syntactic category disambiguation can be assigned its own module within a modular architecture. We will show that the model defined by Corley and Crocker can account for a considerable amount of variance in read- ing times of naturally occurring texts. In addition, we provide evidence that syntactic category disambiguation may be independent of syntactic top-down expectations, emphasizing the important role of bottom-up processes within an architecture of human language processing. Keywords: sentence processing; reading; eye-tracking; ambiguity; lexical access. Syntactic Category Ambiguity Introduction Successful language processing requires the integration of bottom-up information extracted from the current input and top-down expectations generated from what has been processed so far. When and how bottom-up and top-down processes interact has been a distinguish- ing feature of different processing architectures. On the one hand, there are constraint-based models (e.g. Mac- Donald, Pearlmutter, & Seidenberg, 1994; Trueswell & Tanenhaus, 1994; Tabor, Juliano, & Tanenhaus, 1997), which assume one single processing unit, in which all available information is considered simultaneously. Mod- ular architectures, on the other hand, consist of several distinct processing modules (e.g. Frazier, 1987; Frazier & Clifton, 1996; Corley & Crocker, 2000). These mod- ules are restricted to each having its own internal rep- resentation, and they are independently predictive and informationally encapsulated (Crocker & Corley, 2002). Assuming this definition of modules in terms of informa- tion flow, bottom-up processes are more likely to be mod- ular than top-down processes (Appelbaum, 1998; Fodor, One particular process, for which constraint-based and modular models make contradicting predictions, is syntactic category assignment or disambiguation: constraint-based models assume that rich contextual in- formation is utilized to determine the syntactic category (i.e part of speech) of a word, while modular architec- tures only allow context-independent information. Al- though previous research may seem to have provided ev- idence for both positions, Corley and Crocker (2000) (see also Gibson, 2006) have shown that most of the evidence Many words in English (and presumably all other lan- guages) are ambiguous, they can have different senses and/or belong to different syntactic categories or part- of-speech (i.e. noun, verb, adjective, etc.). The following example (from Boland, 1997) illustrates these ambigui- ties: a. b. I saw her duck . . . . . . under the porch to eat some potato chips. . . . under the porch eat some potato chips In (1), the word duck is ambiguous between its verb and noun readings, and only the following context can disam- biguate between the two syntactic categories and senses. Syntactic category ambiguity and lexical ambiguity (in terms of different senses) need not come together like in (1). Lexical ambiguity often occurs within the same syn- tactic category as in the word cabinet, which as a noun can denote either a group of advisors or a closet. Syn- tactic category ambiguity, on the other hand, does not require lexical ambiguity, as evidenced by the English verbal system, where for all regular verbs there is only one form for the past-tense and the past-participle. This ambiguity is crucial to many garden-path sentences. The horse raced past the barn fell. The horse ridden past the barn fell. While example (2) is a classical garden-path sentence, which upon first encounter may be nearly impossible to understand, example (3) is unambiguous and relatively easy to process. The fact that example (2) is derived

Acquiring English dative verbs: proficiency effects in German L2 learners Christoph Wolk * (christoph.wolk@frias.uni-freiburg.de), Sascha Wolfer † , Peter Baumann † , Barbara Hemforth ‡ , Lars Konieczny *† * Freiburg Institute for Advanced Studies, Starkenstr. 44 D-79104 Freiburg i. Br., Germany † Center for Cognitive Science, University of Freiburg, Friedrichstr. 50 D-79098 Freiburg i. Br., Germany ‡ Laboratoire de Psychologie et de Neuropsychologie Cognitives, CNRS, Universite Paris Descartes, 71 ave Edouard Vaillant, 92100 Boulogne-Billancourt, France Abstract This paper investigates the influence of probabilistic informa- tion in the second language on the processing of English da- tive alternation constructions in German learners of English. We present two eye-tracking studies (visual world and reading) with evidence that the probabilistic patterns of the target lan- guage influence L2 processing when the initial preference is vi- olated, and indications that these patterns have a greater effect on more experienced speakers. We also observed a constrast- effect of L1, such that comprehenders expected constructions that occur more often in L2 than in L1, even if L2 lexical statis- tics suggested otherwise. Keywords: Sentence processing, dative alternation, second language acquisition, expectation-based language processing Introduction In many languages, semantically dative sentences can be re- alized with two different object orders that only slightly dif- fer in meaning, one in which the recipient comes before the theme and one where the reverse is true. In English, the for- mer ordering is achieved by two bare noun phrases, as in (1-a), and the latter by having the recipient as a prepositional phrase, as in (1-b). (1) a. b. double object dative (DO) I gave [her] recipient [the book] theme . prepositional dative (PO) I gave [the book] theme to [her] recipient . The dative alternation has received considerable attention from first- and second language acquisition researchers dur- ing the 1980s, especially from the perspective of genera- tive grammar. These studies focused primarily on investigat- ing the following two questions by means of grammatical- ity judgments and sentence completion tasks: First, how well do learners acquire hard constraints on the possibility of al- ternation, such as the fixed prepositional realization of most verbs of Latin origin such as donate; second, what is the order in which speakers acquire the possible realizations for verbs that do alternate. Major results (e.g. in Mazurkewich, 1985; Mazurkewich & White, 1984) were that verb-specific con- straints are acquirable as hard constraints for first language learners with rare errors, but are only learned as softer con- straints — or sometimes not learned at all — for second lan- guage learners. With regard to acquisition order, the preposi- tional dative realization tends to be acquired earlier and easier for second language learners. Recent research on first language (L1) dative alternation patterns, however, has switched the focus from presumably ’hard’ constraints on the possibility of alternation to the softer, probabilistic determinants of actually observed vari- ation. This was motivated by cross-linguistic similarities in grammatical preferences (Bresnan, Dingare, & Manning, 2001) as well as the fact that in both naturally occurring lan- guage and experimental investigation ’hard’ constraints turn out to be surprisingly violable (Bresnan & Nikitina, 2008; Bresnan, 2007) while simultaneous consideration of multi- ple ’soft’ constraints led to considerable success in predic- tion of realizations, reading time, and fluency of production (Bresnan, Cueni, Nikitina, & Baayen, 2007; Bresnan & Ford, 2010; Tily et al., 2009). With regard to acquisition, children have been shown to mirror the probabilistic realization pat- terns of their environment (Marneffe, Grimm, Arnon, Kirby, & Bresnan, to appear). Second language studies within this probabilistic paradigm, however, are still rare; one exception is the study by Frishkoff, Levin, Pavlik, Idemaru, and Jong (2008), who used the results of (Bresnan et al., 2007) to in- vestigate how both native and second language (L2) speak- ers learn to predict dative choice from examples, and found that L2 learners improve quickly when presented with stimuli containing a high degree of contrast between alternation pref- erences. Individual factors that were found to be reliable pre- dictors for L1 speakers in corpus models have, however, also been shown to influence L2 learning at various proficiency levels. These include among others pronominality (Le Com- pagnon, 1984), givenness and persistence (Marefat, 2005), and weight (Tanaka, 1968; Callies & Szczesniak, 2008). The goal of this paper is to investigate how attuned L2 learners are to fine probabilistic details of their target lan- guage. One predictor that is, due to its inherently probabilis- tic nature, especially suited for this research question is verb bias. More specifically, each dative verb has a specific id- iosyncratic degree of preference in alternation choice; this preference is in general not predictable from semantics or morphology. A learner’s acquisition of verb bias should thus be seen as direct instances of fundamentally experience-based learning. The experiments reported here are based on English L2 learners with German as L1. Like English, and unlike most L1s of previous studies, German has a double object da- tive; in contrast, the use of the prepositional dative is limited

Typically, modern dialects show contact-induced rather than endogenous phonological change, i.e. a given dialect absorbs features of other varieties of the same language with which it is in contact, often replacing the dialect’s own sounds. In most cases, this process is lexicalized, i.e. it proceeds word by word, although the entire sound structure of the variety may be affected in the end. One of the unresolved questions of research on phonological dialect change is the nature of the spread of these changes. In principle, they can be due to horizontal (neighbouring dialects) or vertical contact (the standard variety or other forms with a wider range influence the dialects in the area “below” it). This paper discusses some quantitative measures used to estimate the magnitude of the two alternative forms of the propagation of a change, and reports some findings for the traditional dialects of southwest Germany in the last century. On the basis of a large corpus of spontaneous speech, we present an aggregated analysis using statistical correlations and a mixed logistic regression model.

Pragmatic expectations and coreference: how alternative constructions and referring expressions can serve as cues Peter Baumann (peter.baumann@mars.uni-freiburg.de) Department of Romance Languages and Literature, University of Freiburg, Platz der Universitat 3 D-79098 Freiburg i. Br., Germany Lars Konieczny (lars@cognition.uni-freiburg.de) Center for Cognitive Science, University of Freiburg, Friedrichstr. 50 D-79098 Freiburg i. Br., Germany Barbara Hemforth (barbara.hemforth@parisdescartes.fr) Laboratoire de Psychologie et de Neuropsychologie Cognitives, CNRS, Universite Paris Descartes, 71 ave Edouard Vaillant, 92100 Boulogne-Billancourt, France Abstract This paper addresses the question whether possible alternative constructions and alternative choices of referring expressions influence the resolution of anaphoric expressions. We present a questionnaire, a self-paced reading study and a corpus anal- ysis, suggesting that alternative constructions and referring ex- pressions help to constitute preferences for anaphora in refer- entially ambiguous sentences and also affect on-line sentence processing. Keywords: Anaphora resolution, pragmatics, conversational implicatures, pronouns, sentence processing Introduction In many languages, the interpretation of non-reflexive pro- nouns is sensitive to a variety of factors. On the structural and syntactic level, there is evidence indicating a general prefer- ence for the first-mentioned antecedent (Gernsbacher & Har- greaves, 1988) and for the subject (Jarvikivi, Gompel, Hyona, & Bertram, 2005), while on the pragmatic/discourse level, topicality (Givon, 1983), the chain of causality, and general discourse relations (Sanders, 1997; Kehler, 2002) have been shown to influence anaphora resolution. One may thus ar- gue that anaphora resolution is driven by the interaction of grammatical rules and pragmatic constraints, with the former being language-specific and the latter reflecting general prin- ciples of human cognition (cf. Thornton, Gil, & MacDonald, One prominent set of pragmatic constraints are the conver- sational maxims formulated by Grice (1975), which listeners rely on and speakers exploit to convey meaning beyond the level of what is actually said. Out of the four maxims, two are of special interest here: the maxim of quantity (‘do not make your contribution more informative than is required’) and the maxim of manner (‘avoid ambiguity’). They are the basis for most conversational implicatures used to describe the roles of different referring expressions in anaphora resolution. One such example is the contrast between pronouns and reflexives in English (Levinson, 1987): reflexives are more informative than pronouns in the sense that they entail immediate coref- erence and are to be bound within the local domain. When hearing a pronoun, a listener can thus infer that the speaker does not mean immediate coreference, because if so he would have used a reflexive. While most work on conversational implicatures in anaphora resolution focuses on the role of alternative refer- ring expressions, we would like to suggest that implicatures can also be formed on the basis of constructions, more specif- ically on possible alternative constructions that were not used. Alternative Constructions For the domain of relative clause attachment one such exam- ple of a conversational implicature based on alternative con- structions can be found in Frazier and Clifton (1996): in or- der to explain why in sentences like (1) English shows low attachment (i.e. the colonel had an accident), while Spanish and many other languages prefer high attachment (as reported by Mitchell, Cuetos, & Corley, 1992), one may consider that English has an alternative construction to unambiguously ex- press high attachment: the Saxon genitive (2). An English listener can presuppose that the speaker obeys the Gricean maxim of manner and chooses the most appropriate construc- tion. In the referentially ambiguous sentence (1a), the listener can thus assume that the speaker intended a low-attachment reading, because he could have used the alternative construc- tion to unambigously express high attachment. For a Spanish speaker, on the other hand, there is only one way to express sentence (1b), and so the listener must assume that the more ‘prominent’ attachment site was intended by the speaker. (1) a. b. The daughter of the colonel who had an accident . . . La hija del coronel que tuvo un accidente . . . The colonel’s daughter who had an accident . . . First evidence that alternative constructions might also play a role in anaphora resolution was presented by Hemforth, Colonna, Pynte, and Konieczny (2004). In a cross-linguistic comparison they showed that sentences like (3) are inter- preted differently in German and French: while Germans consider the subject ‘Polizist’ (‘policeman’) as the antecedent of the pronoun, speakers of French prefer the object ‘facteur’ (‘postman’).

We present two visual world studies indicating that local syntactic coherences interact with binding constraints (Chomsky, 1981) of both reflexives and pronouns. Gazes to depicted referents or events suggest that when sentences containing a local coherence with a pronoun or reflexive are presented, locally coherent antecedents become activated. Our results strengthen the assumption that local syntactic coherences are interpreted and extend the effect to online anaphora resolution and complementary binding constraints.

We present two visual world studies indicating that local syntactic coherences interact with binding constraints (Chomsky, 1981) of both reflexives and pronouns. Gazes to depicted referents or events suggest that when sentences containing a local coherence with a pronoun or reflexive are presented, locally coherent antecedents become activated. Our results strengthen the assumption that local syntactic coherences are interpreted and extend the effect to online anaphora resolution and complementary binding constraints.

Anaphora and Local Coherences Lars Konieczny (lars@cognition.uni-freiburg.de) Helmut Weldle (helmut@cognition.uni-freiburg.de) Sascha Wolfer (sascha@cognition.uni-freiburg.de) Daniel M uller (daniel@cognition.uni-freiburg.de) Peter Baumann (peter@cognition.uni-freiburg.de) Center for Cognitive Science, University of Freiburg, Friedrichstr. 50 D-79098 Freiburg i. Br., Germany Abstract sensitivity towards different sources of information for pro- nouns and reflexives. Sturt (2003) on the other hand showed that the constrain- ing principle for reflexives operates at the very earliest stages of processing. In eye-tracking-while-reading experiments, he found early effects of binding preferences. He concludes that the responsible binding principle is an early filter for the pro- cessing of referring expressions. We present two visual world studies indicating that local syn- tactic coherences interact with binding constraints (Chomsky, 1981) of both reflexives and pronouns. Gazes to depicted ref- erents or events suggest that when sentences containing a local coherence with a pronoun or reflexive are presented, locally coherent antecedents become activated. Our results strengthen the assumption that local syntactic coherences are interpreted and extend the effect to online anaphora resolution and com- plementary binding constraints. Keywords: anaphora, anaphora resolution, local syntactic co- herences Interpretation of locally syntactic coherences Introduction To arrive at a coherent interpretation of a sentence, we need to bind referring expressions to their correct referent. Bind- ing theory (Chomsky, 1981) provides a syntax-driven struc- tural account for the dependencies of co-reference within sen- tences. Principles based on c-command are assumed to con- strain the possible co-referents of anaphoric expressions on a global level. Reflexives and pronouns have complementary structural binding domains, i.e. within sentence boundaries, the accessible antecedents for both anaphora types are mu- tually exclusive. In sentences like (1-a) and (1-b) determin- ing the referent of the anaphoric expression himself or him is straightforward. 1 (1) a. Ken i who likes John j saw himself i/∗ j in the mirror. b. Ken j who likes John i saw him i/∗ j in the mirror. Recent findings question strictly structure-driven accounts of anaphora resolution. Runner, Sussman, and Tanenhaus (2006), for instance, report violations of the binding domain complementarity assumption. They examined preferences for pronoun and reflexive binding in picture noun phrases in a series of visual-world studies. Fixation probabilities on de- picted referents revealed violations of the binding theory as- sumption. They concluded that reflexives should rather be explained in terms of logophors, deposing reflexives beyond the scope of Binding theory explanations. Converging evidence was found by Kaiser, Runner, Suss- man, and Tanenhaus (2009), suggesting that the interpretation of reflexives is not only sensitive to structural but also seman- tic information. Moreover, they found differing degrees of 1 Subscripted indices mark coreference. A star * indicates that coreference with the indexed referent is not acceptable considering the global parse but suggested considering the local parse. Local syntactic coherences (LSCs) have been shown to in- terfere with the global sentence interpretation. Tabor, Galan- tucci, and Richardson (2004) found increased reading times on the spill-over of tossed in sentences like The coach chided the player tossed a frisbee by the opposing team. Moreover, Konieczny, M¨uller, Baumann, Hachmann, and Wolfer (2009) have shown that LSCs temporarily affect the interpretation of globally unambiguous sentences. Interestingly, there are locally coherent substrings in (1), leading to the opposite binding of the reflexive or the pro- noun: In Sentence (1-a), himself is restricted t bound to John if only the local subparse John saw himself is taken into ac- count. In the global parse however, himself is bound to Ken. In Sentence (1-b), him is bound to Ken or any other (unmen- tioned) referent if the LSC John saw him is interpreted. It is still an open question though, whether or not LSCs can affect anaphora resolution, and if pronouns and reflexives are affected equally. In the remainder of the paper, two experiments will be re- ported providing insight into the time-course of binding and its interaction with local syntactic coherences. We chose the visual-world paradigm (Cooper, 1974; Tanenhaus, Spivey- Knowlton, Eberhard, & Sedivy, 1995) over reading times, as fixations on visual objects can indicate binding preferences in a much more direct way, without inferencing over processing difficulty. The results suggest that LSCs indeed have an effect on the binding of reflexives and pronouns in a way that strong constraints like Principle A are also applied to the local parse, temporarily overriding the globally correct binding. Experiment 1: Depicted referents in the visual world In the first experiment, the visual stimuli showed the de- pictions of three persons, two of which were depictions of the referents introduced in the spoken sentence. When the