Publications

Cross-sentence relation extraction deals with the extraction of relations beyond the sentence boundary. This thesis focuses on two of the NLP tasks which are of importance to the successful extraction of cross-sentence relation mentions: event extraction and coreference resolution. The first part of the thesis focuses on addressing data sparsity issues in event extraction. We propose a self-training approach for obtaining additional labeled examples for the task. The process starts off with a Bi-LSTM event tagger trained on a small labeled data set which is used to discover new event instances in a large collection of unstructured text. The high confidence model predictions are selected to construct a data set of automatically-labeled training examples. We present several ways in which the resulting data set can be used for re-training the event tagger in conjunction with the initial labeled data. The best configuration achieves statistically significant improvement over the baseline on the ACE 2005 test set (macro-F1), as well as in a 10-fold cross validation (micro- and macro-F1) evaluation. Our error analysis reveals that the augmentation approach is especially beneficial for the classification of the most under-represented event types in the original data set. The second part of the thesis focuses on the problem of coreference resolution. While a certain level of precision can be reached by modeling surface information about entity mentions, their successful resolution often depends on semantic or world knowledge. This thesis investigates an unsupervised source of such knowledge, namely distributed word representations. We present several ways in which word embeddings can be utilized to extract features for a supervised coreference resolver. Our evaluation results and error analysis show that each of these features helps improve over the baseline coreference system’s performance, with a statistically significant improvement (CoNLL F1) achieved when the proposed features are used jointly. Moreover, all features lead to a reduction in the amount of precision errors in resolving references between common nouns, demonstrating that they successfully incorporate semantic information into the process.

• https://publikationen.sulb.uni-saarland.de/handle/20.500.11880/32255
• https://doi.org/https://dx.doi.org/10.22028/D291-35277

@phdthesis{Simova_Diss_2021,
title = {Towards the extraction of cross-sentence relations through event extraction and entity coreference},
author = {Iliana Simova},
url = {https://publikationen.sulb.uni-saarland.de/handle/20.500.11880/32255},
doi = {https://doi.org/https://dx.doi.org/10.22028/D291-35277},
year = {2021},
date = {2021},
school = {Saarland University},
address = {Saarbruecken, Germany},
abstract = {Cross-sentence relation extraction deals with the extraction of relations beyond the sentence boundary. This thesis focuses on two of the NLP tasks which are of importance to the successful extraction of cross-sentence relation mentions: event extraction and coreference resolution. The first part of the thesis focuses on addressing data sparsity issues in event extraction. We propose a self-training approach for obtaining additional labeled examples for the task. The process starts off with a Bi-LSTM event tagger trained on a small labeled data set which is used to discover new event instances in a large collection of unstructured text. The high confidence model predictions are selected to construct a data set of automatically-labeled training examples. We present several ways in which the resulting data set can be used for re-training the event tagger in conjunction with the initial labeled data. The best configuration achieves statistically significant improvement over the baseline on the ACE 2005 test set (macro-F1), as well as in a 10-fold cross validation (micro- and macro-F1) evaluation. Our error analysis reveals that the augmentation approach is especially beneficial for the classification of the most under-represented event types in the original data set. The second part of the thesis focuses on the problem of coreference resolution. While a certain level of precision can be reached by modeling surface information about entity mentions, their successful resolution often depends on semantic or world knowledge. This thesis investigates an unsupervised source of such knowledge, namely distributed word representations. We present several ways in which word embeddings can be utilized to extract features for a supervised coreference resolver. Our evaluation results and error analysis show that each of these features helps improve over the baseline coreference system’s performance, with a statistically significant improvement (CoNLL F1) achieved when the proposed features are used jointly. Moreover, all features lead to a reduction in the amount of precision errors in resolving references between common nouns, demonstrating that they successfully incorporate semantic information into the process.},
pubstate = {published},
type = {phdthesis}
}

Copy BibTeX to Clipboard

Alzheimer’s Disease (AD) has a huge impact on an ever-aging society in highly developed industrialized countries such as the EU member states: according to the World Alzheimer’s Association the number one risk factor for AD is age. AD patients suffer from neurodegenerative processes driving cognitive decline which eventually results in the loss of patients’ ability of independent living. Episodic memory impairment is the most prominent cognitive symptom of AD in its clinical stage. In addition, also executive function and semantic memory impairments significantly affect activities of daily living and are discussed as important cognitive symptoms during prodromal as well as acute clinical stages of AD. Most of the research on semantic memory impairments in AD draws evidence from the Semantic Verbal Fluency (SVF) task which evidentially also places high demands on the executive function level. At the same time, the SVF is one of the most-applied routine assessments in clinical neuropsychology especially in the diagnosis of AD. Therefore, the SVF is a prime task to study semantic memory and executive function impairment side-by-side and draw conclusions about their parallel or successive impairments across the clinical trajectory of AD. To effectively investigate semantic memory and executive function processes in the SVF, novel computational measures have been proposed that tap into data-driven semantic as well as temporal metrics scoring an SVF performance on the item-level. With a better and more differentiated understanding of AD-related executive function and semantic memory impairments in the SVF, the SVF can grow from a well-established screening into a more precise diagnostic tool for early AD. As the SVF is one of the most-applied easy-to-use and low-burden neurocognitive assessments in AD, such advancements have a direct impact on clinical practice as well. For the last decades huge efforts have been put on the discovery of disease-modifying compounds responding to specific AD biomarker-related cognitive decline characteristics. However, as most pharmaceutical trials failed, the focus has shifted towards population-wide early screening with cost-effective and scalable cognitive tests representing an effective mid-term strategy. Computer-supported SVF analysis responds to this demand. This thesis pursues a two-fold objective: (1) improve our understanding of the progressive executive function and semantic memory impairments and their interplay in clinical AD as measured by the SVF and (2) harness those insights for applied early and specific AD screening. To achieve both objectives, this thesis comprises work on subjects from different clinical stages of AD (Healthy Aging, amnestic Mild Cognitive Impairment—aMCI, and AD dementia) and in different languages (German & French). All results are based on SVF speech data generated either as a one-time assessment or a repeated within-participant testing. From these SVF speech samples, qualitative markers are extracted with different amount of computational support (ranging from manual processing of speech to fully automated evaluation). The results indicate, that semantic memory is structurally affected from an early clinical—amnestic Mild Cognitive Impairment (aMCI)—stage on and is even more affected in the later acute dementia stage. The semantic memory impairment in AD is particularly worsened through the patients’ inability to compensate by engaging executive functions. Hence, over the course of the disease, hampered executive functioning and therefore the inability to compensate for corrupt semantic memory structures might be the main driver of later-stage AD patients’ notably poor cognitive performance. These insights generated on the SVF alone are only made possible through computer-supported qualitative analysis on an item-per-item level which leads the way towards potential applications in clinical decision support. The more fine-grained qualitative analysis of the SVF is clinically valuable for AD diagnosis and screening but very time-consuming if performed manually. This thesis shows though that automatic analysis pipelines can reliably and validly generate this diagnostic information from the SVF. Automatic transcription of speech plus automatic extraction of the novel qualitative SVF features result in clinical interpretation comparable to manual transcripts and improved diagnostic decision support simulated through machine learning classification experiments. This indicates that the computer-supported SVF could ultimately be used for cost-effective fully automated early clinical AD screening. This thesis advances current AD research in a two-fold manner. First it improves the understanding of the decline of executive function and semantic memory in AD as measured through computational qualitative analysis of the SVF. Secondly, this thesis embeds these theoretical advances into practical clinical decision support concepts that help screen population-wide and cost-effective for early-stage AD.

---

Die Alzheimer-Krankheit (AD) stellt eine enorme Herausforderung für die immer älter werdende Gesellschaft in hochentwickelten Industrieländern wie den EU-Mitgliedsstaaten dar. Nach Angaben der World Alzheimer’s Association ist der größte Risikofaktor für AD das Alter. Alzheimer-Patienten leiden unter neurodegenerativen Prozessen, die kognitiven Abbau verursachen und schließlich dazu führen, dass Patienten nicht länger selbstbestimmt leben können. Die Beeinträchtigung des episodischen Gedächtnisses ist das prominenteste kognitive Symptom von AD im klinischen Stadium. Darüber hinaus führen auch Störungen der Exekutivfunktionen sowie der semantischen Gedächtnisleistung zu erheblichen Einschränkungen bei Aktivitäten des täglichen Lebens und werden als wichtige kognitive Symptome sowohl im Prodromal- als auch im akuten klinischen Stadium von AD diskutiert. Der Großteil der Forschung zu semantischen Gedächtnisbeeinträchtigungen bei AD stützt sich auf Ergebnisse aus dem Semantic Verbal Fluency Tests (SVF), der auch die Exekutivfunktionen stark fordert. In der Praxis ist die SVF eines der am häufigsten eingesetzten Routine- Assessments in der klinischen Neuropsychologie, insbesondere bei der Diagnose von AD. Daher ist die SVF eine erstklassige Aufgabe, um die Beeinträchtigung des semantischen Gedächtnisses und der exekutiven Funktionen Seite an Seite zu untersuchen und Rückschlüsse auf ihre parallelen oder sukzessiven Beeinträchtigungen im klinischen Verlauf von AD zu ziehen. Um semantische Gedächtnis- und Exekutivfunktionsprozesse in der SVF effektiv zu untersuchen, wurden jüngst neuartige computergestützte Verfahren vorgeschlagen, die sowohl datengetriebene semantische als auch temporäre Maße nutzen, die eine SVF-Leistung auf Item-Ebene bewerten. Mit einem besseren und differenzierteren Verständnis von ADbedingten Beeinträchtigungen der Exekutivfunktionen und des semantischen Gedächtnisses in der SVF kann sich die SVF von einem gut etablierten Screening zu einem präziseren Diagnoseinstrument für frühe AD entwickeln. Da die SVF eines der am häufigsten angewandten, einfach zu handhabenden und wenig belastenden neurokognitiven Assessments bei AD ist, haben solche Fortschritte auch einen direkten Einfluss auf die klinische Praxis. In den letzten Jahrzehnten wurden enorme Anstrengungen unternommen, um krankheitsmodifizierende Substanzen zu finden, die auf spezifische, mit AD-Biomarkern verbundene Merkmale des kognitiven Abbaus reagieren. Da jedoch die meisten pharmazeutischen Studien in jüngster Vergangenheit fehlgeschlagen sind, wird heute als mittelfristige Strategie bevölkerungsweite Früherkennung mit kostengünstigen und skalierbaren kognitiven Tests gefordert. Die computergestützte SVF-Analyse ist eine Antwort auf diese Forderung. Diese Arbeit verfolgt deshalb zwei Ziele: (1) Verbesserung des Verständnisses der fortschreitenden Beeinträchtigungen der Exekutivfunktionen und des semantischen Gedächtnisses und ihres Zusammenspiels bei klinischer AD, gemessen durch die SVF, und (2) Nutzung dieser Erkenntnisse für angewandte AD-Früherkennung. Um beide Ziele zu erreichen, umfasst diese Thesis Forschung mit Probanden aus verschiedenen klinischen AD Stadien (gesundes Altern, amnestisches Mild Cognitive Impairment-aMCI, und AD-Demenz) und in verschiedenen Sprachen (Deutsch & Französisch). Alle Ergebnisse basieren auf SVF Sprachdaten, erhoben im Querschnittdesign oder als wiederholte Testung in einem Längsschnittdesign. Aus diesen SVF-Sprachproben werden mit unterschiedlicher rechnerischer Unterstützung qualitative Marker extrahiert (von manueller Verarbeitung der Sprache bis hin zu vollautomatischer Auswertung). Die Ergebnisse zeigen, dass das semantische Gedächtnis bereits im frühen aMCI Stadium strukturell beeinträchtigt ist und im späteren akuten Demenzstadium noch stärker betroffen ist. Die strukturelle Beeinträchtigung des semantischen Gedächtnisses bei Alzheimer wird insbesondere dadurch verschlimmert, dass die Patienten nicht in der Lage sind, dies durch den Einsatz exekutiver Funktionen zu kompensieren. Daher könnten im Verlauf der Erkrankung eingeschränkte Exekutivfunktionen und damit die Unfähigkeit, degenerierte semantische Gedächtnisstrukturen zu kompensieren, die Hauptursache für die auffallend schlechten kognitiven Leistungen von AD-Patienten im Akutstadium sein. Diese Erkenntnisse basierend auf der SVF alleine werden erst durch die computergestützte qualitative Analyse auf Item-per-Item-Ebene möglich und weisen den Weg zu möglichen Anwendungen in der klinischen Entscheidungsunterstützung. Die feinkörnigere qualitative Analyse der SVF ist klinisch wertvoll für die AD-Diagnose und das Screening, aber sehr zeitaufwändig, wenn sie manuell durchgeführt wird. Diese Arbeit zeigt jedoch, dass automatische Analysepipelines diese diagnostischen Informationen zuverlässig und valide aus der SVF generieren können. Die automatische Transkription von Sprache plus die automatische Extraktion der neuartigen qualitativen SVF-Merkmale führen zu einer klinischen Interpretation, die mit manuellen Analysen vergleichbar ist. Diese Verarbeitung führt auch zu einer verbesserten diagnostischen Entscheidungsunterstützung, die durch Klassifikationsexperimente mit maschinellem Lernen simuliert wurde. Dies deutet darauf hin, dass die computergestützte SVF letztendlich für ein kostengünstiges vollautomatisches klinisches AD-Frühscreening eingesetzt werden könnte. Diese Arbeit bringt die aktuelle AD-Forschung auf zweifache Weise voran. Erstens verbessert sie unser Verständnis der kognitiven Einschränkungen im Bereich der Exekutivfunktionen und des semantischen Gedächtnisses bei AD, gemessen durch die computergestützte qualitative Analyse der SVF. Zweitens bettet diese Arbeit diese theoretischen Fortschritte in ein praktisches Konzept zur klinischen Entscheidungsunterstützung ein, das zukünftig ein bevölkerungsweites und kosteneffektives Screening für AD im Frühstadium ermöglichen könnte.

• https://publikationen.sulb.uni-saarland.de/handle/20.500.11880/31994
• https://doi.org/10.22028/D291-35033

@phdthesis{Tröger_Diss_2021,
title = {Executive function & semantic memory impairments in Alzheimer’s disease — investigating the decline of executive function and semantic memory in Alzheimer’s disease through computer-supported qualitative analysis of semantic verbal fluency and its applications in clinical decision support},
author = {Johannes Tr{\"o}ger},
url = {https://publikationen.sulb.uni-saarland.de/handle/20.500.11880/31994},
doi = {https://doi.org/10.22028/D291-35033},
year = {2021},
date = {2021-12-07},
school = {Saarland University},
address = {Saarbruecken, Germany},
abstract = {Alzheimer’s Disease (AD) has a huge impact on an ever-aging society in highly developed industrialized countries such as the EU member states: according to the World Alzheimer’s Association the number one risk factor for AD is age. AD patients suffer from neurodegenerative processes driving cognitive decline which eventually results in the loss of patients’ ability of independent living. Episodic memory impairment is the most prominent cognitive symptom of AD in its clinical stage. In addition, also executive function and semantic memory impairments significantly affect activities of daily living and are discussed as important cognitive symptoms during prodromal as well as acute clinical stages of AD. Most of the research on semantic memory impairments in AD draws evidence from the Semantic Verbal Fluency (SVF) task which evidentially also places high demands on the executive function level. At the same time, the SVF is one of the most-applied routine assessments in clinical neuropsychology especially in the diagnosis of AD. Therefore, the SVF is a prime task to study semantic memory and executive function impairment side-by-side and draw conclusions about their parallel or successive impairments across the clinical trajectory of AD. To effectively investigate semantic memory and executive function processes in the SVF, novel computational measures have been proposed that tap into data-driven semantic as well as temporal metrics scoring an SVF performance on the item-level. With a better and more differentiated understanding of AD-related executive function and semantic memory impairments in the SVF, the SVF can grow from a well-established screening into a more precise diagnostic tool for early AD. As the SVF is one of the most-applied easy-to-use and low-burden neurocognitive assessments in AD, such advancements have a direct impact on clinical practice as well. For the last decades huge efforts have been put on the discovery of disease-modifying compounds responding to specific AD biomarker-related cognitive decline characteristics. However, as most pharmaceutical trials failed, the focus has shifted towards population-wide early screening with cost-effective and scalable cognitive tests representing an effective mid-term strategy. Computer-supported SVF analysis responds to this demand. This thesis pursues a two-fold objective: (1) improve our understanding of the progressive executive function and semantic memory impairments and their interplay in clinical AD as measured by the SVF and (2) harness those insights for applied early and specific AD screening. To achieve both objectives, this thesis comprises work on subjects from different clinical stages of AD (Healthy Aging, amnestic Mild Cognitive Impairment—aMCI, and AD dementia) and in different languages (German & French). All results are based on SVF speech data generated either as a one-time assessment or a repeated within-participant testing. From these SVF speech samples, qualitative markers are extracted with different amount of computational support (ranging from manual processing of speech to fully automated evaluation). The results indicate, that semantic memory is structurally affected from an early clinical—amnestic Mild Cognitive Impairment (aMCI)—stage on and is even more affected in the later acute dementia stage. The semantic memory impairment in AD is particularly worsened through the patients’ inability to compensate by engaging executive functions. Hence, over the course of the disease, hampered executive functioning and therefore the inability to compensate for corrupt semantic memory structures might be the main driver of later-stage AD patients’ notably poor cognitive performance. These insights generated on the SVF alone are only made possible through computer-supported qualitative analysis on an item-per-item level which leads the way towards potential applications in clinical decision support. The more fine-grained qualitative analysis of the SVF is clinically valuable for AD diagnosis and screening but very time-consuming if performed manually. This thesis shows though that automatic analysis pipelines can reliably and validly generate this diagnostic information from the SVF. Automatic transcription of speech plus automatic extraction of the novel qualitative SVF features result in clinical interpretation comparable to manual transcripts and improved diagnostic decision support simulated through machine learning classification experiments. This indicates that the computer-supported SVF could ultimately be used for cost-effective fully automated early clinical AD screening. This thesis advances current AD research in a two-fold manner. First it improves the understanding of the decline of executive function and semantic memory in AD as measured through computational qualitative analysis of the SVF. Secondly, this thesis embeds these theoretical advances into practical clinical decision support concepts that help screen population-wide and cost-effective for early-stage AD.

---

Die Alzheimer-Krankheit (AD) stellt eine enorme Herausforderung f{\"u}r die immer {\"a}lter werdende Gesellschaft in hochentwickelten Industriel{\"a}ndern wie den EU-Mitgliedsstaaten dar. Nach Angaben der World Alzheimer's Association ist der gr{\"o}{\ss}te Risikofaktor f{\"u}r AD das Alter. Alzheimer-Patienten leiden unter neurodegenerativen Prozessen, die kognitiven Abbau verursachen und schlie{\ss}lich dazu f{\"u}hren, dass Patienten nicht l{\"a}nger selbstbestimmt leben k{\"o}nnen. Die Beeintr{\"a}chtigung des episodischen Ged{\"a}chtnisses ist das prominenteste kognitive Symptom von AD im klinischen Stadium. Dar{\"u}ber hinaus f{\"u}hren auch St{\"o}rungen der Exekutivfunktionen sowie der semantischen Ged{\"a}chtnisleistung zu erheblichen Einschr{\"a}nkungen bei Aktivit{\"a}ten des t{\"a}glichen Lebens und werden als wichtige kognitive Symptome sowohl im Prodromal- als auch im akuten klinischen Stadium von AD diskutiert. Der Gro{\ss}teil der Forschung zu semantischen Ged{\"a}chtnisbeeintr{\"a}chtigungen bei AD st{\"u}tzt sich auf Ergebnisse aus dem Semantic Verbal Fluency Tests (SVF), der auch die Exekutivfunktionen stark fordert. In der Praxis ist die SVF eines der am h{\"a}ufigsten eingesetzten Routine- Assessments in der klinischen Neuropsychologie, insbesondere bei der Diagnose von AD. Daher ist die SVF eine erstklassige Aufgabe, um die Beeintr{\"a}chtigung des semantischen Ged{\"a}chtnisses und der exekutiven Funktionen Seite an Seite zu untersuchen und R{\"u}ckschl{\"u}sse auf ihre parallelen oder sukzessiven Beeintr{\"a}chtigungen im klinischen Verlauf von AD zu ziehen. Um semantische Ged{\"a}chtnis- und Exekutivfunktionsprozesse in der SVF effektiv zu untersuchen, wurden j{\"u}ngst neuartige computergest{\"u}tzte Verfahren vorgeschlagen, die sowohl datengetriebene semantische als auch tempor{\"a}re Ma{\ss}e nutzen, die eine SVF-Leistung auf Item-Ebene bewerten. Mit einem besseren und differenzierteren Verst{\"a}ndnis von ADbedingten Beeintr{\"a}chtigungen der Exekutivfunktionen und des semantischen Ged{\"a}chtnisses in der SVF kann sich die SVF von einem gut etablierten Screening zu einem pr{\"a}ziseren Diagnoseinstrument f{\"u}r fr{\"u}he AD entwickeln. Da die SVF eines der am h{\"a}ufigsten angewandten, einfach zu handhabenden und wenig belastenden neurokognitiven Assessments bei AD ist, haben solche Fortschritte auch einen direkten Einfluss auf die klinische Praxis. In den letzten Jahrzehnten wurden enorme Anstrengungen unternommen, um krankheitsmodifizierende Substanzen zu finden, die auf spezifische, mit AD-Biomarkern verbundene Merkmale des kognitiven Abbaus reagieren. Da jedoch die meisten pharmazeutischen Studien in j{\"u}ngster Vergangenheit fehlgeschlagen sind, wird heute als mittelfristige Strategie bev{\"o}lkerungsweite Fr{\"u}herkennung mit kosteng{\"u}nstigen und skalierbaren kognitiven Tests gefordert. Die computergest{\"u}tzte SVF-Analyse ist eine Antwort auf diese Forderung. Diese Arbeit verfolgt deshalb zwei Ziele: (1) Verbesserung des Verst{\"a}ndnisses der fortschreitenden Beeintr{\"a}chtigungen der Exekutivfunktionen und des semantischen Ged{\"a}chtnisses und ihres Zusammenspiels bei klinischer AD, gemessen durch die SVF, und (2) Nutzung dieser Erkenntnisse f{\"u}r angewandte AD-Fr{\"u}herkennung. Um beide Ziele zu erreichen, umfasst diese Thesis Forschung mit Probanden aus verschiedenen klinischen AD Stadien (gesundes Altern, amnestisches Mild Cognitive Impairment-aMCI, und AD-Demenz) und in verschiedenen Sprachen (Deutsch & Franz{\"o}sisch). Alle Ergebnisse basieren auf SVF Sprachdaten, erhoben im Querschnittdesign oder als wiederholte Testung in einem L{\"a}ngsschnittdesign. Aus diesen SVF-Sprachproben werden mit unterschiedlicher rechnerischer Unterst{\"u}tzung qualitative Marker extrahiert (von manueller Verarbeitung der Sprache bis hin zu vollautomatischer Auswertung). Die Ergebnisse zeigen, dass das semantische Ged{\"a}chtnis bereits im fr{\"u}hen aMCI Stadium strukturell beeintr{\"a}chtigt ist und im sp{\"a}teren akuten Demenzstadium noch st{\"a}rker betroffen ist. Die strukturelle Beeintr{\"a}chtigung des semantischen Ged{\"a}chtnisses bei Alzheimer wird insbesondere dadurch verschlimmert, dass die Patienten nicht in der Lage sind, dies durch den Einsatz exekutiver Funktionen zu kompensieren. Daher k{\"o}nnten im Verlauf der Erkrankung eingeschr{\"a}nkte Exekutivfunktionen und damit die Unf{\"a}higkeit, degenerierte semantische Ged{\"a}chtnisstrukturen zu kompensieren, die Hauptursache f{\"u}r die auffallend schlechten kognitiven Leistungen von AD-Patienten im Akutstadium sein. Diese Erkenntnisse basierend auf der SVF alleine werden erst durch die computergest{\"u}tzte qualitative Analyse auf Item-per-Item-Ebene m{\"o}glich und weisen den Weg zu m{\"o}glichen Anwendungen in der klinischen Entscheidungsunterst{\"u}tzung. Die feink{\"o}rnigere qualitative Analyse der SVF ist klinisch wertvoll f{\"u}r die AD-Diagnose und das Screening, aber sehr zeitaufw{\"a}ndig, wenn sie manuell durchgef{\"u}hrt wird. Diese Arbeit zeigt jedoch, dass automatische Analysepipelines diese diagnostischen Informationen zuverl{\"a}ssig und valide aus der SVF generieren k{\"o}nnen. Die automatische Transkription von Sprache plus die automatische Extraktion der neuartigen qualitativen SVF-Merkmale f{\"u}hren zu einer klinischen Interpretation, die mit manuellen Analysen vergleichbar ist. Diese Verarbeitung f{\"u}hrt auch zu einer verbesserten diagnostischen Entscheidungsunterst{\"u}tzung, die durch Klassifikationsexperimente mit maschinellem Lernen simuliert wurde. Dies deutet darauf hin, dass die computergest{\"u}tzte SVF letztendlich f{\"u}r ein kosteng{\"u}nstiges vollautomatisches klinisches AD-Fr{\"u}hscreening eingesetzt werden k{\"o}nnte. Diese Arbeit bringt die aktuelle AD-Forschung auf zweifache Weise voran. Erstens verbessert sie unser Verst{\"a}ndnis der kognitiven Einschr{\"a}nkungen im Bereich der Exekutivfunktionen und des semantischen Ged{\"a}chtnisses bei AD, gemessen durch die computergest{\"u}tzte qualitative Analyse der SVF. Zweitens bettet diese Arbeit diese theoretischen Fortschritte in ein praktisches Konzept zur klinischen Entscheidungsunterst{\"u}tzung ein, das zuk{\"u}nftig ein bev{\"o}lkerungsweites und kosteneffektives Screening f{\"u}r AD im Fr{\"u}hstadium erm{\"o}glichen k{\"o}nnte.},
pubstate = {published},
type = {phdthesis}
}

Copy BibTeX to Clipboard

The current work investigates how preexisting mental representations of the meaning of an utterance (top-down processing) affect the comprehension of external perceptual properties of the linguistic input (bottom-up processing). When it comes to top-down bottom-up processing in the brain previous findings report a division of focus between left and right hemispheric mechanisms. The PARLO sentence comprehension model posits that the LH employs top-down mechanisms which allow for efficient anticipatory processing, while the RH relies more on bottom-up mechanisms. A shortcoming of the PARLO model is that it’s based on experiments manipulating solely top-down contextual constraint, leading to conclusions that hemispheric asymmetries are a function of differences in the efficiency of top-down rather than bottom-up mechanisms. Up until now, there has been no investigation of asymmetries in bottom-up processing, nor an investigation of the potential interactions between that and top-down processing for each hemisphere. This thesis consists of four event-related potential (ERP) experiments divided into two parts. Experiments 1 (central presentation) and 2 (hemispheric presentation) manipulate the bottom-up lexical frequency of critical words in high and low predictability contexts. Experiments 3 (central presentation) and 4 (hemispheric presentation) manipulate bottom-up word status, presenting critical words and pseudowords in the same high and low predictability contexts. The results allow us to extend previous findings and present the Spotlight Theory of Hemispheric Comprehension. We argue that the LH employs a kind of spotlight focus, which affords very efficient top-down processing of the expected input, since only highly predictable inputs receive additional facilitation based their bottom-up features. Alternatively, the RH lack of spotlight mechanism and focus on bottom-up lexical properties allows for the reliable processing of less predictable and irregular inputs. In combination, these complementary processing strategies provide the comprehension system with the efficiency and robustness required in a wide range of communicative situations.

• https://publikationen.sulb.uni-saarland.de/handle/20.500.11880/31806
• https://doi.org/https://dx.doi.org/10.22028/D291-33976

@phdthesis{Vergilova_Diss_2021,
title = {The Lateralization of Expectations: Hemispheric Differences in Top-down and Bottom-up Word Processing in Context},
author = {Yoana Vergilova},
url = {https://publikationen.sulb.uni-saarland.de/handle/20.500.11880/31806},
doi = {https://doi.org/https://dx.doi.org/10.22028/D291-33976},
year = {2021},
date = {2021},
school = {Saarland University},
address = {Saarbruecken, Germany},
abstract = {The current work investigates how preexisting mental representations of the meaning of an utterance (top-down processing) affect the comprehension of external perceptual properties of the linguistic input (bottom-up processing). When it comes to top-down bottom-up processing in the brain previous findings report a division of focus between left and right hemispheric mechanisms. The PARLO sentence comprehension model posits that the LH employs top-down mechanisms which allow for efficient anticipatory processing, while the RH relies more on bottom-up mechanisms. A shortcoming of the PARLO model is that it’s based on experiments manipulating solely top-down contextual constraint, leading to conclusions that hemispheric asymmetries are a function of differences in the efficiency of top-down rather than bottom-up mechanisms. Up until now, there has been no investigation of asymmetries in bottom-up processing, nor an investigation of the potential interactions between that and top-down processing for each hemisphere. This thesis consists of four event-related potential (ERP) experiments divided into two parts. Experiments 1 (central presentation) and 2 (hemispheric presentation) manipulate the bottom-up lexical frequency of critical words in high and low predictability contexts. Experiments 3 (central presentation) and 4 (hemispheric presentation) manipulate bottom-up word status, presenting critical words and pseudowords in the same high and low predictability contexts. The results allow us to extend previous findings and present the Spotlight Theory of Hemispheric Comprehension. We argue that the LH employs a kind of spotlight focus, which affords very efficient top-down processing of the expected input, since only highly predictable inputs receive additional facilitation based their bottom-up features. Alternatively, the RH lack of spotlight mechanism and focus on bottom-up lexical properties allows for the reliable processing of less predictable and irregular inputs. In combination, these complementary processing strategies provide the comprehension system with the efficiency and robustness required in a wide range of communicative situations.},
pubstate = {published},
type = {phdthesis}
}

Copy BibTeX to Clipboard

Tracing the influence of individuals or groups in social networks is an increasingly popular task in sociolinguistic studies. While methods to determine someone’s influence in shortterm contexts (e.g., social media, on-line political debates) are widespread, influence in longterm contexts is less investigated and may be harder to capture. We study the diffusion of scientific terms in an English diachronic scientific corpus, applying Hawkes Processes to capture the role of individual scientists as „influencers“ or „influencees“ in the diffusion of new concepts. Our findings on two major scientific discoveries in chemistry and astronomy of the 18th century reveal that modelling both the introduction and diffusion of scientific terms in a historical corpus as Hawkes Processes allows detecting patterns of influence between authors on a long-term scale.

• https://aclanthology.org/2021.latechclfl-1.14
• https://doi.org/10.18653/v1/2021.latechclfl-1.14

@inproceedings{bizzoni-etal-2021-diffusion,
title = {The diffusion of scientific terms - tracing individuals' influence in the history of science for English},
author = {Yuri Bizzoni and Stefania Degaetano-Ortlieb and Katrin Menzel and Elke Teich},
url = {https://aclanthology.org/2021.latechclfl-1.14},
doi = {https://doi.org/10.18653/v1/2021.latechclfl-1.14},
year = {2021},
date = {2021-11-30},
booktitle = {Proceedings of the 5th Joint SIGHUM Workshop on Computational Linguistics for Cultural Heritage, Social Sciences, Humanities and Literature},
pages = {120-127},
publisher = {Association for Computational Linguistics},
address = {Punta Cana, Dominican Republic (online)},
abstract = {Tracing the influence of individuals or groups in social networks is an increasingly popular task in sociolinguistic studies. While methods to determine someone's influence in shortterm contexts (e.g., social media, on-line political debates) are widespread, influence in longterm contexts is less investigated and may be harder to capture. We study the diffusion of scientific terms in an English diachronic scientific corpus, applying Hawkes Processes to capture the role of individual scientists as "influencers" or "influencees" in the diffusion of new concepts. Our findings on two major scientific discoveries in chemistry and astronomy of the 18th century reveal that modelling both the introduction and diffusion of scientific terms in a historical corpus as Hawkes Processes allows detecting patterns of influence between authors on a long-term scale.},
pubstate = {published},
type = {inproceedings}
}

Copy BibTeX to Clipboard

This paper aims to find a correlation between Information Density (ID) and extraposition of Relative Clauses (RC) in Early New High German. Since surprisal is connected to perceiving difficulties, the impact on the working memory is lower for frequent combinations with low surprisal-values than it is for rare combinations with higher surprisal-values. To improve text comprehension, producers therefore distribute information as evenly as possible across a discourse. Extraposed RC are expected to have a higher surprisal-value than embedded RC. We intend to find evidence for this idea in RC taken from scientific texts from the 17th to 19th century. We built a corpus of tokenized, lemmatized and normalized papers about medicine from the 17th and 19th century, manually determined the RC-variants and calculated a skipgram-Language Model to compute the 2-Skip-bigram surprisal of every word of the relevant sentences. A logistic regression over the summed up surprisal values shows a significant result, which indicates a correlation between surprisal values and extraposition. So, for these periods it can be said that RC are more likely to be extraposed when they have a high total surprisal value. The influence of surprisal values also seems to be stable across time. The comparison of the analyzed language periods shows no significant change.

• https://doi.org/10.3389/fpsyg.2021.650969
• https://doi.org/10.3389/fpsyg.2021.650969

@article{Voigtmann.Speyer,
title = {Information density and the extraposition of German relative clauses},
author = {Sophia Voigtmann and Augustin Speyer},
url = {https://doi.org/10.3389/fpsyg.2021.650969},
doi = {https://doi.org/10.3389/fpsyg.2021.650969},
year = {2021},
date = {2021-11-26},
journal = {Frontiers in Psychology},
pages = {1-18},
abstract = {This paper aims to find a correlation between Information Density (ID) and extraposition of Relative Clauses (RC) in Early New High German. Since surprisal is connected to perceiving difficulties, the impact on the working memory is lower for frequent combinations with low surprisal-values than it is for rare combinations with higher surprisal-values. To improve text comprehension, producers therefore distribute information as evenly as possible across a discourse. Extraposed RC are expected to have a higher surprisal-value than embedded RC. We intend to find evidence for this idea in RC taken from scientific texts from the 17th to 19th century. We built a corpus of tokenized, lemmatized and normalized papers about medicine from the 17th and 19th century, manually determined the RC-variants and calculated a skipgram-Language Model to compute the 2-Skip-bigram surprisal of every word of the relevant sentences. A logistic regression over the summed up surprisal values shows a significant result, which indicates a correlation between surprisal values and extraposition. So, for these periods it can be said that RC are more likely to be extraposed when they have a high total surprisal value. The influence of surprisal values also seems to be stable across time. The comparison of the analyzed language periods shows no significant change.},
pubstate = {published},
type = {article}
}

Copy BibTeX to Clipboard

@inproceedings{Menzel-etal2021,
title = {Structural complexity in scientific journal articles across time - from negative clausal expressions towards adjectival negative prefixes},
author = {Katrin Menzel and Marie-Pauline Krielke and Stefania Degaetano-Ortlieb},
year = {2021},
date = {2021-11-19},
booktitle = {Workshop on Complexity and Register (CAR21)},
address = {Berlin, Germany, CRC1412 Register},
pubstate = {published},
type = {inproceedings}
}

Copy BibTeX to Clipboard

• https://www.cambridgescholars.com/product/978-1-5275-7432-8

@inbook{Menzel2021_eponyms,
title = {Scientific Eponyms throughout the History of English Scholarly Journal Articles},
author = {Katrin Menzel},
editor = {Hans Van de Velde and Fredric T. Dolezal},
url = {https://www.cambridgescholars.com/product/978-1-5275-7432-8},
year = {2021},
date = {2021-11-08},
booktitle = {Broadening Perspectives in the History of Dictionaries and Word Studies},
isbn = {1-5275-7432-6},
pages = {159-193},
publisher = {Cambridge Scholars Publishing},
address = {Newcastle upon Tyne},
pubstate = {published},
type = {inbook}
}

Copy BibTeX to Clipboard

Recently, Ankener et al. (Frontiers in Psychology, 9, 2387, 2018) presented a visual world study which combined both attention and pupillary measures to demonstrate that anticipating a target results in lower effort to integrate that target (noun). However, they found no indication that the anticipatory processes themselves, i.e., the reduction of uncertainty about upcoming referents, results in processing effort (cf. Linzen and Jaeger, Cognitive Science, 40(6), 1382–1411, 2016). In contrast, Maess et al. (Frontiers in Human Neuroscience, 10, 1–11, 2016) found that more constraining verbs elicited a higher N400 amplitude than unconstraining verbs. The aim of the present study was therefore twofold: Firstly, we examined whether the graded ICA effect, which was previously found on the noun as a result of a likelihood manipulation, replicates in ERP measures. Secondly, we set out to investigate whether the processes leading to the generation of expectations (derived during verb and scene processing) induce an N400 modulation. Our results confirm that visual context is combined with the verb’s meaning to establish expectations about upcoming nouns and that these expectations affect the retrieval of the upcoming noun (modulated N400 on the noun). Importantly, however, we find no evidence for different costs in generating more or less specific expectations for upcoming nouns. Thus, the benefits of generating expectations are not associated with any costs in situated language comprehension.

• https://link.springer.com/article/10.3758/s13423-020-01827-3
• https://doi.org/10.3758/s13423-020-01827-3

@article{staudte2021,
title = {Graded expectations in visually situated comprehension: Costs and benefits as indexed by the N400},
author = {Maria Staudte and Christine Ankener and Heiner Drenhaus and Matthew W. Crocker},
url = {https://link.springer.com/article/10.3758/s13423-020-01827-3},
doi = {https://doi.org/10.3758/s13423-020-01827-3},
year = {2021},
date = {2021},
journal = {Psychonomic Bulletin & Review},
pages = {624-631},
publisher = {Springer},
volume = {28},
number = {2},
abstract = {Recently, Ankener et al. (Frontiers in Psychology, 9, 2387, 2018) presented a visual world study which combined both attention and pupillary measures to demonstrate that anticipating a target results in lower effort to integrate that target (noun). However, they found no indication that the anticipatory processes themselves, i.e., the reduction of uncertainty about upcoming referents, results in processing effort (cf. Linzen and Jaeger, Cognitive Science, 40(6), 1382–1411, 2016). In contrast, Maess et al. (Frontiers in Human Neuroscience, 10, 1–11, 2016) found that more constraining verbs elicited a higher N400 amplitude than unconstraining verbs. The aim of the present study was therefore twofold: Firstly, we examined whether the graded ICA effect, which was previously found on the noun as a result of a likelihood manipulation, replicates in ERP measures. Secondly, we set out to investigate whether the processes leading to the generation of expectations (derived during verb and scene processing) induce an N400 modulation. Our results confirm that visual context is combined with the verb’s meaning to establish expectations about upcoming nouns and that these expectations affect the retrieval of the upcoming noun (modulated N400 on the noun). Importantly, however, we find no evidence for different costs in generating more or less specific expectations for upcoming nouns. Thus, the benefits of generating expectations are not associated with any costs in situated language comprehension.},
pubstate = {published},
type = {article}
}

Copy BibTeX to Clipboard

@miscellaneous{HoeltjeMecklinger2021,
title = {The Secret Life of (Dis-)Confirmed Predictions: Effects of Sentence Constraint and Word Expectedness on Episodic Memory Formation, and how they are Reflected in Event-Related Potentials},
author = {Gerrit H{\"o}ltje and Axel Mecklinger},
year = {2021},
date = {2021},
booktitle = {CNS 2020 Virtual meeting, Abstract Book},
pubstate = {published},
type = {miscellaneous}
}

Copy BibTeX to Clipboard

Endeavors to computationally model language variation and change are ever increasing. While analyses of recent diachronic trends are frequently conducted, long-term trends accounting for sociolinguistic variation are less well-studied. Our work sheds light on the temporal dynamics of language use of British 18th century women as a group in transition across two situational contexts. Our findings reveal that in formal contexts women adapt to register conventions, while in informal contexts they act as innovators of change in language use influencing others. While adopted from other disciplines, our methods inform (historical) sociolinguistic work in novel ways. These methods include diachronic periodization by Kullback-Leibler divergence to determine periods of change and relevant features of variation, and event cascades as influencer models.

• frai-04-609970     (3.66MB)
• https://www.frontiersin.org/article/10.3389/frai.2021.609970
• https://doi.org/10.3389/frai.2021.609970

@article{Degaetano-Ortlieb2021,
title = {Registerial Adaptation vs. Innovation Across Situational Contexts: 18th Century Women in Transition},
author = {Stefania Degaetano-Ortlieb and Tanja S{\"a}ily and Yuri Bizzoni},
url = {https://www.frontiersin.org/article/10.3389/frai.2021.609970},
doi = {https://doi.org/10.3389/frai.2021.609970},
year = {2021},
date = {2021},
journal = {Frontiers in Artificial Intelligence, section Language and Computation},
volume = {4},
abstract = {Endeavors to computationally model language variation and change are ever increasing. While analyses of recent diachronic trends are frequently conducted, long-term trends accounting for sociolinguistic variation are less well-studied. Our work sheds light on the temporal dynamics of language use of British 18th century women as a group in transition across two situational contexts. Our findings reveal that in formal contexts women adapt to register conventions, while in informal contexts they act as innovators of change in language use influencing others. While adopted from other disciplines, our methods inform (historical) sociolinguistic work in novel ways. These methods include diachronic periodization by Kullback-Leibler divergence to determine periods of change and relevant features of variation, and event cascades as influencer models.},
pubstate = {published},
type = {article}
}

Copy BibTeX to Clipboard

The present paper deals with a computational analysis of translationese in professional and student English-to-German translations belonging to different registers. Building upon an information-theoretical approach, we test translation conformity to source and target language in terms of a neural language model’s perplexity over Part of Speech (PoS) sequences. Our primary focus is on register diversification vs. convergence, reflected in the use of constructions eliciting a higher vs. lower perplexity score. Our results show that, against our expectations, professional translations elicit higher perplexity scores from a target language model than students’ translations. An analysis of the distribution of PoS patterns across registers shows that this apparent paradox is the effect of higher stylistic diversification and register sensitivity in professional translations. Our results contribute to the understanding of human translationese and shed light on the variation in texts generated by different translators, which is valuable for translation studies, multilingual language processing, and machine translation.

• https://aclanthology.org/2021.nodalida-main.6

@inproceedings{Bizzoni2021,
title = {Measuring Translationese across Levels of Expertise: Are Professionals more Surprising than Students?},
author = {Yuri Bizzoni and Ekaterina Lapshinova-Koltunski},
url = {https://aclanthology.org/2021.nodalida-main.6},
year = {2021},
date = {2021},
booktitle = {Proceedings of the 23rd Nordic Conference on Computational Linguistics (NoDaLiDa)},
pages = {53-63},
publisher = {Link{\"o}ping University Electronic Press, Sweden},
abstract = {The present paper deals with a computational analysis of translationese in professional and student English-to-German translations belonging to different registers. Building upon an information-theoretical approach, we test translation conformity to source and target language in terms of a neural language model’s perplexity over Part of Speech (PoS) sequences. Our primary focus is on register diversification vs. convergence, reflected in the use of constructions eliciting a higher vs. lower perplexity score. Our results show that, against our expectations, professional translations elicit higher perplexity scores from a target language model than students’ translations. An analysis of the distribution of PoS patterns across registers shows that this apparent paradox is the effect of higher stylistic diversification and register sensitivity in professional translations. Our results contribute to the understanding of human translationese and shed light on the variation in texts generated by different translators, which is valuable for translation studies, multilingual language processing, and machine translation.},
pubstate = {published},
type = {inproceedings}
}

Copy BibTeX to Clipboard

Expectation-based theories of language processing, such as Surprisal theory, are supported by evidence of anticipation effects in both behavioural and neurophysiological measures. Online measures of language processing, however, are known to be influenced by factors such as lexical association that are distinct from—but often confounded with—expectancy. An open question therefore is whether a specific locus of expectancy related effects can be established in neural and behavioral processing correlates. We address this question in an event-related potential experiment and a self-paced reading experiment that independently cross expectancy and lexical association in a context manipulation design. We find that event-related potentials reveal that the N400 is sensitive to both expectancy and lexical association, while the P600 is modulated only by expectancy. Reading times, in turn, reveal effects of both association and expectancy in the first spillover region, followed by effects of expectancy alone in the second spillover region. These findings are consistent with the Retrieval-Integration account of language comprehension, according to which lexical retrieval (N400) is facilitated for words that are both expected and associated, whereas integration difficulty (P600) will be greater for unexpected words alone. Further, an exploratory analysis suggests that the P600 is not merely sensitive to expectancy violations, but rather, that there is a continuous relation. Taken together, these results suggest that the P600, like reading times, may reflect a meaning-centric notion of Surprisal in language comprehension.

• https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0257430
• https://doi.org/10.1371/journal.pone.0257430

@article{aurnhammer2021retrieval,
title = {Retrieval (N400) and Integration (P600) in Expectation-based Comprehension},
author = {Christoph Aurnhammer and Francesca Delogu and Miriam Schulz and Harm Brouwer and Matthew W. Crocker},
url = {https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0257430},
doi = {https://doi.org/10.1371/journal.pone.0257430},
year = {2021},
date = {2021-09-28},
journal = {PLoS ONE},
pages = {e0257430},
volume = {16},
number = {9},
abstract = {Expectation-based theories of language processing, such as Surprisal theory, are supported by evidence of anticipation effects in both behavioural and neurophysiological measures. Online measures of language processing, however, are known to be influenced by factors such as lexical association that are distinct from—but often confounded with—expectancy. An open question therefore is whether a specific locus of expectancy related effects can be established in neural and behavioral processing correlates. We address this question in an event-related potential experiment and a self-paced reading experiment that independently cross expectancy and lexical association in a context manipulation design. We find that event-related potentials reveal that the N400 is sensitive to both expectancy and lexical association, while the P600 is modulated only by expectancy. Reading times, in turn, reveal effects of both association and expectancy in the first spillover region, followed by effects of expectancy alone in the second spillover region. These findings are consistent with the Retrieval-Integration account of language comprehension, according to which lexical retrieval (N400) is facilitated for words that are both expected and associated, whereas integration difficulty (P600) will be greater for unexpected words alone. Further, an exploratory analysis suggests that the P600 is not merely sensitive to expectancy violations, but rather, that there is a continuous relation. Taken together, these results suggest that the P600, like reading times, may reflect a meaning-centric notion of Surprisal in language comprehension.},
pubstate = {published},
type = {article}
}

Copy BibTeX to Clipboard

DiscAlign for Penn and RST Discourse Treebanks was developed by Saarland University. It consists of alignment information for the discourse annotations contained in Penn Discourse Treebank Version 2.0 (LDC2008T05) (PDTB 2.0) and RST Discourse Treebank (LDC2002T07) (RST-DT). PDTB 2.0 and RST-DT annotations overlap for 385 newspaper articles in sections 6, 11, 13, 19 and 23 of the Wall Street Journal corpus contained in Treebank-2 (LDC95T7). DiscAlign for Penn and RST Discourse Treebanks contains approximately 6,700 alignments between PDTB 2.0 and RST-DT relations.

• https://catalog.ldc.upenn.edu/LDC2021T16
• https://doi.org/10.35111/cf0q-c454

@miscellaneous{Demberg_etal_DiscAlign,
title = {DiscAlign for Penn and RST Discourse Treebanks},
author = {Vera Demberg and Fatemeh Torabi Asr and Merel Scholman},
url = {https://catalog.ldc.upenn.edu/LDC2021T16},
doi = {https://doi.org/10.35111/cf0q-c454},
year = {2021},
date = {2021},
isbn = {1-58563-975-3},
publisher = {Linguistic Data Consortium},
address = {Philadelphia},
abstract = {DiscAlign for Penn and RST Discourse Treebanks was developed by Saarland University. It consists of alignment information for the discourse annotations contained in Penn Discourse Treebank Version 2.0 (LDC2008T05) (PDTB 2.0) and RST Discourse Treebank (LDC2002T07) (RST-DT). PDTB 2.0 and RST-DT annotations overlap for 385 newspaper articles in sections 6, 11, 13, 19 and 23 of the Wall Street Journal corpus contained in Treebank-2 (LDC95T7). DiscAlign for Penn and RST Discourse Treebanks contains approximately 6,700 alignments between PDTB 2.0 and RST-DT relations.},
pubstate = {published},
type = {miscellaneous}
}

Copy BibTeX to Clipboard

In this paper, we investigate grammatical complexity as a register feature of scientific English and German. Specifically, we carry out a diachronic comparison between general and scientific discourse in the two languages from the 17th to the 19th century, using relativizers as proxies for grammatical complexity. We ground our study in register theory (Halliday and Hasan, 1985), assuming that language use reflects contextual factors, which contribute to the formation of registers (Quirk et al., 1985; Biber et al., 1999; Teich et al., 2016). Our findings show a clear tendency towards grammatical simplification in scientific discourse in both languages with English spearheading the trend early on and German following later.

• https://doi.org/10.15845/bells.v11i1.3440
• https://doi.org/10.15845/bells.v11i1.3440

@article{Krielke2021relativizers,
title = {Relativizers as markers of grammatical complexity: A diachronic, cross-register study of English and German},
author = {Marie-Pauline Krielke},
url = {https://doi.org/10.15845/bells.v11i1.3440},
doi = {https://doi.org/10.15845/bells.v11i1.3440},
year = {2021},
date = {2021-09-15},
journal = {Bergen Language and Linguistics Studies},
pages = {91-120},
volume = {11},
number = {1},
abstract = {In this paper, we investigate grammatical complexity as a register feature of scientific English and German. Specifically, we carry out a diachronic comparison between general and scientific discourse in the two languages from the 17th to the 19th century, using relativizers as proxies for grammatical complexity. We ground our study in register theory (Halliday and Hasan, 1985), assuming that language use reflects contextual factors, which contribute to the formation of registers (Quirk et al., 1985; Biber et al., 1999; Teich et al., 2016). Our findings show a clear tendency towards grammatical simplification in scientific discourse in both languages with English spearheading the trend early on and German following later.},
pubstate = {published},
type = {article}
}

Copy BibTeX to Clipboard

Previous studies have shown that at moderate levels of spectral degradation, semantic predictability facilitates language comprehension. It is argued that when speech is degraded, listeners have narrowed expectations about the sentence endings; i.e., semantic prediction may be limited to only most highly predictable sentence completions. The main objectives of this study were to (i) examine whether listeners form narrowed expectations or whether they form predictions across a wide range of probable sentence endings, (ii) assess whether the facilitatory effect of semantic predictability is modulated by perceptual adaptation to degraded speech, and (iii) use and establish a sensitive metric for the measurement of language comprehension. For this, we created 360 German Subject-Verb-Object sentences that varied in semantic predictability of a sentence-final target word in a graded manner (high, medium, and low) and levels of spectral degradation (1, 4, 6, and 8 channels noise-vocoding). These sentences were presented auditorily to two groups: One group (n =48) performed a listening task in an unpredictable channel context in which the degraded speech levels were randomized, while the other group (n =50) performed the task in a predictable channel context in which the degraded speech levels were blocked. The results showed that at 4 channels noise-vocoding, response accuracy was higher in high-predictability sentences than in the medium-predictability sentences, which in turn was higher than in the low-predictability sentences. This suggests that, in contrast to the narrowed expectations view, comprehension of moderately degraded speech, ranging from low- to high- including medium-predictability sentences, is facilitated in a graded manner; listeners probabilistically preactivate upcoming words from a wide range of semantic space, not limiting only to highly probable sentence endings. Additionally, in both channel contexts, we did not observe learning effects; i.e., response accuracy did not increase over the course of experiment, and response accuracy was higher in the predictable than in the unpredictable channel context. We speculate from these observations that when there is no trial-by-trial variation of the levels of speech degradation, listeners adapt to speech quality at a long timescale; however, when there is a trial-by-trial variation of the high-level semantic feature (e.g., sentence predictability), listeners do not adapt to low-level perceptual property (e.g., speech quality) at a short timescale.

• https://www.frontiersin.org/articles/10.3389/fpsyg.2021.714485/full
• https://doi.org/10.3389/fpsyg.2021.714485

@article{bhandari2021semantic,
title = {Semantic Predictability Facilitates Comprehension of Degraded Speech in a Graded Manner},
author = {Pratik Bhandari and Vera Demberg and Jutta Kray},
url = {https://www.frontiersin.org/articles/10.3389/fpsyg.2021.714485/full},
doi = {https://doi.org/10.3389/fpsyg.2021.714485},
year = {2021},
date = {2021-09-09},
journal = {Frontiers in Psychology},
pages = {3769},
publisher = {Frontiers},
abstract = {Previous studies have shown that at moderate levels of spectral degradation, semantic predictability facilitates language comprehension. It is argued that when speech is degraded, listeners have narrowed expectations about the sentence endings; i.e., semantic prediction may be limited to only most highly predictable sentence completions. The main objectives of this study were to (i) examine whether listeners form narrowed expectations or whether they form predictions across a wide range of probable sentence endings, (ii) assess whether the facilitatory effect of semantic predictability is modulated by perceptual adaptation to degraded speech, and (iii) use and establish a sensitive metric for the measurement of language comprehension. For this, we created 360 German Subject-Verb-Object sentences that varied in semantic predictability of a sentence-final target word in a graded manner (high, medium, and low) and levels of spectral degradation (1, 4, 6, and 8 channels noise-vocoding). These sentences were presented auditorily to two groups: One group (n =48) performed a listening task in an unpredictable channel context in which the degraded speech levels were randomized, while the other group (n =50) performed the task in a predictable channel context in which the degraded speech levels were blocked. The results showed that at 4 channels noise-vocoding, response accuracy was higher in high-predictability sentences than in the medium-predictability sentences, which in turn was higher than in the low-predictability sentences. This suggests that, in contrast to the narrowed expectations view, comprehension of moderately degraded speech, ranging from low- to high- including medium-predictability sentences, is facilitated in a graded manner; listeners probabilistically preactivate upcoming words from a wide range of semantic space, not limiting only to highly probable sentence endings. Additionally, in both channel contexts, we did not observe learning effects; i.e., response accuracy did not increase over the course of experiment, and response accuracy was higher in the predictable than in the unpredictable channel context. We speculate from these observations that when there is no trial-by-trial variation of the levels of speech degradation, listeners adapt to speech quality at a long timescale; however, when there is a trial-by-trial variation of the high-level semantic feature (e.g., sentence predictability), listeners do not adapt to low-level perceptual property (e.g., speech quality) at a short timescale.},
pubstate = {published},
type = {article}
}

Copy BibTeX to Clipboard

Due to a lack of annotated data, theories of historical syntax are often based on very small, manually compiled data sets. To enable the empirical evaluation of existing hypotheses, the present study explores the automatic recognition of phrases in historical German. Using modern and historical treebanks, training data for a neural sequence labeling tool and a probabilistic parser is created, and both methods are compared on a variety of data sets. The evaluation shows that the unlexicalized parser outperforms the sequence labeling approach, achieving F1-scores of 87%–91% on modern German and between 73% and 85% on different historical corpora. An error analysis indicates that accuracy decreases especially for longer phrases, but most of the errors concern incorrect phrase boundaries, suggesting further potential for improvement.

• 2021.konvens-1.11     (1.27MB)
• https://aclanthology.org/2021.konvens-1.11

@inproceedings{ortmann-2021b,
title = {Automatic Phrase Recognition in Historical German},
author = {Katrin Ortmann},
url = {https://aclanthology.org/2021.konvens-1.11},
year = {2021},
date = {2021-09-06},
booktitle = {Proceedings of the 17th Conference on Natural Language Processing (KONVENS 2021)},
pages = {127–136},
publisher = {KONVENS 2021 Organizers},
address = {D{\"u}sseldorf, Germany},
abstract = {Due to a lack of annotated data, theories of historical syntax are often based on very small, manually compiled data sets. To enable the empirical evaluation of existing hypotheses, the present study explores the automatic recognition of phrases in historical German. Using modern and historical treebanks, training data for a neural sequence labeling tool and a probabilistic parser is created, and both methods are compared on a variety of data sets. The evaluation shows that the unlexicalized parser outperforms the sequence labeling approach, achieving F1-scores of 87%–91% on modern German and between 73% and 85% on different historical corpora. An error analysis indicates that accuracy decreases especially for longer phrases, but most of the errors concern incorrect phrase boundaries, suggesting further potential for improvement.},
pubstate = {published},
type = {inproceedings}
}

Copy BibTeX to Clipboard

We present an extended version of a tool developed for calculating linguistic distances and asymmetries in auditory perception of closely related languages. Along with evaluating the metrics available in the initial version of the tool, we introduce word adaptation entropy as an additional metric of linguistic asymmetry. Potential predictors of speech intelligibility are validated with human performance in spoken cognate recognition experiments for Bulgarian and Russian. Special attention is paid to the possibly different contributions of vowels and consonants in oral intercomprehension. Using incom.py 2.0 it is possible to calculate, visualize, and validate three measurement methods of linguistic distances and asymmetries as well as carrying out regression analyses in speech intelligibility between related languages.

• https://aclanthology.org/2021.ranlp-1.110/

@inproceedings{mosbach-etal-2021-incom,
title = {incom.py 2.0 - Calculating Linguistic Distances and Asymmetries in Auditory Perception of Closely Related Languages},
author = {Marius Mosbach and Irina Stenger and Tania Avgustinova and Bernd M{\"o}bius and Dietrich Klakow},
url = {https://aclanthology.org/2021.ranlp-1.110/},
year = {2021},
date = {2021-09-01},
booktitle = {Proceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2021)},
pages = {968-977},
publisher = {INCOMA Ltd.},
address = {Held Online},
abstract = {We present an extended version of a tool developed for calculating linguistic distances and asymmetries in auditory perception of closely related languages. Along with evaluating the metrics available in the initial version of the tool, we introduce word adaptation entropy as an additional metric of linguistic asymmetry. Potential predictors of speech intelligibility are validated with human performance in spoken cognate recognition experiments for Bulgarian and Russian. Special attention is paid to the possibly different contributions of vowels and consonants in oral intercomprehension. Using incom.py 2.0 it is possible to calculate, visualize, and validate three measurement methods of linguistic distances and asymmetries as well as carrying out regression analyses in speech intelligibility between related languages.},
pubstate = {published},
type = {inproceedings}
}

Copy BibTeX to Clipboard

Traditional hand-crafted linguistically-informed features have often been used for distinguishing between translated and original non-translated texts. By contrast, to date, neural architectures without manual feature engineering have been less explored for this task. In this work, we (i) compare the traditional feature-engineering-based approach to the feature-learning-based one and (ii) analyse the neural architectures in order to investigate how well the hand-crafted features explain the variance in the neural models’ predictions. We use pre-trained neural word embeddings, as well as several end-to-end neural architectures in both monolingual and multilingual settings and compare them to feature-engineering-based SVM classifiers. We show that (i) neural architectures outperform other approaches by more than 20 accuracy points, with the BERT-based model performing the best in both the monolingual and multilingual settings; (ii) while many individual hand-crafted translationese features correlate with neural model predictions, feature importance analysis shows that the most important features for neural and classical architectures differ; and (iii) our multilingual experiments provide empirical evidence for translationese universals across languages.

• 2021.emnlp-main.676     (0.41MB)
• https://aclanthology.org/2021.emnlp-main.676/
• https://doi.org/10.18653/v1/2021.emnlp-main.676

@inproceedings{pylypenko-etal-2021-comparing,
title = {Comparing Feature-Engineering and Feature-Learning Approaches for Multilingual Translationese Classification},
author = {Daria Pylypenko and Kwabena Amponsah-Kaakyire and Koel Dutta Chowdhury and Josef van Genabith and Cristina Espa{\~n}a-Bonet},
url = {https://aclanthology.org/2021.emnlp-main.676/},
doi = {https://doi.org/10.18653/v1/2021.emnlp-main.676},
year = {2021},
date = {2021},
booktitle = {Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing},
pages = {8596–8611},
publisher = {Association for Computational Linguistics},
address = {Online and Punta Cana, Dominican Republic},
abstract = {Traditional hand-crafted linguistically-informed features have often been used for distinguishing between translated and original non-translated texts. By contrast, to date, neural architectures without manual feature engineering have been less explored for this task. In this work, we (i) compare the traditional feature-engineering-based approach to the feature-learning-based one and (ii) analyse the neural architectures in order to investigate how well the hand-crafted features explain the variance in the neural models’ predictions. We use pre-trained neural word embeddings, as well as several end-to-end neural architectures in both monolingual and multilingual settings and compare them to feature-engineering-based SVM classifiers. We show that (i) neural architectures outperform other approaches by more than 20 accuracy points, with the BERT-based model performing the best in both the monolingual and multilingual settings; (ii) while many individual hand-crafted translationese features correlate with neural model predictions, feature importance analysis shows that the most important features for neural and classical architectures differ; and (iii) our multilingual experiments provide empirical evidence for translationese universals across languages.},
pubstate = {published},
type = {inproceedings}
}

Copy BibTeX to Clipboard

Previous research has used linguistic features to show that translations exhibit traces of source language interference and that phylogenetic trees between languages can be reconstructed from the results of translations into the same language. Recent research has shown that instances of translationese (source language interference) can even be detected in embedding spaces, comparing embeddings spaces of original language data with embedding spaces resulting from translations into the same language, using a simple Eigenvectorbased divergence from isomorphism measure. To date, it remains an open question whether alternative graph-isomorphism measures can produce better results. In this paper, we (i) explore Gromov-Hausdorff distance, (ii) present a novel spectral version of the Eigenvectorbased method, and (iii) evaluate all approaches against a broad linguistic typological database (URIEL). We show that language distances resulting from our spectral isomorphism approaches can reproduce genetic trees on a par with previous work without requiring any explicit linguistic information and that the results can be extended to non-Indo-European languages. Finally, we show that the methods are robust under a variety of modeling conditions.

• https://aclanthology.org/2021.ranlp-1.43/

@inproceedings{Chowdhury2021tracing,
title = {Tracing Source Language Interference in Translation with Graph-Isomorphism Measures},
author = {Koel Dutta Chowdhury and Cristina Espa{\~n}a-Bonet and Josef van Genabith},
url = {https://aclanthology.org/2021.ranlp-1.43/},
year = {2021},
date = {2021},
booktitle = {Proceedings of Recent Advances in Natural Language Processing (RANLP 2021)},
issn = {2603-2813},
pages = {380-390},
address = {Online},
abstract = {Previous research has used linguistic features to show that translations exhibit traces of source language interference and that phylogenetic trees between languages can be reconstructed from the results of translations into the same language. Recent research has shown that instances of translationese (source language interference) can even be detected in embedding spaces, comparing embeddings spaces of original language data with embedding spaces resulting from translations into the same language, using a simple Eigenvectorbased divergence from isomorphism measure. To date, it remains an open question whether alternative graph-isomorphism measures can produce better results. In this paper, we (i) explore Gromov-Hausdorff distance, (ii) present a novel spectral version of the Eigenvectorbased method, and (iii) evaluate all approaches against a broad linguistic typological database (URIEL). We show that language distances resulting from our spectral isomorphism approaches can reproduce genetic trees on a par with previous work without requiring any explicit linguistic information and that the results can be extended to non-Indo-European languages. Finally, we show that the methods are robust under a variety of modeling conditions.},
pubstate = {published},
type = {inproceedings}
}

Copy BibTeX to Clipboard

@miscellaneous{Menzel2021epic,
title = {EPIC-UdS - ein mehrsprachiges Korpus als Grundlage f{\"u}r die korpusbasierte Dolmetsch- und {\"U}bersetzungswissenschaft},
author = {Katrin Menzel and Heike Przybyl and Ekaterina Lapshinova-Koltunski},
year = {2021},
date = {2021},
booktitle = {TRANSLATA IV - 4. Internationale Konferenz zur Translationswissenschaft},
address = {Innsbruck},
pubstate = {published},
type = {miscellaneous}
}

Copy BibTeX to Clipboard