Studies of metaphors in teaching and learning have underlined the important role of metaphors in reasoning, but have sometimes failed to show the effect of metaphor on how scientific concepts are represented, and have sometimes overlooked hidden metaphors in their attempts to be explicit about how metaphor functions. This study investigates metaphor in the context of teaching environmental science. It does not assume any simple correlation between surface linguistic cues and the presence or kind of metaphor. Two theoretical approaches have been chosen, Systemic Functional Linguistics (M. Halliday) which sees language as a social construction of meaning, and Image Schema (M Johnson and G Lakoff) which has developed in cognitive science and cognitive linguistics. These two approaches are used to discuss examples of metaphors from a number of lessons which have been observed and video-recorded, and in a variety of textbooks used as resource materials in teaching environmental science. The choice of environmental science as the subject matter arises from two of its distinct characteristics. One is the fact that ideology triggers and shapes the interests, decisions and choices of materials, issues, arguments, reasons, etc. But there is nothing like one unique ideology, on the contrary conflicts of different ideologies raise differences about what will be selected and how it will be represented. At this point there is a special role taken on by metaphor. Metaphors provide the means for creating differences and similarities, thus bringing together or keeping apart ideologies. Second, the teaching of environmental science does not appear as the teaching of science only, bounded from anything else, but is a blend of accounts of scientific and commonsense knowledge. Metaphors appear at the overlapping points where this blending takes place. It is not the purpose of the thesis to question, or to contribute to, the theoretical perspectives adopted. Rather, its interest is in how these perspectives provide, each in their own way, insights into the nature of the discourse of teaching environmental science, and thus to raise questions about its effectiveness.

With the prevalence of large data stored in the cloud, including unstructured information in the form of text, there is now an increased emphasis on text mining. A broad range of techniques are now used for text mining, including algorithms adapted from machine learning, NLP, computational linguistics, and data mining. Applications are also multi-fold, including classification, clustering, segmentation, relationship discovery, and practically any task that discovers latent information from written natural language. Classical mining algorithms have traditionally focused on shallow representations such as bag-of-words and similar feature-based models. With the advent of modern high performance computing, deep sentence level linguistic analysis of large scale text corpora has become practical. In this dissertation, we evaluate the utility of dependency parses as textual features for different text mining applications. Dependency parsing is one form of syntactic parsing, based on the dependency grammar implicit in sentences. While dependency parsing has traditionally been used for text understanding, we investigate here its application to supply features for text mining applications. We specifically focus on three methods to construct textual features from dependency parses. First, we consider a dependency parse as a general feature akin to a traditional bag-of-words model. Second, we consider the dependency parse as the basis to build a feature graph representation. Finally, we use dependency parses in a supervised collocation mining method for feature selection. To investigate these three methods, several applications are studied, including: (i) movie spoiler detection, (ii) text segmentation, (iii) query expansion, and (iv) recommender systems. / Ph. D.

Diese Arbeit beschäftigt sich mit der Rolle von Konkordanzmarkierungen innerhalb einer Phrase für die Segmentierung eines fremden Sprachstroms. Das Merkmal der Konkordanz tritt auf, wenn alle Bestandteile einer Phrase gleichermaßen durch eine identische Markierung gekennzeichnet sind (z. B. los muchachos ricosSpanisch = die reichen Männer). Da diese wiederkehrenden Markierungen zumeist aus Affixen bestehen, kann Konkordanz als ein Sonderfall der Flexionsmorphologie betrachtet werden. Es wurde untersucht, ob die formale Korrespondenz zwischen den Bestandteilen konkordanter Phrasen als Hinweis auf die Grenzen der linguistisch relevanten Einheit Phrase im Spracherwerb fungieren kann. Zusätzlich wird auf das Zusammenspiel einzelner Hinweisreize untereinander eingegangen. Mit Kindern im Alter von zehn Monaten wurden vier Experimente mit dem Headturn Preference Paradigma (Jusczyk & Aslin, 1995) durchgeführt. Es wurde zunächst bei deutschen und englischen Kleinkindern untersucht, ob sie sensibel für gleich bleibende Suffixe innerhalb einer Phrase sind und diese für die Segmentierung nutzen können. Außerdem wurde das Zusammenspiel der Hinweise Konkordanz und Prosodie bei der Auffindung von Phrasengrenzen betrachtet. Es zeigte sich, dass deutsche Kinder in besonderer Weise auf konkordante Markierungen reagieren. Neben einer Sensitivität für Konkordanzmarkierungen zeigte das Ergebnis der deutschen Kinder auch, dass sie Flexionssuffixe im Deutschen bereits im Sprachstrom bemerken können. Ein solches Ergebnismuster ließ sich bei den Englisch lernenden Kindern nicht beobachten. Verschiedene Erklärungsmöglichkeiten für diesen Unterschied werden erläutert. Insgesamt weisen die Daten aus den Kindersprachexperimenten darauf hin, dass bereits im Alter von zehn Monaten bei Kindern eine Sensibilität für wiederholt in ähnlicher / gleicher Form auftretende sprachliche Elemente innerhalb der Domäne der Phrase vorhanden ist. Außerdem lassen die Resultate darauf schließen, dass Konkordanzmarkierungen bereits früh zur Segmentierung von kontinuierlicher Sprache verwendet werden. Diese Leistung steht in Zusammenhang mit der Beachtung von statistischen Regularitäten im Sprachstrom. Untersuchungen dazu zeigen, dass m. H. statistischer Lernmechanismen wiederkehrende Elemente im Sprachstrom erkannt werden können (Bonatti, Peña, Nespor, & Mehler, 2005; Newport & Aslin, 2004; Saffran, 2001; Saffran, Aslin & Newport, 1996). Anscheinend ist das Auftreten identischer Segmente innerhalb einer relativ kleinen Domäne im Sprachstrom für Lerner ein hervorstechendes Merkmal, das dazu beiträgt, diese Domäne aus dem Signal hervorzuheben und somit die Segmentierung des Sprachstroms in kleinere Anteile zu unterstützen. Neben den Untersuchungen mit den Kleinkindern wurden zusätzlich drei Reaktionszeitexperimente mit deutschen und englischen Erwachsenen zur Rolle von Konkordanzmarkierungen bei der Verarbeitung der Fremdsprachen Spanisch, Suaheli und (für die englischen Probanden) Deutsch durchgeführt. Das erste Experiment befasste sich mit der Stimulussprache Spanisch, in der es bei Konkordanz zum mehrfachen Auftreten von identischen Suffixen mit Vollvokalen kommt. Dabei war zu beobachten, dass deutsche und englische Muttersprachler die zu erinnernden Phrasen besser in einem kontinuierlichen spanischen Sprachstrom wieder erkannten, wenn die kritischen Phrasen konkordant waren, als wenn sie nicht konkordant waren. Das zweite Experiment verwendete die Stimulussprache Suaheli (konkordante vs. nicht konkordante Präfixe). Dabei zeigte sich ein solches Muster ausschließlich bei den englischen Muttersprachlern. Das dritte Experiment untersuchte englische Muttersprachler mit deutschem Stimulusmaterial, wobei Konkordanz durch Suffixe markiert wird, die aus einer Schwa-Silbe bestehen. Hier ergab sich kein Hinweis für eine Nutzung konkordanter Markierungen bei der Erkennung von Phrasen. Als Grund dafür wird die reduzierte Vokalqualität angenommen, die Schwa-Silben u.U. schwerer wahrnehmbar macht als Vollvokalsilben (z.B. Widera & Portele, 1999; Goméz Lacabex, García Lecumberri, & Cooke, 2005). Es werden weitere Erklärungshypothesen bzgl. der Ergebnisunterschiede bei deutschen und englischen Muttersprachlern beschrieben, die auch auf den Unterschied zwischen der Verarbeitung von konkordanten Suffixen vs. Präfixen eingehen. Zusätzlich erfolgt eine Diskussion der Ergebnisse vor dem Hintergrund von Annahmen über Arten von (nicht-)sprachlichen Ähnlichkeiten und ihren Einfluss auf die Wahrnehmung von ähnlichen Elementen. Die vorliegenden Daten stützen die Annahme von Morgan (1986), dass der Input für einen Sprachlerner bereits zahlreiche Hinweise über die Struktur der jeweiligen Sprache enthält. Sowohl Kleinkinder als auch erwachsene Sprachlerner scheinen für einen beachtlichen Teil dieser Hinweisreize sensibel zu sein. Die bislang kaum beachteten konkordante Markierungen innerhalb von Phrasen scheinen zumindest einen Teil dieser Hinweisreize auszumachen. / This thesis investigates the role of concordant markings in a phrase for the segmentation of a foreign speech stream. The feature concordance appears, when all elements of a syntactic phrase carry the same marker (e.g. los muchachos ricosSpanish = the rich men). These recurring markers are mostly affixes which means that they can be viewed as a special case in inflectional morphology. It was investigated whether the formal correspondance between the elements of concordant phrases might function as a cue to phrase boundaries in language acquisition. Four experiments were done with infants at the age of 10 months using the Headturn Preference Paradigm (Jusczyk & Aslin, 1995). At first it was investigated with German- and English-learning infants whether they are sensitive to recurring suffixes in a phrase and whether they are able to use these elements for phrase segmentation. Furthermore, the interplay between the cues concordance and prosody for the location of phrase boundaries was investigated. It was shown that German-learning infants where sensitive to concordant markings and that they notice inflectional suffixes in the speech stream. Such a result was not observed with the English-learning infants. Various possibilities to explain this difference between German- and English-learning infants are proposed. The results from the experiments with infants show that already 10-months old infants are sensitive to linguistic elements which recur in the domain of the phrase in similar or identical form. Furthermore, the results suggest that concordant markers are used for segmentation of the continuous speech stream from early on. This capacity is closely related with the attention for statistical regularities in the speech stream. Concerning this it was shown that supported by statistical learning mechanisms recurring elements in the speech stream or certain patterns of co-occurrence can be tracked (Bonatti, Peña, Nespor, & Mehler, 2005; Newport & Aslin, 2004; Saffran, 2001; Saffran, Aslin & Newport, 1996). Consequently, it seems that the occurrence of identical segments in a relatively small domain in the speech stream is a salient feature for learners which points out this domain from the signal and therfore supports the segmentation of the speech stream into smaller units. Besides the experiments with infants also three reaction time experiments with German and English adults where carried out. These experiments were concerned with the role of concordant markings for the processing of the foreign languages Spanish, Swahili and (for English participants) German. The first experment used Spanish as the stimulus language. In Spanish concordant phrases each element carries an identical suffix containing a full vowel. For German and English participants it was observed that they recognized a critical phrase form a continuous speech stream significantly better when this phrase was concordant than when it was non-concordant. In the second experiment with adults Swahili stimuli were used where concordant vs. non-concordant prefixes occur. Now, for English but not for German participants an advantage for the recognition of concordant phrases was observed. In the third experiment English participants were tested with German stimuli where concordance is marked by suffixes containing a Schwa-vowel (reduced vowel). The results of this experiments were not significant, suggesting no use of German concordant markings for the recognition of phrases by native English speakers. One reason that might be proposed to explain this result is the reduced vowel appearing in German concordant suffixes which might make them less perceptually salient than syllables containing a full vowel (e.g. Widera & Portele, 1999; Goméz Lacabex, García Lecumberri, & Cooke, 2005). More possiblities to explain the differences in the results between the German and the English participants are given. Furthermore, there is a discussion of the results against the backround of assumptions about kinds of (non-)linguistic similarities and their influence of the perception of similar elements. Altogether the experimental results of this thesis support the assumption of Morgan (1986) that the linguistic input for a learner already contains various cues about the structure of the target language. Infants as well as adults seem to be sensitve for a considerable amount of these cues. Concordant markings in the domain of a phrase seem to be one of these cues.

The present study investigates whether advanced adult L2 speakers comprehend the subtle linguistic cues that modal particles entail and seek to find if modal particles affect them in their responsibility attribution. Two groups of advanced L2 speakers of Swedish were tested; one group of L1 German speakers and one group of L1 English speakers. In an experiment that investigated responsibility attribution, participants read short stories that were manipulated with the modal particles nog and ju, to see if the use of these modal particles affected how they attributed  responsibility to a character in the short story. The L2 learners were tested to see if L1 background affects the L2 acquisition of modal particles. A control group of native Swedish speakers were also tested. As an exploratory and complementary measure, reading times were recorded for the critical sentences modified with ju and nog. The results show a main effect of group and a main effect of condition, but no interaction between the two. However, upon closer inspections of the numerical values in the groups, possible trends and curious directions are seen. The results yielded no significant differences between groups and conditions, and are presented as possible trends, and discussed. Contrary to the hypotheses, these trends are indicative of the English speakers being affected by the modal particles in the way that natives were expected to, while Germans showed a pattern that was different from native speakers. The results show no significant differences for the different conditions in the native control group. The results show no support for L1 transfer facilitation in the acquisition of modal particles.

Les travaux de recherche de cette thèse portent sur l'amélioration des capacités sociales d'un système conversationnel en interaction avec un humain. Lorsque le système n'est pas dédié à une tâche particulière, il doit tenir compte de difficultés relevant de l'interaction sociale elle-même. L'humour est un mécanisme naturel dans les interactions sociales. Nous considérons un humour-machine comme une simulation de comportements simplifiés des capacités humoristiques humaines: dérision, blagues, jeux de mots. Les travaux de cette thèse s'appuient sur des théories issues de domaines variés en sociologie, psychologie, neurosciences et linguistique pour l'intégration de cet humour-machine dans un système robotique. Implémentées dans certains systèmes de dialogue, ces capacités humoristiques sont cependant rarement utilisées pour pouvoir choisir les comportements à générer du robot. Dans nos travaux, la mise en œuvre des comportements humoristiques du système en conversation est réalisée en utilisant la théorie des rites d'interaction. L'estimation de la face de l'interlocuteur permet de diriger le comportement du robot dans la conversation casuelle humoristique. Les facteurs expressifs d'une perte ou d'une valorisation de la face sont liés à des états mentaux exprimés dans une situation donnée. Pour réaliser cette estimation de la face, nous étudions, à partir de corpus créés à cet effet, les réactions comportementales, affectives et expressives des participants à différents types d'humour réalisés par le robot et ayant des impacts variables sur celle-ci (l'humour prenant pour cible le robot, le participant ou un autre sujet). Les réactions des participants à l'humour sont établies sur une représentation multi-niveaux d'indices émotionnels, comportementaux et linguistiques, extraits à partir de l'audio.Des règles sont ainsi construites à partir de l'apprentissage automatique de ces indices issus des corpus, concernant l'appréciation de la réaction des participants à l'humour et la mise à jour de l'estimation de la face présentée du participant. Leur implémentation dans un système automatique nous permet de les évaluer. De nombreuses expérimentations ont été menées avec des publics variés : personnes âgées, adultes, adolescents. Enfin, l'utilisation des préférences du participant à l'humour dans la conversation fait émerger des questions éthiques, notamment face au pouvoir persuasif et manipulateur de l'humour. / This doctoral dissertation is about the improvement of social capacities of a conversationnal system to interact with humans. When the system is not dedicated to one particular task, it must take into account the inherent difficulties of social interaction it-self. Humor is a natural mechanism present in social interactions. We consider humour in a robotic system as a simulation of simplified behaviors from human humor : derision, jokes, puns. This work is based on theories issued from various research domains as sociology, psychology, neurosciences and linguistics to enable integration of humor in a robotic system. Implemented in some dialog systems, humorous capacities are however rarely used when programming the robot’s behavior. In our study, the humourous behavior is implemented in the system by using the ritual theory of face-work. The face analysis of the interlocutor can be used to direct the robot’s reactions during a casual humorous talk. In order to evaluate the faces of participants in interaction, we study, using data collections created for this purpose, the participant’s behavior, emotionnal and expressive responses to different types of humor (humorous act targeting the robot, the participant or a neutral subject). Participant's reaction to humor are made upon a multi-level processing of emotionnal, linguistic and behavioral cues.Machine learning is used to extract rules defining appreciation or not and update the participant's face evaluation in regards of the humorous act produced by the robot. An implementation of these rules in an automatic dialog system allows us to evaluate their accuracy. Numerous experiments were carried out on various populations : elderly persons, adults, teenagers. Finally, the use of the participant’s preferences in the conversation raises ethical questions, in particular against the persuasive and manipulative power of humor.