Refine
Year of publication
Document Type
- Doctoral Thesis (97) (remove)
Language
- English (97) (remove)
Keywords
- Satzverarbeitung (8)
- Spracherwerb (7)
- psycholinguistics (7)
- sentence processing (7)
- Informationsstruktur (6)
- Psycholinguistik (6)
- eye-tracking (6)
- information structure (6)
- prosody (6)
- Fokus (5)
Institute
- Department Linguistik (97) (remove)
Successful sentence comprehension requires the comprehender to correctly figure out who did what to whom. For example, in the sentence John kicked the ball, the comprehender has to figure out who did the action of kicking and what was being kicked. This process of identifying and connecting the syntactically-related words in a sentence is called dependency completion. What are the cognitive constraints that determine dependency completion? A widely-accepted theory is cue-based retrieval. The theory maintains that dependency completion is driven by a content-addressable search for the co-dependents in memory. The cue-based retrieval explains a wide range of empirical data from several constructions including subject-verb agreement, subject-verb non-agreement, plausibility mismatch configurations, and negative polarity items.
However, there are two major empirical challenges to the theory: (i) Grammatical sentences’ data from subject-verb number agreement dependencies, where the theory predicts a slowdown at the verb in sentences like the key to the cabinet was rusty compared to the key to the cabinets was rusty, but the data are inconsistent with this prediction; and, (ii) Data from antecedent-reflexive dependencies, where a facilitation in reading times is predicted at the reflexive in the bodybuilder who worked with the trainers injured themselves vs. the bodybuilder who worked with the trainer injured themselves, but the data do not show a facilitatory effect.
The work presented in this dissertation is dedicated to building a more general theory of dependency completion that can account for the above two datasets without losing the original empirical coverage of the cue-based retrieval assumption. In two journal articles, I present computational modeling work that addresses the above two empirical challenges.
To explain the grammatical sentences’ data from subject-verb number agreement dependencies, I propose a new model that assumes that the cue-based retrieval operates on a probabilistically distorted representation of nouns in memory (Article I). This hybrid distortion-plus-retrieval model was compared against the existing candidate models using data from 17 studies on subject-verb number agreement in 4 languages. I find that the hybrid model outperforms the existing models of number agreement processing suggesting that the cue-based retrieval theory must incorporate a feature distortion assumption.
To account for the absence of facilitatory effect in antecedent-reflexive dependencies, I propose an individual difference model, which was built within the cue-based retrieval framework (Article II). The model assumes that individuals may differ in how strongly they weigh a syntactic cue over a number cue. The model was fitted to data from two studies on antecedent-reflexive dependencies, and the participant-level cue-weighting was estimated. We find that one-fourth of the participants, in both studies, weigh the syntactic cue higher than the number cue in processing reflexive dependencies and the remaining participants weigh the two cues equally. The result indicates that the absence of predicted facilitatory effect at the level of grouped data is driven by some, not all, participants who weigh syntactic cues higher than the number cue. More generally, the result demonstrates that the assumption of differential cue weighting is important for a theory of dependency completion processes. This differential cue weighting idea was independently supported by a modeling study on subject-verb non-agreement dependencies (Article III).
Overall, the cue-based retrieval, which is a general theory of dependency completion, needs to incorporate two new assumptions: (i) the nouns stored in memory can undergo probabilistic feature distortion, and (ii) the linguistic cues used for retrieval can be weighted differentially. This is the cumulative result of the modeling work presented in this dissertation.
The dissertation makes an important theoretical contribution: Sentence comprehension in humans is driven by a mechanism that assumes cue-based retrieval, probabilistic feature distortion, and differential cue weighting. This insight is theoretically important because there is some independent support for these three assumptions in sentence processing and the broader memory literature. The modeling work presented here is also methodologically important because for the first time, it demonstrates (i) how the complex models of sentence processing can be evaluated using data from multiple studies simultaneously, without oversimplifying the models, and (ii) how the inferences drawn from the individual-level behavior can be used in theory development.
The aim of this dissertation was to conduct a larger-scale cross-linguistic empirical investigation of similarity-based interference effects in sentence comprehension.
Interference studies can offer valuable insights into the mechanisms that are involved in long-distance dependency completion.
Many studies have investigated similarity-based interference effects, showing that syntactic and semantic information are employed during long-distance dependency formation (e.g., Arnett & Wagers, 2017; Cunnings & Sturt, 2018; Van Dyke, 2007, Van Dyke & Lewis, 2003; Van Dyke & McElree, 2011). Nevertheless, there are some important open questions in the interference literature that are critical to our understanding of the constraints involved in dependency resolution.
The first research question concerns the relative timing of syntactic and semantic interference in online sentence comprehension. Only few interference studies have investigated this question, and, to date, there is not enough data to draw conclusions with regard to their time course (Van Dyke, 2007; Van Dyke & McElree, 2011).
Our first cross-linguistic study explores the relative timing of syntactic and semantic interference in two eye-tracking reading experiments that implement the study design used in Van Dyke (2007). The first experiment tests English sentences. The second, larger-sample experiment investigates the two interference types in German.
Overall, the data suggest that syntactic and semantic interference can arise simultaneously during retrieval.
The second research question concerns a special case of semantic interference: We investigate whether cue-based retrieval interference can be caused by semantically similar items which are not embedded in a syntactic structure.
This second interference study builds on a landmark study by Van Dyke & McElree (2006). The study design used in their study is unique in that it is able to pin down the source of interference as a consequence of cue overload during retrieval, when semantic retrieval cues do not uniquely match the retrieval target. Unlike most other interference studies, this design is able to rule out encoding interference as an alternative explanation. Encoding accounts postulate that it is not cue overload at the retrieval site but the erroneous encoding of similar linguistic items in memory that leads to interference (Lewandowsky et al., 2008; Oberauer & Kliegl, 2006). While Van Dyke & McElree (2006) reported cue-based retrieval interference from sentence-external distractors, the evidence for this effect was weak. A subsequent study did not show interference of this type (Van Dyke et al., 2014). Given these inconclusive findings, further research is necessary to investigate semantic cue-based retrieval interference.
The second study in this dissertation provides a larger-scale cross-linguistic investigation of cue-based retrieval interference from sentence-external items. Three larger-sample eye-tracking studies in English, German, and Russian tested cue-based interference in the online processing of filler-gap dependencies. This study further extends the previous research by investigating interference in each language under varying task demands (Logačev & Vasishth, 2016; Swets et al., 2008).
Overall, we see some very modest support for proactive cue-based retrieval interference in English. Unexpectedly, this was observed only under a low task demand. In German and Russian, there is some evidence against the interference effect. It is possible that interference is attenuated in languages with richer case marking.
In sum, the cross-linguistic experiments on the time course of syntactic and semantic interference from sentence-internal distractors support existing evidence of syntactic and semantic interference during sentence comprehension. Our data further show that both types of interference effects can arise simultaneously. Our cross-linguistic experiments investigating semantic cue-based retrieval interference from sentence-external distractors suggest that this type of interference may arise only in specific linguistic contexts.
Adverb positioning is guided by syntactic, semantic, and pragmatic considerations and is subject to cross-linguistic as well as language-specific variation. The goal of the thesis is to identify the factors that determine adverb placement in general (Part I) as well as in constructions in which the adverb's sister constituent is deprived of its phonetic material by movement or ellipsis (gap constructions, Part II) and to provide an Optimality Theoretic approach to the contrasts in the effects of these factors on the distribution of adverbs in English, French, and German. In Optimality Theory (Prince & Smolensky 1993), grammaticality is defined as optimal satisfaction of a hierarchy of violable constraints: for a given input, a set of output candidates are produced out of which that candidate is selected as grammatical output which optimally satisfies the constraint hierarchy. Since grammaticality crucially relies on the hierarchic relations of the constraints, cross-linguistic variation can be traced back to differences in the language-specific constraint rankings. Part I shows how diverse phenomena of adverb placement can be captured by corresponding constraints and their relative rankings: - contrasts in the linearization of adverbs and verbs/auxiliaries in English and French - verb placement in German and the filling of the prefield position - placement of focus-sensitive adverbs - fronting of topical arguments and adverbs Part II extends the analysis to a particular phenomenon of adverb positioning: the avoidance of adverb attachment to a phonetically empty constituent (gap). English and French are similar in that the acceptability of pre-gap adverb placement depends on the type of adverb, its scope, and the syntactic construction (English: wh-movement vs. topicalization / VP Fronting / VP Ellipsis, inverted vs. non-inverted clauses; French: CLLD vs. Cleft, simple vs. periphrastic tense). Yet, the two languages differ in which strategies a specific type of adverb may pursue to escape placement in front of a certain type of gap. In contrast to English and French, placement of an adverb in front of a gap never gives rise to ungrammaticality in German. Rather, word ordering has to obey the syntactic, semantic, and pragmatic principles discussed in Part I; whether or not it results in adverb attachment to a phonetically empty constituent seems to be irrelevant: though constraints are active in every language, the emergence of a visible effect of their requirements in a given language depends on their relative ranking. The complex interaction of the diverse factors as well as their divergent effects on adverb placement in the various languages are accounted for by the universal constraints and their language-specific hierarchic relations in the OT framework.
Age of acquisition (AOA) is a psycholinguistic variable that significantly influences behavioural measures (response times and accuracy rates) in tasks that require lexical and semantic processing. Its origin is – unlike the origin of semantic typicality (TYP), which is assumed at the semantic level – controversially discussed. Different theories propose AOA effects to originate either at the semantic level or at the link between semantics and phonology (lemma-level).
The dissertation aims at investigating the influence of AOA and its interdependence with the semantic variable TYP on particularly semantic processing in order to pinpoint the origin of AOA effects. Therefore, three studies have been conducted that considered the variables AOA and TYP in semantic processing tasks (category verifications and animacy decisions) by means of behavioural and partly electrophysiological (ERP) data and in different populations (healthy young and elderly participants and in semantically impaired individuals with aphasia (IWA)).
The behavioural and electrophysiological data of the three studies provide evidence for distinct processing levels of the variables AOA and TYP. The data further support previous assumptions on a semantic origin for TYP but question the same for AOA. The findings, however, support an origin of AOA effects at the transition between the word form (phonology) and the semantic level that can be captured at the behavioural but not at the electrophysiological level.
This dissertation explores whether the processing of ellipsis is affected by changes in the complexity of the antecedent, either due to added linguistic material or to the presence of a temporary ambiguity. Murphy (1985) hypothesized that ellipsis is resolved via a string copying procedure when the antecedent is within the same sentence, and that copying longer strings takes more time. Such an account also implies that the antecedent is copied without its structure, which in turn implies that recomputing its syntax and semantics may be necessary at the ellipsis gap. Alternatively, several accounts predict null effects of antecedent complexity, as well as no reparsing. These either involve a structure copying mechanism that is cost-free and whose finishing time is thus independent of the form of the antecedent (Frazier & Clifton, 2001), treat ellipsis as a pointer into content-addressable memory with direct access (Martin & McElree, 2008, 2009), or assume that one structure is ‘shared’ between antecedent and gap (Frazier & Clifton, 2005).
In a self-paced reading study on German sluicing, temporarily ambiguous garden-path clauses were used as antecedents, but no evidence of reparsing in the form of a slowdown at the ellipsis site was found. Instead, results suggest that antecedents which had been reanalyzed from an initially incorrect structure were easier to retrieve at the gap. This finding that can be explained within the framework of cue-based retrieval parsing (Lewis & Vasishth, 2005), where additional syntactic operations on a structure yield memory reactivation effects.
Two further self-paced reading studies on German bare argument ellipsis and English verb phrase ellipsis investigated if adding linguistic content to the antecedent would increase processing times for the ellipsis, and whether insufficiently demanding comprehension tasks may have been responsible for earlier null results (Frazier & Clifton, 2000; Martin & McElree, 2008). It has also been suggested that increased antecedent complexity should shorten rather than lengthen retrieval times by providing more unique memory features (Hofmeister, 2011). Both experiments failed to yield reliable evidence that antecedent complexity affects ellipsis processing times in either direction, irrespectively of task demands.
Finally, two eye-tracking studies probed more deeply into the proposed reactivation-induced speedup found in the first experiment. The first study used three different kinds of French garden-path sentences as antecedents, with two of them failing to yield evidence for reactivation. Moreover, the third sentence type showed evidence suggesting that having failed to assign a structure to the antecedent leads to a slowdown at the ellipsis site, as well as regressions towards the ambiguous part of the sentence. The second eye-tracking study used the same materials as the initial self-paced reading study on German, with results showing a pattern similar to the one originally observed, with some notable differences.
Overall, the experimental results are compatible with the view that adding linguistic material to the antecedent has no or very little effect on the ease with which ellipsis is resolved, which is consistent with the predictions of cost-free copying, pointer-based approaches and structure sharing. Additionally, effects of the antecedent’s parsing history on ellipsis processing may be due to reactivation, the availability of multiple representations in memory, or complete failure to retrieve a matching target.
This project describes the nominal, verbal and ‘truncation’ systems of Awing and explains the syntactic and semantic functions of the multifunctional l<-><-> (LE) morpheme in copular and wh-focused constructions. Awing is a Bantu Grassfields language spoken in the North West region of Cameroon. The work begins with morphological processes viz. deverbals, compounding, reduplication, borrowing and a thorough presentation of the pronominal system and takes on verbal categories viz. tense, aspect, mood, verbal extensions, negation, adverbs and triggers of a homorganic N(asal)-prefix that attaches to the verb and other verbal categories. Awing grammar also has a very unusual phenomenon whereby nouns and verbs take long and short forms. A chapter entitled truncation is dedicated to the phenomenon. It is observed that the truncation process does not apply to bare singular NPs, proper names and nouns derived via morphological processes. On the other hand, with the exception of the 1st person non-emphatic possessive determiner and the class 7 noun prefix, nouns generally take the truncated form with modifiers (i.e., articles, demonstratives and other possessives). It is concluded that nominal truncation depicts movement within the DP system (Abney 1987). Truncation of the verb occurs in three contexts: a mass/plurality conspiracy (or lattice structuring in terms of Link 1983) between the verb and its internal argument (i.e., direct object); a means to align (exhaustive) focus (in terms of Fery’s 2013), and a means to form polar questions.
The second part of the work focuses on the role of the LE morpheme in copular and wh-focused clauses. Firstly, the syntax of the Awing copular clause is presented and it is shown that copular clauses in Awing have ‘subject-focus’ vs ‘topic-focus’ partitions and that the LE morpheme indirectly relates such functions. Semantically, it is shown that LE does not express contrast or exhaustivity in copular clauses. Turning to wh-constructions, the work adheres to Hamblin’s (1973) idea that the meaning of a question is the set of its possible answers and based on Rooth’s (1985) underspecified semantic notion of alternative focus, concludes that the LE morpheme is not a Focus Marker (FM) in Awing: LE does not generate or indicate the presence of alternatives (Krifka 2007); The LE morpheme can associate with wh-elements as a focus-sensitive operator with semantic import that operates on the focus alternatives by presupposing an exhaustive answer, among other notions. With focalized categories, the project further substantiates the claim in Fominyam & Šimík (2017), namely that exhaustivity is part of the semantics of the LE morpheme and not derived via contextual implicature, via a number of diagnostics. Hence, unlike in copular clauses, the LE morpheme with wh-focused categories is analysed as a morphological exponent of a functional head Exh corresponding to Horvath's (2010) EI (Exhaustive Identification). The work ends with the syntax of verb focus and negation and modifies the idea in Fominyam & Šimík (2017), namely that the focalized verb that associates with the exhaustive (LE) particle is a lower copy of the finite verb that has been moved to Agr. It is argued that the LE-focused verb ‘cluster’ is an instantiation of adjunction. The conclusion is that verb doubling with verb focus in Awing is neither a realization of two copies of one and the same verb (Fominyam and Šimík 2017), nor a result of a copy triggered by a focus marker (Aboh and Dyakonova 2009). Rather, the focalized copy is said to be merged directly as the complement of LE forming a type of adjoining cluster.
The main goal of this thesis is to explore the feasibility of using cross-lingual annotation projection as a method of alleviating the task of manual coreference annotation.
To reach our goal, we build a first trilingual parallel coreference corpus that encompasses multiple genres. For the annotation of the corpus, we develop common coreference annotation guidelines that are applicable to three languages (English, German, Russian) and include a novel domain-independent typology of bridging relations as well as state-of-the-art near-identity categories.
Thereafter, we design and perform several annotation projection experiments. In the first experiment, we implement a direct projection method with only one source language. Our results indicate that, already in a knowledge-lean scenario, our projection approach is superior to the most closely related work of Postolache et al. (2006). Since the quality of the resulting annotations is to a high degree dependent on the word alignment, we demonstrate how using limited syntactic information helps to further improve mention extraction on the target side. As a next step, in our second experiment, we show how exploiting two source languages helps to improve the quality of target annotations for both language pairs by concatenating annotations projected from two source languages. Finally, we assess the projection quality in a fully automatic scenario (using automatically produced source annotations), and propose a pilot experiment on manual projection of bridging pairs.
For each of the experiments, we carry out an in-depth error analysis, and we conclude that noisy word alignments, translation divergences and morphological and syntactic differences between languages are responsible for projection errors. We systematically compare and evaluate our projection methods, and we investigate the errors both qualitatively and quantitatively in order to identify problematic cases. Finally, we discuss the applicability of our method to coreference annotations and propose several avenues of future research.
The aim of this thesis is to develop approaches to automatically recognise the structure of argumentation in short monological texts. This amounts to identifying the central claim of the text, supporting premises, possible objections, and counter-objections to these objections, and connecting them correspondingly to a structure that adequately describes the argumentation presented in the text.
The first step towards such an automatic analysis of the structure of argumentation is to know how to represent it. We systematically review the literature on theories of discourse, as well as on theories of the structure of argumentation against a set of requirements and desiderata, and identify the theory of J. B. Freeman (1991, 2011) as a suitable candidate to represent argumentation structure. Based on this, a scheme is derived that is able to represent complex argumentative structures and can cope with various segmentation issues typically occurring in authentic text.
In order to empirically test our scheme for reliability of annotation, we conduct several annotation experiments, the most important of which assesses the agreement in reconstructing argumentation structure. The results show that expert annotators produce very reliable annotations, while the results of non-expert annotators highly depend on their training in and commitment to the task.
We then introduce the 'microtext' corpus, a collection of short argumentative texts. We report on the creation, translation, and annotation of it and provide a variety of statistics. It is the first parallel corpus (with a German and English version) annotated with argumentation structure, and -- thanks to the work of our colleagues -- also the first annotated according to multiple theories of (global) discourse structure.
The corpus is then used to develop and evaluate approaches to automatically predict argumentation structures in a series of six studies: The first two of them focus on learning local models for different aspects of argumentation structure. In the third study, we develop the main approach proposed in this thesis for predicting globally optimal argumentation structures: the 'evidence graph' model. This model is then systematically compared to other approaches in the fourth study, and achieves state-of-the-art results on the microtext corpus. The remaining two studies aim to demonstrate the versatility and elegance of the proposed approach by predicting argumentation structures of different granularity from text, and finally by using it to translate rhetorical structure representations into argumentation structures.
Breaking down complexity
(2015)
The unbounded expressive capacity of human language cannot boil down to an infinite list of sentences stored in a finite brain. Our linguistic knowledge is rather grounded around a rule-based universal syntactic computation—called Merge—which takes categorized units in input (e.g. this and ship), and generates structures by binding words recursively into more complex hierarchies of any length (e.g. this ship; this ship sinks…). Here we present data from different fMRI datasets probing the cortical implementation of this fundamental process. We first pushed complexity down to a three-word level, to explore how Merge creates minimally hierarchical phrases and sentences. We then moved to the most fundamental two-word level, to directly assess the universal invariant nature of Merge, when no additive mechanisms are involved. Our most general finding is that Merge as the basic syntactic operation is primarily performed by confined area, namely BA 44 in the IFG. Activity reduces to its most ventral-anterior portion at the most fundamental level, following fine-grained sub-anatomical parcellation proposed for the region. The deep frontal operculum/anterior-dorsal insula (FOP/adINS), a phylogenetically older and less specialized region, rather appears to support word-accumulation processing in which the categorical information of the word is first accessed based on its lexical status, and then maintained on hold before further processing takes place. The present data confirm the general notion of BA 44 being activated as a function of complex structural hierarchy, but they go beyond this view by proposing that structural sensitivity in BA 44 is already appreciated at the lowest levels of complexity during which minimal phrase-structures are build up, and syntactic Merge is assessed. Further, they call for a redefinition of BA 44 from multimodal area to a macro-region with internal localizable functional profiles
Cleft exhaustivity
(2020)
In this dissertation a series of experimental studies are presented which demonstrate that the exhaustive inference of focus-background it-clefts in English and their cross-linguistic counterparts in Akan, French, and German is neither robust nor systematic. The inter-speaker and cross-linguistic variability is accounted for with a discourse-pragmatic approach to cleft exhaustivity, in which -- following Pollard & Yasavul 2016 -- the exhaustive inference is derived from an interaction with another layer of meaning, namely, the existence presupposition encoded in clefts.
The present dissertation investigates changes in lingual coarticulation across childhood in German-speaking children from three to nine years of age and adults. Coarticulation refers to the mismatch between the abstract phonological units and their seemingly commingled realization in continuous speech. Being a process at the intersection of phonology and phonetics, addressing its changes across childhood allows for insights in speech motor as well as phonological developments. Because specific predictions for changes in coarticulation across childhood can be derived from existing speech production models, investigating children’s coarticulatory patterns can help us model human speech production.
While coarticulatory changes may shed light on some of the central questions of speech production development, previous studies on the topic were sparse and presented a puzzling picture of conflicting findings. One of the reasons for this lack is the difficulty in articulatory data acquisition in a young population. Within the research program this dissertation is embedded in, we accepted this challenge and successfully set up the hitherto largest corpus of articulatory data from children using ultrasound tongue imaging. In contrast to earlier studies, a high number of participants in tight age cohorts across a wide age range and a thoroughly controlled set of pseudowords allowed for statistically powerful investigations of a process known as variable and complicated to track.
The specific focus of my studies is on lingual vocalic coarticulation as measured in the horizontal position of the highest point of the tongue dorsum. Based on three studies on a) anticipatory coarticulation towards the left, b) carryover coarticulation towards the right side of the utterance, and c) anticipatory coarticulatory extent in repeated versus read aloud speech, I deduct the following main theses:
1. Maturing speech motor control is responsible for some developmental changes in coarticulation.
2. Coarticulation can be modeled as the coproduction of articulatory gestures.
3. The developmental change in coarticulation results from a decrease of vocalic activation width.
Discourse production is crucial for communicative success and is in the core of aphasia assessment and treatment. Coherence differentiates discourse from a series of utterances/sentences; it is internal unity and connectedness, and, as such, perhaps the most inherent property of discourse. It is unclear whether people with aphasia, who experience various language production difficulties, preserve the ability to produce coherent discourse. A more general question of how coherence is established and represented linguistically has been addressed in the literature, yet remains unanswered. This dissertation presents an investigation of discourse production in aphasia and the linguistic mechanisms of establishing coherence.
Previous studies on the acquisition of verb inflection in normally developing children have revealed an astonishing pattern: children use correctly inflected verbs in their own speech but fail to make use of verb inflections when comprehending sentences uttered by others. Thus, a three-year old might well be able to say something like ‘The cat sleeps on the bed’, but fails to understand that the same sentence, when uttered by another person, refers to only one sleeping cat but not more than one. The previous studies that have examined children's comprehension of verb inflections have employed a variant of a picture selection task in which the child was asked to explicitly indicate (via pointing) what semantic meaning she had inferred from the test sentence. Recent research on other linguistic structures, such as pronouns or focus particles, has indicated that earlier comprehension abilities can be found when methods are used that do not require an explicit reaction, like preferential looking tasks. This dissertation aimed to examine whether children are truly not able to understand the connection the the verb form and the meaning of the sentence subject until the age of five years or whether earlier comprehension can be found when a different measure, preferential looking, is used. Additionally, children's processing of subject-verb agreement violations was examined. The three experiments of this thesis that examined children's comprehension of verb inflections revealed the following: German-speaking three- to four-year old children looked more to a picture showing one actor when hearing a sentence with a singular inflected verb but only when their eye gaze was tracked and they did not have to perform a picture selection task. When they were asked to point to the matching picture, they performed at chance-level. This pattern indicates asymmetries in children's language performance even within the receptive modality. The fourth experiment examined sensitivity to subject-verb agreement violations and did not reveal evidence for sensitivity toward agreement violations in three- and four-year old children, but only found that children's looking patterns were influenced by the grammatical violations at the age of five. The results from these experiments are discussed in relation to the existence of a production-comprehension asymmetry in the use of verb inflections and children's underlying grammatical knowledge.
Connective ties in discourse: Three ERP studies on causal, temporal and concessive connective ties and their influence on language processing. Questions In four experiments the influence of lexical connectives such as " darum", therefore, " danach", afterwards, and " trotzdem", nevertheless, on the processing of short two-sentence discourses was examined and compared to the processing of deictical sentential adverbs such as " gestern", yesterday, and " lieber", rather. These latter words do not have the property of signaling a certain discourse relation between two sentences, as connective ties do. Three questions were central to the work: * Do the processing contrasts found between connective and non-connective elements extend to connective ties and deictical sentential adverbs (experiments 2 and 3)? * Does the semantic content of the connective ties play the primary role, i.e is the major distinction to be made indeed between connective and non-connective or instead between causal, temporal and concessive? * When precisely is the information provided by connective ties used? There is some evidence that connective ties can have an immediate influence on the integration of subsequent elements, but the end of the second sentences appears to play an important role as well: experiments 2, 3, and 4. Conclusions First of all, the theoretical distinction between connective and non-connective elements does indeed have " cognitive reality" . This has already been shown in previous studies. The present studies do however show, that there is also a difference between one-place discourse elements (deictical sentential adverbs) and two-place discourse elements, namely connective ties, since all experiments examining this contrast found evidence for qualitatively and quantitatively different processing (experiments 1, 2, and 3). Secondly, the semantic type of the connective ties also plays a role. This was not shown for the LAN, found for all connective ties when compared to non-connective elements, and consequently interpreted as a more abstract reflection of the integration of connective ties. There was also no difference between causal and temporal connective ties before the end of the discourses in experiment 3. However, the N400 found for incoherent discourses in experiment 2, larger for connective incoherent than non-connective incoherent discourses, as well as the P3b found for concessive connective ties in the comparison between causal and concessive connective ties gave reason to assume that the semantic content of connective ties is made use of in incremental processing, and that the relation signaled by the connective tie is the one that readers attempt to construct. Concerning when the information provided by connective ties is used, it appears as if connectivity is generally and obligatorily taken at face value. As long as the meaning of a connective tie did not conflict with a preferred canonical discourse relation, there were no differences found for varying connective discourses (experiment 3). However, the fact that concessive connective ties announce the need for a more complex text representation was recognized and made use of immediately (experiment 4). Additionally, a violation of the discourse relation resulted in more difficult semantic integration if a connective tie was present (experiment 2). It is therefore concluded here that connective ties influence processing immediately. This claim has to be modified somewhat, since the sentence-final elements suggested that connective ties trigger different integration processes than non-connective elements. It seems as if the answer to the question of when connective ties are processed is neither exclusively immediately nor exclusively afterwards, but that both viewpoints are correct. It is suggested here that before the end of a discourse economy plays a central role in that a canonical relation is assumed unless there is evidence to the contrary. A connective tie could have the function of reducing the dimensions evaluated in a discourse to the one signaled by the connective tie. At the end of the discourse the representation is evaluated and verified, and an integrated situation model constructed. Here, the complexity of the different discourse relations that connective ties can signal, is expressed.
My thesis focused on the predictions of the activation-based model of Lewis and Vasishth (2005) to investigate the evidence for the use of the memory system in the formation of non-local dependencies in sentence comprehension.
The activation-based model, which follows the Adaptive Control of Thought-Rational framework (ACT-R; Anderson et al., 2004), has been used to explain locality effects and similarity-based interference by assuming that dependencies are resolved by a cue-based retrieval mechanism, and that the retrieval mechanism is affected by decay and interference.
Both locality effects and (inhibitory) similarity-based interference cause increased difficulty (e.g., longer reading times) at the site of the dependency completion where a retrieval is assumed: (I) Locality effects are attributed to the increased difficulty in the retrieval of a dependent when the distance from its retrieval site is increased. (II) Similarity-based interference is attributed to the retrieval being affected by the presence of items which have similar features as the dependent that needs to be retrieved.
In this dissertation, I investigated some findings problematic to the activation-based model, namely, facilitation where locality effects are expected (e.g., Levy, 2008), and the lack of similarity-based interference from the number feature in grammatical sentences (e.g., Wagers et al., 2009). In addition, I used individual differences in working memory capacity and reading fluency as a way to validate the theories investigated (Underwood, 1975), and computational modeling to achieve a more precise account of the phenomena.
Regarding locality effects, by using self-paced reading and eye-tracking-while reading methods with Spanish and German data, this dissertation yielded two main findings: (I) Locality effects seem to be modulated by working memory capacity, with high-capacity participants showing expectation-driven facilitation. (II) Once expectations and other potential confounds are controlled using baselines, with increased distance, high-capacity readers can show a slow-down (i.e., locality effects) and low-capacity readers can show a speedup. While the locality effects are compatible with the activation-based model, simulations show that the speedup of low-capacity readers can only be accounted for by changing some of the assumptions of the activation-based model.
Regarding similarity-based interference, two relatively high-powered self-paced reading experiments in German using grammatical sentences yielded a slowdown at the verb as predicted by the activation-based model. This provides evidence in favor of dependency creation via cue-based retrieval, and in contrast with the view that cue-based retrieval is a reanalysis mechanism (Wagers et al., 2009).
Finally, the same experimental results that showed inhibitory interference from the number feature are used for a finer grain evaluation of the retrieval process. Besides Lewis and Vasishth’s (2005) activation-based model, also McElree’s (2000) direct-access model can account for inhibitory interference. These two models assume a cue-based retrieval mechanism to build dependencies, but they are based on different assumptions. I present a computational evaluation of the predictions of these two theories of retrieval. The models were compared by implementing them in a Bayesian hierarchical framework. The evaluation of the models reveals that some aspects of the data fit better under the direct access model than under the activation-based model. However, a simple extension of the activation-based model provides a comparable fit to the direct access model. This serves as a proof of concept showing potential ways to improve the original activation-based model.
In conclusion, this thesis adds to the body of evidence that argues for the use of the general memory system in dependency resolution, and in particular for a cue-based retrieval mechanism. However, it also shows that some of the default assumptions inherited from ACT-R in the activation-based model need to be revised.
This thesis gives formal definitions of discourse-givenness, coreference and reference, and reports on experiments with computational models of discourse-givenness of noun phrases for English and German. Definitions are based on Bach's (1987) work on reference, Kibble and van Deemter's (2000) work on coreference, and Kamp and Reyle's Discourse Representation Theory (1993). For the experiments, the following corpora with coreference annotation were used: MUC-7, OntoNotes and ARRAU for Englisch, and TueBa-D/Z for German. As for classification algorithms, they cover J48 decision trees, the rule based learner Ripper, and linear support vector machines. New features are suggested, representing the noun phrase's specificity as well as its context, which lead to a significant improvement of classification quality.
Distributed optimality
(2001)
In this thesis I propose a synthesis (Distributed Optimality, DO) between Optimality Theory (OT, Prince & Smolensky, 1993) and a morphological framework in a genuine derivational tradition, namely Distributed Morphology (DM) as developed by Halle & Marantz (1993). By carrying over the apparatus of OT to DM, phenomena which are captured in DM by language-specific rules or features of lexical entries, are given a more principled account in the terms of ranked universal constraints. On the other hand, also the DM part makes two contributions, namely strong locality and impoverishment. The first gives rise to a simple formal interpretation of DO, while the latter is shown to be indispensable in any theoretically satisfying account of agreement morphology. The empirical basis of the work is given by the complex agreement morphology of genetically different languages. Theoretical focus is mainly on two areas: First, so-called direction marking which is shown to be preferably treated in terms of constraints on feature realization. Second, the effects of precedence constraints which are claimed to regulate the status of agreement affixes as prefixes or suffixes and their respective order. A universal typology for the order of agreement categories by means of OT-constraints is proposed.
Does it have to be trees? : Data-driven dependency parsing with incomplete and noisy training data
(2011)
We present a novel approach to training data-driven dependency parsers on incomplete annotations. Our parsers are simple modifications of two well-known dependency parsers, the transition-based Malt parser and the graph-based MST parser. While previous work on parsing with incomplete data has typically couched the task in frameworks of unsupervised or semi-supervised machine learning, we essentially treat it as a supervised problem. In particular, we propose what we call agnostic parsers which hide all fragmentation in the training data from their supervised components. We present experimental results with training data that was obtained by means of annotation projection. Annotation projection is a resource-lean technique which allows us to transfer annotations from one language to another within a parallel corpus. However, the output tends to be noisy and incomplete due to cross-lingual non-parallelism and error-prone word alignments. This makes the projected annotations a suitable test bed for our fragment parsers. Our results show that (i) dependency parsers trained on large amounts of projected annotations achieve higher accuracy than the direct projections, and that (ii) our agnostic fragment parsers perform roughly on a par with the original parsers which are trained only on strictly filtered, complete trees. Finally, (iii) when our fragment parsers are trained on artificially fragmented but otherwise gold standard dependencies, the performance loss is moderate even with up to 50% of all edges removed.
Early sensitivity to prosodic phrase boundary cues: Behavioral evidence from German-learning infants
(2023)
This dissertation seeks to shed light on the relation of phrasal prosody and developmental speech perception in German-learning infants. Three independent empirical studies explore the role of acoustic correlates of major prosodic boundaries, specifically pitch change, final lengthening, and pause, in infant boundary perception. Moreover, it was examined whether the sensitivity to prosodic phrase boundary markings changes during the first year of life as a result of perceptual attunement to the ambient language (Aslin & Pisoni, 1980).
Using the headturn preference procedure six- and eight-month-old monolingual German-learning infants were tested on their discrimination of two different prosodic groupings of the same list of coordinated names either with or without an internal IPB after the second name, that is, [Moni und Lilli] [und Manu] or [Moni und Lilli und Manu]. The boundary marking was systematically varied with respect to single prosodic cues or specific cue combinations.
Results revealed that six- and eight-month-old German-learning infants successfully detect the internal prosodic boundary when it is signaled by all the three main boundary cues pitch change, final lengthening, and pause. For eight-, but not for six-month-olds, the combination of pitch change and final lengthening, without the occurrence of a pause, is sufficient. This mirrors an adult-like perception by eight-months (Holzgrefe-Lang et al., 2016). Six-month-olds detect a prosodic phrase boundary signaled by final lengthening and pause. The findings suggest a developmental change in German prosodic boundary cue perception from a strong reliance on the pause cue at six months to a differentiated sensitivity to the more subtle cues pitch change and final lengthening at eight months. Neither for six- nor for eight-month-olds the occurrence of pitch change or final lengthening as single cues is sufficient, similar to what has been observed for adult speakers of German (Holzgrefe-Lang et al., 2016).
The present dissertation provides new scientific knowledge on infants’ sensitivity to individual prosodic phrase boundary cues in the first year of life. Methodologically, the studies are pathbreaking since they used exactly the same stimulus materials – phonologically thoroughly controlled lists of names – that have also been used with adults (Holzgrefe-Lang et al., 2016) and with infants in a neurophysiological paradigm (Holzgrefe-Lang, Wellmann, Höhle, & Wartenburger, 2018), allowing for comparisons across age (six/ eight months and adults) and method (behavioral vs. neurophysiological methods). Moreover, materials are suited to be transferred to other languages allowing for a crosslinguistic comparison. Taken together with a study with similar French materials (van Ommen et al., 2020) the observed change in sensitivity in German-learning infants can be interpreted as a language-specific one, from an initial language-general processing mechanism that primarily focuses on the presence of pauses to a language-specific processing that takes into account prosodic properties available in the ambient language. The developmental pattern is discussed as an interplay of acoustic salience, prosodic typology (prosodic regularity) and cue reliability.
Difficulties with object relative clauses (ORC), as compared to subject relative clauses (SR), are widely attested across different languages, both in adults and in children. This SR-ORC asymmetry is reduced, or even eliminated, when the embedded constituent in the ORC is a pronoun, rather than a lexical noun phrase. The studies included in this thesis were designed to explore under what circumstances the pronoun facilitation occurs; whether all pronouns have the same effect; whether SRs are also affected by embedded pronouns; whether children perform like adults on such structures; and whether performance is related to cognitive abilities such as memory or grammatical knowledge. Several theoretical approaches that explain the pronoun facilitation in relative clauses are evaluated. The experimental data have been collected in three languages–German, Italian and Hebrew–stemming from both children and adults.
In the German study (Chapter 2), ORCs with embedded 1st- or 3rd-person pronouns are compared to ORCs with an embedded lexical noun phrase. Eye-movement data from 5-year-old children show that the 1st-person pronoun facilitates processing, but not the 3rd-person pronoun. Moreover, children’s performance is modulated by additive effects of their memory and grammatical skills. In the Italian study (Chapter 3), the 1st-person pronoun advantage over the 3rd-person pronoun is tested in ORCs and SRs that display a similar word order. Eye-movement data from 5-year-olds and adult controls and reading times data from adults are pitted against the outcome of a corpus analysis, showing that the 1st-/3rd-person pronoun asymmetry emerges in the two relative clause types to an equal extent. In the Hebrew study (Chapter 4), the goal is to test the effect of a special kind of pronoun–a non-referential arbitrary subject pronoun–on ORC comprehension, in the light of potential confounds in previous studies that used this pronoun. Data from a referent-identification task with 4- to 5-year-olds indicate that, when the experimental material is controlled, the non-referential pronoun does not necessarily facilitate ORC comprehension. Importantly, however, children have even more difficulties when the embedded constituent is a referential pronoun. The non-referentiality / referentiality asymmetry is emphasized by the relation between children’s performance on the experimental task and their memory skills.
Together, the data presented in this thesis indicate that sentence processing is not only driven by structural (or syntactic) factors, but also by discourse-related ones, like pronouns’ referential properties or their discourse accessibility mechanism, which is defined as the level of ease or difficulty with which referents of pronouns are identified and retrieved from the discourse model. Although independent in essence, these structural and discourse factors can in some cases interact in a way that affects sentence processing. Moreover, both types of factors appear to be strongly related to memory. The data also support the idea that, from early on, children are sensitive to the same factors that affect adults’ sentence processing, and that the processing strategies of both populations are qualitatively similar.
In sum, this thesis suggests that a comprehensive theory of human sentence processing needs to account for effects that are due to both structural and discourse-related factors, which operate as a function of memory capacity.
Exhaustivity
(2016)
The dissertation proposes an answer to the question of how to model exhaustive inferences and what the meaning of the linguistic material that triggers these inferences is. In particular, it deals with the semantics of exclusive particles, clefts, and progressive aspect in Ga, an under-researched language spoken in Ghana. Based on new data coming from the author’s original fieldwork in Accra, the thesis points to a previously unattested variation in the semantics of exclusives in a cross-linguistic perspective, analyzes the connections between exhaustive interpretation triggered by clefts and the aspectual interpretation of the sentence, and identifies a cross-categorial definite determiner. By that it sheds new light on several exhaustivity-related phenomena in both the nominal and the verbal domain and shows that both domains are closely connected.
This work addresses issues in the automatic preprocessing of historical German input text for use by conventional natural language processing techniques. Conventional techniques cannot adequately account for historical input text due to conventional tools' reliance on a fixed application-specific lexicon keyed by contemporary orthographic surface form on the one hand, and the lack of consistent orthographic conventions in historical input text on the other. Historical spelling variation is treated here as an error-correction problem or "canonicalization" task: an attempt to automatically assign each (historical) input word a unique extant canonical cognate, thus allowing direct application-specific processing (tagging, parsing, etc.) of the returned canonical forms without need for any additional application-specific modifications. In the course of the work, various methods for automatic canonicalization are investigated and empirically evaluated, including conflation by phonetic identity, conflation by lemma instantiation heuristics, canonicalization by weighted finite-state rewrite cascade, and token-wise disambiguation by a dynamic Hidden Markov Model.
The main research question of this thesis concerns the relation between focus interpretation, focus realization, and association with focus in the West Chadic language Ngamo.
Concerning the relation between focus realization and interpretation, this thesis contributes to the question, cross-linguistically, what factors influence a marked realization of the focus/background distinction. There is background-marking rather than focus-marking in Ngamo, and the background marker is related to the definite determiner in the language. Using original fieldwork data as a basis, a formal semantic analysis of the background marker as a definite determiner of situations is proposed.
Concerning the relation between focus and association with focus, the thesis adds to the growing body of crosslinguistic evidence that not all so-called focus-sensitive operators always associate with focus. The thesis shows that while the exclusive particle yak('i) (= "only") in Ngamo conventionally associates
with focus, the particles har('i) (= "even, as far as, until, already"), and ke('e) (= "also, and") do not.
The thesis provides an analysis of these phenomena in a situation semantic framework.
Èto-clefts are Russian focus constructions with the demonstrative pronoun èto ‘this’ at the beginning: “Èto Mark vyigral gonku” (“It was Mark who won the race”). They are often being compared with English it-clefts, German es-clefts, as well as the corresponding focus-background structures in other languages.
In terms of semantics, èto-clefts have two important properties which are cross-linguistically typical for clefts: existence presupposition (“Someone won the race”) and exhaustivity (“Nobody except Mark won the race”). However, the exhaustivity effects are not as strong as exhaustivity effects in structures with the exclusive only and require more research.
At the same time, the question if the syntactic structure of èto-clefts matches the biclausal structure of English and German clefts, remains open. There are arguments in favor of biclausality, as well as monoclausality. Besides, there is no consistency regarding the status of èto itself.
Finally, the information structure of èto-clefts has remained underexplored in the existing literature.
This research investigates the information-structural, syntactic, and semantic properties of Russian clefts, both theoretically (supported by examples from Russian text corpora and judgments from native speakers) and experimentally. It is determined which desired changes in the information structure motivate native speakers to choose an èto-cleft and not the canonical structure or other focus realization tools. Novel syntactic tests are conducted to find evidence for bi-/monoclausality of èto-clefts, as well as for base-generation or movement of the cleft pivot. It is hypothesized that èto has a certain important function in clefts, and its status is investigated. Finally, new experiments on the nature of exhaustivity in èto-clefts are conducted. They allow for direct cross-linguistic comparison, using an incremental-information paradigm with truth-value judgments.
In terms of information structure, this research makes a new proposal that presents èto-clefts as structures with an inherent focus-background bipartitioning. Even though èto-clefts are used in typical focus contexts, evidence was found that èto-clefts (as well as Russian thetic clefts) allow for both new information focus and contrastive focus. Èto-clefts are pragmatically acceptable when a singleton answer to the implied question is expected (e.g. “It was Mark who won the race” but not “It was Mark who came to the party”). Importantly, èto in Russian clefts is neither dummy, nor redundant, but is a topic expression; conveys familiarity which triggers existence presupposition; refers to an instantiated event, or a known/perceivable situation; finally, èto plays an important role in the spoken language as a tool for speech coherency and a focus marker.
In terms of syntax, this research makes a new monoclausal proposal and shows evidence that the cleft pivot undergoes movement to the left peripheral position. Èto is proposed to be TopP.
Finally, in terms of semantics, a novel cross-linguistic evaluation of Russian clefts is made. Experiments show that the exhaustivity inference in èto-clefts is not robust. Participants used different strategies in resolving exhaustivity, falling into 2 groups: one group considered èto-clefts exhaustive, while another group considered them non-exhaustive. Hence, there is evidence for the pragmatic nature of exhaustivity in èto-clefts. The experimental results for èto-clefts are similar to the experimental results for clefts in German, French and Akan. It is concluded that speakers use different tools available in their languages to produce structures with similar interpretive properties.
The present work is a compilation of three original research articles submitted (or already published) in international peer-reviewed venues of the field of speech science. These three articles address the topics of fundamental motor laws in speech and dynamics of corresponding speech movements:
1. Kuberski, Stephan R. and Adamantios I. Gafos (2019). "The speed-curvature power law in tongue movements of repetitive speech". PLOS ONE 14(3). Public Library of Science. doi: 10.1371/journal.pone.0213851.
2. Kuberski, Stephan R. and Adamantios I. Gafos (In press). "Fitts' law in tongue movements of repetitive speech". Phonetica: International Journal of Phonetic Science. Karger Publishers. doi: 10.1159/000501644
3. Kuberski, Stephan R. and Adamantios I. Gafos (submitted). "Distinct phase space topologies of identical phonemic sequences". Language. Linguistic Society of America.
The present work introduces a metronome-driven speech elicitation paradigm in which participants were asked to utter repetitive sequences of elementary consonant-vowel syllables. This paradigm, explicitly designed to cover speech rates from a substantially wider range than has been explored so far in previous work, is demonstrated to satisfy the important prerequisites for assessing so far difficult to access aspects of speech. Specifically, the paradigm's extensive speech rate manipulation enabled elicitation of a great range of movement speeds as well as movement durations and excursions of the relevant effectors. The presence of such variation is a prerequisite to assessing whether invariant relations between these and other parameters exist and thus provides the foundation for a rigorous evaluation of the two laws examined in the first two contributions of this work.
In the data resulting from this paradigm, it is shown that speech movements obey the same fundamental laws as movements from other domains of motor control do. In particular, it is demonstrated that speech strongly adheres to the power law relation between speed and curvature of movement with a clear speech rate dependency of the power law's exponent. The often-sought or reported exponent of one third in the statement of the law is unique to a subclass of movements which corresponds to the range of faster rates under which a particular utterance is produced. For slower rates, significantly larger values than one third are observed. Furthermore, for the first time in speech this work uncovers evidence for the presence of Fitts' law. It is shown that, beyond a speaker-specific speech rate, speech movements of the tongue clearly obey Fitts' law by emergence of its characteristic linear relation between movement time and index of difficulty. For slower speech rates (when temporal pressure is small), no such relation is observed. The methods and datasets obtained in the two assessment above provide a rigorous foundation both for addressing implications for theories and models of speech as well as for better understanding the status of speech movements in the context of human movements in general.
All modern theories of language rely on a fundamental segmental hypothesis according to which the phonological message of an utterance is represented by a sequence of segments or phonemes. It is commonly assumed that each of these phonemes can be mapped to some unit of speech motor action, a so-called speech gesture.
For the first time here, it is demonstrated that the relation between the phonological description of simple utterances and the corresponding speech motor action is non-unique. Specifically, by the extensive speech rate manipulation in the herein used experimental paradigm it is demonstrated that speech exhibits clearly distinct dynamical organizations underlying the production of simple utterances. At slower speech rates, the dynamical organization underlying the repetitive production of elementary /CV/ syllables can be described by successive concatenations of closing and opening gestures, each with its own equilibrium point. As speech rate increases, the equilibria of opening and closing gestures are not equally stable yielding qualitatively different modes of organization with either a single equilibrium point of a combined opening-closing gesture or a periodic attractor unleashed by the disappearance of both equilibria. This observation, the non-uniqueness of the dynamical organization underlying what on the surface appear to be identical phonemic sequences, is an entirely new result in the domain of speech. Beyond that, the demonstration of periodic attractors in speech reveals that dynamical equilibrium point models do not account for all possible modes of speech motor behavior.
This dissertation examines the impact of the type of referring expression on the acquisition of word order variation in German-speaking preschoolers. A puzzle in the area of language acquisition concerns the production-comprehension asymmetry for non-canonical sentences like "Den Affen fängt die Kuh." (“The monkey, the cow chases.”), that is, preschoolers usually have difficulties in accurately understanding non-canonical sentences approximately until age six (e.g., Dittmar et al., 2008) although they produce non-canonical sentences already around age three (e.g., Poeppel & Wexler, 1993; Weissenborn, 1990). This dissertation investigated the production and comprehension of non-canonical sentences to address this issue.
Three corpus analyses were conducted to investigate the impact of givenness, topic status and the type of referring expression on word order in the spontaneous speech of two- to four-year-olds and the child-directed speech produced by their mothers. The positioning of the direct object in ditransitive sentences was examined; in particular, sentences in which the direct object occurred before or after the indirect object in the sentence-medial positions and sentences in which it occurred in the sentence-initial position. The results reveal similar ordering patterns for children and adults. Word order variation was to a large extent predictable from the type of referring expression, especially with respect to the word order involving the sentence-medial positions. Information structure (e.g., topic status) had an additional impact only on word order variation that involved the sentence-initial position.
Two comprehension experiments were conducted to investigate whether the type of referring expression and topic status influences the comprehension of non-canonical transitive sentences in four- and five-year-olds. In the first experiment, the topic status of the one of the sentential arguments was established via a preceding context sentence, and in the second experiment, the type of referring expression for the sentential arguments was additionally manipulated by using either a full lexical noun phrase (NP) or a personal pronoun. The results demonstrate that children’s comprehension of non-canonical sentences improved when the topic argument was realized as a personal pronoun and this improvement was independent of the grammatical role of the arguments. However, children’s comprehension was not improved when the topic argument was realized as a lexical NP.
In sum, the results of both production and comprehension studies support the view that referring expressions may be seen as a sentence-level cue to word order and to the information status of the sentential arguments. The results highlight the important role of the type of referring expression on the acquisition of word order variation and indicate that the production-comprehension asymmetry is reduced when the type of referring expression is considered.
Interactive generation of effective discourse in situated context : a planning-based approach
(2013)
As our modern-built structures are becoming increasingly complex, carrying out basic tasks such as identifying points or objects of interest in our surroundings can consume considerable time and cognitive resources. In this thesis, we present a computational approach to converting contextual information about a person's physical environment into natural language, with the aim of helping this person identify given task-related entities in their environment. Using efficient methods from automated planning - the field of artificial intelligence concerned with finding courses of action that can achieve a goal -, we generate discourse that interactively guides a hearer through completing their task. Our approach addresses the challenges of controlling, adapting to, and monitoring the situated context. To this end, we develop a natural language generation system that plans how to manipulate the non-linguistic context of a scene in order to make it more favorable for references to task-related objects. This strategy distributes a hearer's cognitive load of interpreting a reference over multiple utterances rather than one long referring expression. Further, to optimize the system's linguistic choices in a given context, we learn how to distinguish speaker behavior according to its helpfulness to hearers in a certain situation, and we model the behavior of human speakers that has been proven helpful. The resulting system combines symbolic with statistical reasoning, and tackles the problem of making non-trivial referential choices in rich context. Finally, we complement our approach with a mechanism for preventing potential misunderstandings after a reference has been generated. Employing remote eye-tracking technology, we monitor the hearer's gaze and find that it provides a reliable index of online referential understanding, even in dynamically changing scenes. We thus present a system that exploits hearer gaze to generate rapid feedback on a per-utterance basis, further enhancing its effectiveness. Though we evaluate our approach in virtual environments, the efficiency of our planning-based model suggests that this work could be a step towards effective conversational human-computer interaction situated in the real world.
Neural conversation models aim to predict appropriate contributions to a (given) conversation by using neural networks trained on dialogue data. A specific strand focuses on non-goal driven dialogues, first proposed by Ritter et al. (2011): They investigated the task of transforming an utterance into an appropriate reply. Then, this strand evolved into dialogue system approaches using long dialogue histories and additional background context. Contributing meaningful and appropriate to a conversation is a complex task, and therefore research in this area has been very diverse: Serban et al. (2016), for example, looked into utilizing variable length dialogue histories, Zhang et al. (2018) added additional context to the dialogue history, Wolf et al. (2019) proposed a model based on pre-trained Self-Attention neural networks (Vasvani et al., 2017), and Dinan et al. (2021) investigated safety issues of these approaches. This trend can be seen as a transformation from trying to somehow carry on a conversation to generating appropriate replies in a controlled and reliable way.
In this thesis, we first elaborate the meaning of appropriateness in the context of neural conversation models by drawing inspiration from the Cooperative Principle (Grice, 1975). We first define what an appropriate contribution has to be by operationalizing these maxims as demands on conversation models: being fluent, informative, consistent towards given context, coherent and following a social norm. Then, we identify different targets (or intervention points) to achieve the conversational appropriateness by investigating recent research in that field.
In this thesis, we investigate the aspect of consistency towards context in greater detail, being one aspect of our interpretation of appropriateness.
During the research, we developed a new context-based dialogue dataset (KOMODIS) that combines factual and opinionated context to dialogues. The KOMODIS
dataset is publicly available and we use the data in this thesis to gather new insights in context-augmented dialogue generation.
We further introduced a new way of encoding context within Self-Attention based neural networks. For that, we elaborate the issue of space complexity from knowledge graphs,
and propose a concise encoding strategy for structured context inspired from graph neural networks (Gilmer et al., 2017) to reduce the space complexity of the additional context. We discuss limitations of context-augmentation for neural conversation models, explore the characteristics of knowledge graphs, and explain how we create and augment knowledge graphs for our experiments.
Lastly, we analyzed the potential of reinforcement and transfer learning to improve context-consistency for neural conversation models. We find that current reward functions need to be more precise to enable the potential of reinforcement learning, and that sequential transfer learning can improve the subjective quality of generated dialogues.
In this thesis sentence processing was investigated using a psychophysiological measure known as pupillometry as well as Event-Related Potentials (ERP). The scope of the the- sis was broad, investigating the processing of several different movement constructions with native speakers of English and second language learners of English, as well as word order and case marking in German speaking adults and children. Pupillometry and ERP allowed us to test competing linguistic theories and use novel methodologies to investigate the processing of word order. In doing so we also aimed to establish pupillometry as an effective way to investigate the processing of word order thus broadening the methodological spectrum.
This thesis investigates the processing and representation of (ir-)regularity in inflectional verb morphology in German and English. The focus lies on the predictions from models of morphological processing about the production of subtypes of irregular verbs which are usually subsumed under the category `irregular verbs'. Thus, this dissertation presents three journal articles investigating the language production of healthy speakers and speakers with agrammatic aphasia in order to fill a gap both for the availability of language production data and systematically tested patterns of irregularity. The second Chapter set out to investigate whether regularity of a verb or its phonological complexity (measured in number of phonemes) better predict the production accuracies of German speakers with agrammatic aphasia. While regular verbs were significantly more often correct than mixed and irregular verbs, production accuracies of irregular and mixed verbs for impaired participants did not differ. Thus, no influence of phonological complexity was observed. Chapter 3 aimed at teasing apart the influence of stem changes and affix type on the production accuracies of English speaking individuals with agrammatic aphasia. The analyses revealed that the presence of stem changes but not the type of affix had a significant effect on the production accuracies. Moreover, as four different verb types were tested, results showed that production accuracies did not conform to a regular-irregular distinction but that accuracies differed by the degree of regularity. In Chapter 4, long-lag primed picture naming design was used to study if the differences found in the production accuracies of Chapter 3 were also associated with differences in production latencies of non-brain damaged speakers. A morphological priming effect was found, however, in neither experiment the effect differed of the three verb types tested. In addition to standard frequentist analysis, Bayesian analysis were performed. In this way the absence of a difference of the morphological priming effect between verb types was interpreted as actual evidence for the lack of such a difference. Hence, this thesis presents diverging results on the production of subtypes of irregular verbs in healthy and impaired adult speakers. However, at the same time these results provided evidence that the conventional regular-irregular distinction is not adequate for testing models of morphological processing.
Learners' Little Helper
(2015)
This dissertation is about factors that contribute to the surface forms of tones in connected speech in Akan. Akan is an African tone language, which is spoken in Ghana. It has two level tones (low and high), automatic and non-automatic downstep. Downstep is the major factor that influences the surface forms of tones. The thesis shows that downstep is caused by declination. It is argued that declination is an intonational property of Akan, which serves to signal coherence. A phonological representation using a high and a low register tone, associating to the left and right edge of an intonational phrase (IP), respectively, is proposed. Declination/downstep is modelled using a (phonetic) pitch implementation algorithm (Liberman & Pierrehumbert, 1984). An innovative application of the algorithm is presented, which naturally captures the relation between declination and downstep in Akan. Another important factor is the prosodic manifestation of sentence level pragmatic meanings, such as sentence mode and focus. Regarding the former, the thesis shows that a post-lexical low tone, which associates with the right edge of an IP, signals interrogativity. Additionally, lexical tones in Yes – No questions are realized in a higher pitch register, which does not lead to a reduction of declination. It is claimed that the higher register is not part of the phonological representation in Akan, but that it emerges at the phonetic level to compensate for the ‘unnatural’ form of the question morpheme and to satisfy the Frequency code (Gussenhoven, 2002; 2004). An extension of Rialland’s (2007) typology in terms of a new category called “low tense” question prosody is proposed. Concerning focus marking, it is argued that the use of the morpho-syntactic focus marking strategy is related to extra grammatical factors, such as hearer expectation, discourse expectability (Zimmermann, 2007) and emphasis (Hartmann, 2008). If a speaker of Akan wants to highlight a particular element in a sentence, in-situ, i.e. by means of prosody, the default prosodic structure is modified in such a way that the focused element forms its own phonological phrase (pP). If it is already contained in a pP, the boundary deliminating the focused element is enhanced (Féry, 2012). This restructuring/enhancement is accompanied by an interruption of the otherwise continuous melody due to insertion of a pause and/or a glottal stop. Beside declination and intonation, raising of H tones applies in Akan. H raising is analyzed as a local anticipatory planning effect, employed at the phonetic level, which enhances the perceptual distance between low and high tones. Low tones are raised, if they are wedged between two high tones. L raising is argued to be a local carryover effect (co-articulation). Further, it is demonstrated that global anticipatory raising takes place. It is shown that Akan speakers anticipate the length of an IP. Preplanning (anticipatory raising) is argued to be an important process at the level of pitch implementation. It serves to ensure that declination can be maintained throughout the IP, which prevents pitch resetting.
The melody of an Akan sentence is largely determined by the choice of words. The inventory of post-lexical tones is small. It consists of post-lexical register tones, which trigger declination and post-lexical intonational tones, which signal sentence type. The overall melodic shape is falling. At the local level, H raising and L raising occur. At the global level, initial low and high tones are realized higher if they occur in a long and/or complex sentence. This dissertation shows that many factors, which emerge at different levels of the tone production process, contribute to the surface form of tones in Akan.
Interlocutors typically link their utterances to the discourse environment and enrich communication by linguistic (e.g., information packaging) and extra-linguistic (e.g., eye gaze, gestures) means to optimize information transfer. Psycholinguistic studies underline that ‒for meaning computation‒ listeners profit from linguistic and visual cues that draw their focus of attention to salient information. This dissertation is the first work that examines how linguistic compared to visual salience cues influence sentence comprehension using the very same experimental paradigms and materials, that is, German subject-before-object (SO) and object-before-subject (OS) sentences, across the two cue modalities. Linguistic salience was induced by indicating a referent as the aboutness topic. Visual salience was induced by implicit (i.e., unconscious) or explicit (i.e., shared) manipulations of listeners’ attention to a depicted referent.
In Study 1, a selective, facilitative impact of linguistic salience on the context-sensitive OS word order was found using offline comprehensibility judgments. More precisely, during online sentence processing, this impact was characterized by a reduced sentence-initial Late positivity which reflects reduced processing costs for updating the current mental representation of discourse. This facilitative impact of linguistic salience was not replicated by means of an implicit visual cue (Study 2) shown to modulate word order preferences during sentence production. However, a gaze shift to a depicted referent as an indicator of shared attention eased sentence-initial processing similar to linguistic salience as revealed by reduced reading times (Study 3). Yet, this cue did not modulate the strong subject-antecedent preference during later pronoun resolution like linguistic salience. Taken together, these findings suggest a significant impact of linguistic and visual salience cues on sentence comprehension, which substantiates that both the information delivered via language and via the visual environment is integrated into the mental representation of the discourse; but, the way how salience is induced is crucial to its impact.
Infants' lexical processing is modulated by featural manipulations made to words, suggesting that early lexical representations are sufficiently specified to establish a match with the corresponding label. However, the precise degree of detail in early words requires further investigation due to equivocal findings. We studied this question by assessing children’s sensitivity to the degree of featural manipulation (Chapters 2 and 3), and sensitivity to the featural makeup of homorganic and heterorganic consonant clusters (Chapter 4). Gradient sensitivity on the one hand and sensitivity to homorganicity on the other hand would suggest that lexical processing makes use of sub-phonemic information, which in turn would indicate that early words contain sub-phonemic detail. The studies presented in this thesis assess children’s sensitivity to sub-phonemic detail using minimally demanding online paradigms suitable for infants: single-picture pupillometry and intermodal preferential looking. Such paradigms have the potential to uncover lexical knowledge that may be masked otherwise due to cognitive limitations. The study reported in Chapter 2 obtained a differential response in pupil dilation to the degree of featural manipulation, a result consistent with gradient sensitivity. The study reported in Chapter 3 obtained a differential response in proportion of looking time and pupil dilation to the degree of featural manipulation, a result again consistent with gradient sensitivity. The study reported in Chapter 4 obtained a differential response to the manipulation of homorganic and heterorganic consonant clusters, a result consistent with sensitivity to homorganicity. These results suggest that infants' lexical representations are not only specific, but also detailed to the extent that they contain sub-phonemic information.
This dissertation addresses the question of how linguistic structures can be represented in working memory. We propose a memory-based computational model that derives offline and online complexity profiles in terms of a top-down parser for minimalist grammars (Stabler, 2011). The complexity metric reflects the amount of time an item is stored in memory. The presented architecture links grammatical representations stored in memory directly to the cognitive behavior by deriving predictions about sentence processing difficulty.
Results from five different sentence comprehension experiments were used to evaluate the model's assumptions about memory limitations. The predictions of the complexity metric were compared to the locality (integration and storage) cost metric of Dependency Locality Theory (Gibson, 2000). Both metrics make comparable offline and online predictions for four of the five phenomena. The key difference between the two metrics is that the proposed complexity metric accounts for the structural complexity of intervening material. In contrast, DLT's integration cost metric considers the number of discourse referents, not the syntactic structural complexity.
We conclude that the syntactic analysis plays a significant role in memory requirements of parsing. An incremental top-down parser based on a grammar formalism easily computes offline and online complexity profiles, which can be used to derive predictions about sentence processing difficulty.
Minimalism, Features and Parallel Grammars : on the acquisition of German ditransitive structures
(2004)
An important strand of research has investigated the question of how children acquire a morphological system using offline data from spontaneous or elicited child language. Most of these studies have found dissociations in how children apply regular and irregular inflection (Marcus et al. 1992, Weyerts & Clahsen 1994, Rothweiler & Clahsen 1993). These studies have considerably deepened our understanding of how linguistic knowledge is acquired and organised in the human mind. Their methodological procedures, however, do not involve measurements of how children process morphologically complex forms in real time. To date, little is known about how children process inflected word forms. The aim of this study is to investigate children’s processing of inflected words in a series of on-line reaction time experiments. We used a cross-modal priming experiment to test for decompositional effects on the central level. We used a speeded production task and a lexical decision task to test for frequency effects on access level in production and recognition. Children’s behaviour was compared to adults’ behaviour towards three participle types (-t participles, e.g. getanzt ‘danced’ vs. -n participles with stem change, e.g. gebrochen ‘broken’ vs.-n participles without stem change, e.g. geschlafen ‘slept’). For the central level, results indicate that -t participles but not -n participles have decomposed representations. For the access level, results indicate that -t participles are represented according to their morphemes and additionally as full forms, at least from the age of nine years onwards (Pinker 1999 and Clahsen et al. 2004). Further evidence suggested that -n participles are represented as full-form entries on access level and that -n participles without stem change may encode morphological structure (cf. Clahsen et al. 2003). Out data also suggests that processing strategies for -t participles are differently applied in recognition and production. These results provide evidence that children (within the age range tested) employ the same mechanisms for processing participles as adults. The child lexicon grows as children form additional full-form representations for -t participles on access level and elaborate their full-form lexical representations of -n participles on central level. These results are consistent with processing as explained in dual-system theories.
Over the last decades mechanisms of recognition of morphologically complex words have been extensively examined in order to determine whether all word forms are stored and retrieved from the mental lexicon as wholes or whether they are decomposed into their morphological constituents such as stems and affixes. Most of the research in this domain focusses on English. Several factors have been argued to affect morphological processing including, for instance, morphological structure of a word (e.g., existence of allomorphic stem alternations) and its linguistic nature (e.g., whether it is a derived word or an inflected word form). It is not clear, however, whether processing accounts based on experimental evidence from English would hold for other languages. Furthermore, there is evidence that processing mechanisms may differ across various populations including children, adult native speakers and language learners. Recent studies claim that processing mechanisms could also differ between older and younger adults (Clahsen & Reifegerste, 2017; Reifegerste, Meyer, & Zwitserlood, 2017).
The present thesis examined how properties of the morphological structure, types of linguistic operations involved (i.e., the linguistic contrast between inflection and derivation) and characteristics of the particular population such as older adults (e.g., potential effects of ageing as a result of the cognitive decline or greater experience and exposure of older adults) affect initial, supposedly automatic stages of morphological processing in Russian and German. To this end, a series of masked priming experiments was conducted.
In experiments on Russian, the processing of derived -ost’ nouns (e.g., glupost’ ‘stupidity’) and of inflected forms with and without allomorphic stem alternations in 1P.Sg.Pr. (e.g., igraju – igrat’ ‘to play’ vs. košu – kosit’ ‘to mow’) was examined. The first experiment on German examined and directly compared processing of derived -ung nouns (e.g., Gründung ‘foundation’) and inflected -t past participles (e.g., gegründet ‘founded’), whereas the second one investigated the processing of regular and irregular plural forms (-s forms such as Autos ‘cars’ and -er forms such as Kinder ‘children’, respectively).
The experiments on both languages have shown robust and comparable facilitation effects for derived words and regularly inflected forms without stem changes (-t participles in German, forms of -aj verbs in Russian). Observed morphological priming effects could be clearly distinguished from purely semantic or orthographic relatedness between words. At the same time, we found a contrast between forms with and without allomorphic stem alternations in Russian and regular and irregular forms in German, with significantly more priming for unmarked stems (relative to alternated ones) and significantly more priming for regular (compared) word forms. These findings indicate the relevance of morphological properties of a word for initial stages of processing, contrary to claims made in the literature holding that priming effects are determined by surface form and meaning overlap only. Instead, our findings are more consistent with approaches positing a contrast between combinatorial, rule-based and lexically-stored forms (Clahsen, Sonnenstuhl, & Blevins, 2003).
The doctoral dissertation also addressed the role of ageing and age-related cognitive changes on morphological processing. The results obtained on this research issue are twofold. On the one hand, the data demonstrate effects of ageing on general measures of language performance, i.e., overall longer reaction times and/or higher accuracy rates in older than younger individuals. These findings replicate results from previous studies, which have been linked to the general slowing of processing speed at older age and to the larger vocabularies of older adults. One the other hand, we found that more specific aspects of language processing appear to be largely intact in older adults as revealed by largely similar morphological priming effects for older and younger adults. These latter results indicate that initial stages of morphological processing investigated here by means of the masked priming paradigm persist in older age. One caveat should, however, be noted. Achieving the same performance as a younger individual in a behavioral task may not necessarily mean that the same neural processes are involved. Older people may have to recruit a wider brain network than younger individuals, for example. To address this and related possibilities, future studies should examine older people’s neural representations and mechanisms involved in morphological processing.
For several decades, researchers have tried to explain how speakers of more than one language (multilinguals) manage to keep their languages separate and to switch from one language to the other depending on the context. This ability of multilingual speakers to use the intended language, while avoiding interference from the other language(s) has recently been termed “language control”.
A multitude of studies showed that when bilinguals process one language, the other language is also activated and might compete for selection. According to the most influential model of language control developed over the last two decades, competition from the non-intended language is solved via inhibition. In particular, the Inhibitory Control (IC) model proposed by Green (1998) puts forward that the amount of inhibition applied to the non-relevant language depends on its dominance, in that the stronger the language the greater the strength of inhibition applied to it. Within this account, the cost required to reactivate a previously inhibited language depends on the amount of inhibition previously exerted on it, that is, reactivation costs are greater for a stronger compared to a weaker language. In a nutshell, according to the IC model, language control is determined by language dominance.
The goal of the present dissertation is to investigate the extent to which language control in multilinguals is affected by language dominance and whether and how other factors might influence this process. Three main factors are considered in this work: (i) the time speakers have to prepare for a certain language or PREPARATION TIME, (ii) the type of languages involved in the interactional context or LANGUAGE TYPOLOGY, and (iii) the PROCESSING MODALITY, that is, whether the way languages are controlled differs between reception and production.
The results obtained in the four manuscripts, either published or in revision, indicate that language dominance alone does not suffice to explain language switching patterns. In particular, the present thesis shows that language control is profoundly affected by each of the three variables described above. More generally, the findings obtained in the present dissertation indicate that language control in multilingual speakers is a much more dynamic system than previously believed and is not exclusively determined by language dominance, as predicted by the IC model (Green, 1998).
Previous behavioral studies showed that perceptual changes in infancy can be observed in multiple patterns, namely decline (e.g., Mattock et al., 2008; Yeung et al., 2013), maintenance (e.g., Chen & Kager, 2016) and U-shaped development (Liu & Kager, 2014).
This dissertation contributes further to the understanding of the developmental trajectory of phonological acquisition in infancy. The dissertation addresses the questions of how the perceptual sensitivity of lexical tones and vowels changes in infancy and how different experimental procedures contribute to our understanding. We used three experimental procedures to investigate German-learning infants’ discrimination abilities. In Studies 1 and 3 (Chapters 5 and 7) we used behavioral methods (habituation and familiarization procedures) and in Study 2 (Chapter 6) we measured neural correlates.
Study 1 showed a U-shaped developmental pattern: 6- and 18-month-olds discriminated a lexical tone contrast, but not the 9-month-olds. In addition, we found an effect of experimental procedure: infants discriminated the tone contrast at 6 months in a habituation but not in a familiarization procedure. In Study 2, we observed mismatch responses (MMR) to a non-native tone contrast and a native-like vowel in 6- and 9-month-olds. In 6-month-olds, both contrasts elicited positive MMRs. At 9 months, the vowel contrast elicited an adult-like negative MMR, while the tone contrast elicited a positive MMR. Study 3 demonstrated a change in perceptual sensitivity to a vowel contrast between 6 and 9 months. In contrast to the 6-month-old infants, the 9-month-old infants discriminated the tested vowel contrast asymmetrically.
We suggest that the shifts in perceptual sensitivity between 6 and 9 months are functional rather than perceptual. In the case of lexical tone discrimination, infants may have already learned by 9 months of age that pitch is not relevant at the lexical level in German, since the infants in Study 1 showed no perceptual sensitivity to the contrast tested. Nevertheless, the brain responded to the contrast, especially since pitch differences are also part of the German intonation system (Gussenhoven, 2004). The role of the intonation system in pitch discrimination could be supported by the recovery of behavioral discrimination at 18 months of age, as well as behavioral and neural discrimination in German-speaking adults.