TY - JOUR A1 - Genzel, Susanne A1 - Kügler, Frank T1 - Production and perception of question prosody in Akan JF - Journal of the International Phonetic Association N2 - The paper presents a production experiment investigating the phonetic parameters speakers employ to differentiate Yes-No questions from string-identical statements in Akan, a West-African two-tone Kwa language. Results show that, in comparison to the statement, speakers use a higher pitch register throughout the utterance as a global parameter, and falling f0, longer duration and higher intensity as local parameters on the final syllable of the Yes-No question. Further, two perception experiments (forced-choice identification and gating) investigate the perceptual relevance of the global parameter and the local final parameters. Results show that listeners cannot assess the higher pitch register information to identify the mode of a sentence early on. Rather, identification takes place when the local phonetic parameters on the final vowel are available. The findings point to the superiority of language-specific cues in sentence mode perception. It is suggested that Akan uses a low boundary tone that associates with the right edge of the intonation phrase (L%) in Yes-No questions. The results are discussed from the point of view of question intonation typology in African languages. It is argued that a classification along the lines of functionally relevant cues is preferable to an impressionistic analysis. Y1 - 2018 U6 - https://doi.org/10.1017/S0025100318000191 SN - 0025-1003 SN - 1475-3502 VL - 50 IS - 1 SP - 61 EP - 92 PB - Cambridge Univ. Press CY - Cambridge ER - TY - JOUR A1 - Amaechi, Mary Chimaobi A1 - Georgi, Doreen T1 - On optional wh-/focus fronting in Igbo BT - a SYN-SEM-PHON interaction JF - Zeitschrift für Sprachwissenschaft N2 - This paper discusses surface optionality in focus fronting in the Benue-Congo language Igbo. A focused XP can occur in-situ or ex-situ. We argue that the optionality does not have its origins in the syntax: in fact, exactly one focused XP has to move to the designated focus position in the left periphery in the syntax. The alternation between in-situ and ex-situ rather arises at PF: either the lowest or the topmost copy of the focus chain is pronounced. The choice is determined by semantic-pragmatic factors, i. e., we see an interaction between PF and LF. This constitutes a challenge for a strict version of the Y-model of grammar. KW - (A)over-bar-movement KW - focus realization KW - PF-optionality KW - Y-model KW - copy KW - pronounciation KW - Benue-Congo languages Y1 - 2020 U6 - https://doi.org/10.1515/zfs-2020-2017 SN - 0721-9067 SN - 1613-3706 VL - 39 IS - 3 SP - 299 EP - 327 PB - De Gruyter CY - Berlin ER - TY - THES A1 - Yadav, Himanshu T1 - A computational evaluation of feature distortion and cue weighting in sentence comprehension T1 - Eine komputationale Evaluation von Feature-Verfälschung und Cue-Gewichtung in der Satzverarbeitung N2 - Successful sentence comprehension requires the comprehender to correctly figure out who did what to whom. For example, in the sentence John kicked the ball, the comprehender has to figure out who did the action of kicking and what was being kicked. This process of identifying and connecting the syntactically-related words in a sentence is called dependency completion. What are the cognitive constraints that determine dependency completion? A widely-accepted theory is cue-based retrieval. The theory maintains that dependency completion is driven by a content-addressable search for the co-dependents in memory. The cue-based retrieval explains a wide range of empirical data from several constructions including subject-verb agreement, subject-verb non-agreement, plausibility mismatch configurations, and negative polarity items. However, there are two major empirical challenges to the theory: (i) Grammatical sentences’ data from subject-verb number agreement dependencies, where the theory predicts a slowdown at the verb in sentences like the key to the cabinet was rusty compared to the key to the cabinets was rusty, but the data are inconsistent with this prediction; and, (ii) Data from antecedent-reflexive dependencies, where a facilitation in reading times is predicted at the reflexive in the bodybuilder who worked with the trainers injured themselves vs. the bodybuilder who worked with the trainer injured themselves, but the data do not show a facilitatory effect. The work presented in this dissertation is dedicated to building a more general theory of dependency completion that can account for the above two datasets without losing the original empirical coverage of the cue-based retrieval assumption. In two journal articles, I present computational modeling work that addresses the above two empirical challenges. To explain the grammatical sentences’ data from subject-verb number agreement dependencies, I propose a new model that assumes that the cue-based retrieval operates on a probabilistically distorted representation of nouns in memory (Article I). This hybrid distortion-plus-retrieval model was compared against the existing candidate models using data from 17 studies on subject-verb number agreement in 4 languages. I find that the hybrid model outperforms the existing models of number agreement processing suggesting that the cue-based retrieval theory must incorporate a feature distortion assumption. To account for the absence of facilitatory effect in antecedent-reflexive dependen� cies, I propose an individual difference model, which was built within the cue-based retrieval framework (Article II). The model assumes that individuals may differ in how strongly they weigh a syntactic cue over a number cue. The model was fitted to data from two studies on antecedent-reflexive dependencies, and the participant-level cue-weighting was estimated. We find that one-fourth of the participants, in both studies, weigh the syntactic cue higher than the number cue in processing reflexive dependencies and the remaining participants weigh the two cues equally. The result indicates that the absence of predicted facilitatory effect at the level of grouped data is driven by some, not all, participants who weigh syntactic cues higher than the number cue. More generally, the result demonstrates that the assumption of differential cue weighting is important for a theory of dependency completion processes. This differential cue weighting idea was independently supported by a modeling study on subject-verb non-agreement dependencies (Article III). Overall, the cue-based retrieval, which is a general theory of dependency completion, needs to incorporate two new assumptions: (i) the nouns stored in memory can undergo probabilistic feature distortion, and (ii) the linguistic cues used for retrieval can be weighted differentially. This is the cumulative result of the modeling work presented in this dissertation. The dissertation makes an important theoretical contribution: Sentence comprehension in humans is driven by a mechanism that assumes cue-based retrieval, probabilistic feature distortion, and differential cue weighting. This insight is theoretically important because there is some independent support for these three assumptions in sentence processing and the broader memory literature. The modeling work presented here is also methodologically important because for the first time, it demonstrates (i) how the complex models of sentence processing can be evaluated using data from multiple studies simultaneously, without oversimplifying the models, and (ii) how the inferences drawn from the individual-level behavior can be used in theory development. N2 - Bei der Satzverarbeitung muss der Leser richtig herausfinden, wer wem was angetan hat. Zum Beispiel muss der Leser in dem Satz „John hat den Ball getreten“ herausfinden, wer tat die Aktion des Tretens und was getreten wurde. Dieser Prozess des Identifizierens und Verbindens der syntaktisch verwandte Wörter in einem Satz nennt man Dependency-Completion. Was sind die kognitiven Mechanismen, die Dependency-Completion bestimmen? Eine weithin akzeptierte Theorie ist der Cue-based retrieval. Die Theorie besagt, dass die Dependency-Completion durch eine inhaltsadressierbare Suche nach der vorangetrieben wird Co-Abhängige im Gedächtnis. Der Cue-basierte Abruf erklärt ein breites Spektrum an empirischen Daten mehrere Konstruktionen, darunter Subjekt-Verb-Übereinstimmung, Subjekt-Verb-Nichtübereinstimmung, Plausibilität Mismatch-Konfigurationen und Elemente mit negativer Polarität. Es gibt jedoch zwei große empirische Herausforderungen für die Theorie: (i) Grammatische Sätze Daten aus Subjekt-Verb-Nummer-Dependency, bei denen die Theorie eine Verlangsamung vorhersagt das Verb in Sätzen wie „the key to the cabinet was rusty“ im Vergleich zu „the key to the cabinets was rusty“, aber die Daten stimmen nicht mit dieser Vorhersage überein; und (ii) Daten von Antezedenz-Reflexiv Strukturen, wo eine Leseerleichterung beim reflexiven „the bodybuilder who worked with the trainers injured themselves“ vs. „the bodybuilder who worked with the trainers injured themselves", aber die Daten zeigen keine vermittelnde Wirkung. Die in dieser Dissertation vorgestellte Arbeit widmet sich dem Aufbau einer allgemeineren Theorie von Dependency-Completion, die die beiden oben genannten Datensätze berücksichtigen kann, ohne das Original zu verlieren empirische Abdeckung der Cue-based Retrieval-Annahme. In zwei Zeitschriftenartikeln stelle ich Arbeiten zur Computermodellierung vor, die sich mit den beiden oben genannten empirischen Herausforderungen befassen. Um die Daten der grammatikalischen Sätze aus den Abhängigkeiten der Subjekt-Verb-Nummer-Übereinstimmung zu erklären, schlage ich ein neues Modell vor, das davon ausgeht, dass der Cue-basierte Abruf probabilistisch funktioniert verzerrte Darstellung von Substantiven im Gedächtnis (Artikel I). Dieses hybride Distortion-plus-Retrieval-Modell wurde anhand von Daten aus 17 Studien zu Subjekt-Verb mit den bestehenden Kandidatenmodellen verglichen Nummernvereinbarung in 4 Sprachen. Ich finde, dass das Hybridmodell die bestehenden Modelle übertrifft der Nummernvereinbarungsverarbeitung, was darauf hindeutet, dass die Cue-based Retrieval-Theorie Folgendes umfassen muss: a Annahme von Feature-Verfälschung. Um das Fehlen eines unterstützenden Effekts in antezedens-reflexiven Abhängigkeiten zu berücksichtigen, schlage ich ein individuelles Differenzmodell vor, das innerhalb des Cue-based Retrieval-Frameworks erstellt wurde (Artikel II). Das Modell geht davon aus, dass Individuen sich darin unterscheiden können, wie stark sie eine Syntax gewichten Cue über einem Nummern-Cue. Das Modell wurde an Daten aus zwei Studien zum Antezedenz-Reflexiv angepasst Abhängigkeiten, und die Cue-Gewichtung auf Teilnehmerebene wurde geschätzt. Wir finden, dass ein Viertel von Die Teilnehmer in beiden Studien gewichten bei der Verarbeitung den syntaktischen Cue höher als den numerischen Cue reflexive Abhängigkeiten und die verbleibenden Teilnehmer gewichten die beiden Cue gleichermaßen. Das Ergebnis weist darauf hin, dass das Fehlen des prognostizierten Erleichterungseffekts auf der Ebene der gruppierten Daten von einigen, nicht alle Teilnehmer, die syntaktische Cue höher gewichten als Zahlenhinweise. Allgemeiner gesagt, die Das Ergebnis zeigt, dass die Annahme einer differentiellen Hinweisgewichtung wichtig für eine Theorie von ist Dependency-Completion. Diese Idee der differentiellen Cue-Gewichtung wurde unabhängig unterstützt durch eine Modellierungsstudie zu Subjekt-Verb-Nichteinigungsabhängigkeiten (Artikel III). Insgesamt benötigt der Cue-basierte Abruf, der eine allgemeine Theorie der Abhängigkeitsvervollständigung ist um zwei neue Annahmen aufzunehmen: (i) die im Gedächtnis gespeicherten Substantive können einer Wahrscheinlichkeitsanalyse unterzogen werden Feature-Verfälschung, und (ii) die für den Abruf verwendeten sprachlichen Cue können unterschiedlich gewichtet werden. Das ist das kumulative Ergebnis der in dieser Dissertation vorgestellten Modellierungsarbeit.Die Dissertation leistet einen wichtigen theoretischen Beitrag: Satzverständnis in Der Mensch wird von einem Mechanismus getrieben, der einen hinweisbasierten Abruf, eine probabilistische Merkmalsverzerrung und eine differentielle Hinweisgewichtung annimmt. Diese Einsicht ist theoretisch wichtig, weil es einige gibt unabhängige Unterstützung für diese drei Annahmen in der Satzverarbeitung und im weiteren Gedächtnis Literatur. Die hier vorgestellten Modellierungsarbeiten sind auch methodisch wichtig, weil für die Zum ersten Mal wird gezeigt, (i) wie die komplexen Modelle der Satzverarbeitung evaluiert werden können Daten aus mehreren Studien gleichzeitig zu verwenden, ohne die Modelle zu stark zu vereinfachen, und (ii) wie die Schlussfolgerungen aus dem Verhalten auf individueller Ebene können in der Theorieentwicklung verwendet werden. KW - sentence comprehension KW - individual differences KW - cue-based retrieval KW - memory distortion KW - Approximate Bayesian Computation KW - cue reliability KW - ungefähre Bayessche Komputation KW - Cue-Gewichtung KW - Cue-basierter Retrieval KW - individuelle Unterschiede KW - Darstellung Verfälschung KW - Satzverarbeitung Y1 - 2023 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:kobv:517-opus4-585055 ER - TY - JOUR A1 - Gafos, Adamantios I. A1 - Lieshout, Pascal H. H. M. van T1 - Models and theories of speech production BT - editorial JF - Frontiers in psychology KW - speech production KW - motor control KW - dynamical models KW - phonology KW - speech KW - disorders KW - timing Y1 - 2020 U6 - https://doi.org/10.3389/fpsyg.2020.01238 SN - 1664-1078 VL - 11 PB - Frontiers Research Foundation CY - Lausanne ER - TY - JOUR A1 - Fuhrmeister, Pamela A1 - Smith, Garrett A1 - Myers, Emily B. T1 - Overlearning of non-native speech sounds does not result in superior consolidation after a period of sleep JF - The journal of the Acoustical Society of America N2 - Recent studies suggest that sleep-mediated consolidation processes help adults learn non-native speech sounds. However, overnight improvement was not seen when participants learned in the morning, perhaps resulting from native-language interference. The current study trained participants to perceive the Hindi dental/retroflex contrast in the morning and tested whether increased training can lead to overnight improvement. Results showed overnight effects regardless of training amount. In contrast to previous studies, participants in this study heard sounds in limited contexts (i.e., one talker and one vowel context), corroborating other findings, suggesting that overnight improvement is seen in non-native phonetic learning when variability is limited. Y1 - 2020 U6 - https://doi.org/10.1121/10.0000943 SN - 0001-4966 SN - 1520-8524 VL - 147 IS - 3 SP - EL289 EP - EL294 PB - American Institute of Physics CY - Melville ER - TY - THES A1 - Galetzka, Fabian T1 - Investigating and improving background context consistency in neural conversation models N2 - Neural conversation models aim to predict appropriate contributions to a (given) conversation by using neural networks trained on dialogue data. A specific strand focuses on non-goal driven dialogues, first proposed by Ritter et al. (2011): They investigated the task of transforming an utterance into an appropriate reply. Then, this strand evolved into dialogue system approaches using long dialogue histories and additional background context. Contributing meaningful and appropriate to a conversation is a complex task, and therefore research in this area has been very diverse: Serban et al. (2016), for example, looked into utilizing variable length dialogue histories, Zhang et al. (2018) added additional context to the dialogue history, Wolf et al. (2019) proposed a model based on pre-trained Self-Attention neural networks (Vasvani et al., 2017), and Dinan et al. (2021) investigated safety issues of these approaches. This trend can be seen as a transformation from trying to somehow carry on a conversation to generating appropriate replies in a controlled and reliable way. In this thesis, we first elaborate the meaning of appropriateness in the context of neural conversation models by drawing inspiration from the Cooperative Principle (Grice, 1975). We first define what an appropriate contribution has to be by operationalizing these maxims as demands on conversation models: being fluent, informative, consistent towards given context, coherent and following a social norm. Then, we identify different targets (or intervention points) to achieve the conversational appropriateness by investigating recent research in that field. In this thesis, we investigate the aspect of consistency towards context in greater detail, being one aspect of our interpretation of appropriateness. During the research, we developed a new context-based dialogue dataset (KOMODIS) that combines factual and opinionated context to dialogues. The KOMODIS dataset is publicly available and we use the data in this thesis to gather new insights in context-augmented dialogue generation. We further introduced a new way of encoding context within Self-Attention based neural networks. For that, we elaborate the issue of space complexity from knowledge graphs, and propose a concise encoding strategy for structured context inspired from graph neural networks (Gilmer et al., 2017) to reduce the space complexity of the additional context. We discuss limitations of context-augmentation for neural conversation models, explore the characteristics of knowledge graphs, and explain how we create and augment knowledge graphs for our experiments. Lastly, we analyzed the potential of reinforcement and transfer learning to improve context-consistency for neural conversation models. We find that current reward functions need to be more precise to enable the potential of reinforcement learning, and that sequential transfer learning can improve the subjective quality of generated dialogues. N2 - Neuronale Konversationsmodelle versuchen einen angemessenen Beitrag zu einer (gegebenen) Konversation zu erzeugen, indem neuronale Netze auf Dialogdaten trainiert werden. Ein spezieller Forschungszweig beschäftigt sich mit den nicht-zielgeführten Dialogen, erstmals vorgestellt von Ritter et al. (2011): Das Team untersuchte die Aufgabe der Transformation einer Äußerung in eine angemessene Antwort. Im Laufe der Zeit hat dieser Zweig Dialogsystem-Ansätze hervorgebracht, die lange Konversationen und zusätzlichen Kontext verarbeiten können. Einen sinnvollen und angemessenen Beitrag zu einem Gespräch zu leisten, ist eine komplexe Aufgabe, und daher war die Forschung auf diesem Gebiet sehr vielfältig: Serban et al. (2016) untersuchten beispielsweise die Verwendung von Dialogverläufen variabler Länge, Zhang et al. (2018) fügten der Dialoggeschichte zusätzlichen Kontext hinzu, Wolf et al. (2019) schlugen ein Modell vor, das auf vortrainierten neuronalen Self-Attention Schichten basiert (Vasvani et al., 2017), und Dinan et al. (2021) untersuchten Ansätze zur Kontrolle von unangebrachten Inhalten, wie zum Beispiel Beleidigungen. Dieser Trend kann als Transformation gesehen werden, der vom Versuch, ein Gespräch irgendwie fortzuführen, hin zum kontrollierten und zuverlässigen Generieren angemessener Antworten reicht. In dieser Arbeit untersuchen wir den Aspekt der Kontextkonsistenz genauer, der ein Aspekt unserer Interpretation von einem angemessenen Konversationsbeitrag ist. Während der Untersuchungen haben wir einen neuen kontextbasierten Dialogdatensatz (KOMODIS) entwickelt, der sachlichen und meinungsbezogenen Kontext zu Dialogen kombiniert. Der KOMODIS Datensatz ist öffentlich verfügbar, und wir verwenden die Daten in dieser Arbeit, um neue Einblicke in die kontextunterstützte Dialoggenerierung zu gewinnen. Wir haben außerdem eine neue Methode zur Eingabe von Kontext auf Self-Attention basierenden neuronalen Netzen entwickelt. Dazu erörtern wir zunächst das Problem der begrenzten Eingabelänge für Sequenzen aus Wissensgraphen in solche Modelle, und schlagen eine effiziente Codierungsstrategie für strukturierten Kontext vor, die von Graph Neural Networks inspiriert ist (Gilmer et al., 2017), um die Komplexität des zusätzlichen Kontexts zu reduzieren. Wir diskutieren die Grenzen der Kontexterweiterung für neuronale Konversationsmodelle, untersuchen die Eigenschaften von Wissensgraphen und erklären, wie wir Wissensgraphen für unsere Experimente erstellen und erweitern können. Schließlich haben wir das Potenzial von Reinforcement Learning und Transfer Learning analysiert, um die Kontextkonsistenz für neuronale Konversationsmodelle zu verbessern. Wir stellen fest, dass aktuelle Reward Funktionen präziser sein müssen, um das Potenzial von Reinforcement Learning zu nutzen, und dass Sequential Transfer Learning die subjektive Qualität der generierten Dialoge verbessern kann. KW - conversational ai KW - neural conversation models KW - context consistency KW - gpt KW - conversation KW - dialogue KW - deep learning KW - knowledge graphs KW - Kontextkonsistenz KW - Konversation KW - Dialog KI KW - Deep Learning KW - Dialog KW - GPT KW - Wissensgraph KW - neuronale Konversationsmodelle Y1 - 2022 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:kobv:517-opus4-584637 ER - TY - JOUR A1 - Puebla, Cecilia A1 - Garcia, Juan T1 - Advocating the inclusion of older adults in digital language learning technology and research BT - some considerations JF - Bilingualism : language and cognition Y1 - 2021 U6 - https://doi.org/10.1017/S1366728921000742 SN - 1366-7289 SN - 1469-1841 VL - 25 IS - 3 SP - 398 EP - 399 PB - Cambridge Univ. Press CY - New York ER - TY - JOUR A1 - Ghaffarvand Mokari, Payam A1 - Sardhaei, Nasim Mahdinezhad T1 - Predictive power of cepstral coefficients and spectral moments in the classification of Azerbaijani fricatives JF - The journal of the Acoustical Society of America N2 - This study compares the classification of Azerbaijani fricatives based on two sets of features: (a) spectral moments, spectral peak, amplitude, duration, and (b) cepstral coefficients employing Hidden Markov Models to divide each fricative into three regions such that the variances of the measures within each region are minimized. The cepstral coefficients were found to be more reliable predictors in the classification of all nine Azerbaijani fricatives and the cepstral measures yielded highly successful classification rates (91.21% across both genders) in the identification of the full set of fricatives of Azerbaijani. Y1 - 2020 U6 - https://doi.org/10.1121/10.0000830 SN - 0001-4966 SN - 1520-8524 VL - 147 IS - 3 SP - EL228 EP - EL234 PB - American Institute of Physics CY - Melville ER - TY - JOUR A1 - Krasotkina, Anna A1 - Götz, Antonia A1 - Höhle, Barbara A1 - Schwarzer, Gudrun T1 - Perceptual narrowing in face- and speech-perception domains in infancy BT - a longitudinal approach JF - Infant behavior & development : an international and interdisciplinary journal N2 - During the first year of life, infants undergo a process known as perceptual narrowing, which reduces their sensitivity to classes of stimuli which the infants do not encounter in their environment. It has been proposed that perceptual narrowing for faces and speech may be driven by shared domain-general processes. To investigate this theory, our study longitudinally tested 50 German Caucasian infants with respect to these domains first at 6 months of age followed by a second testing at 9 months of age. We used an infant-controlled habituation-dishabituation paradigm to test the infants' ability to discriminate among other-race Asian faces and non-native Cantonese speech tones, as well as same-race Caucasian faces as a control. We found that while at 6 months of age infants could discriminate among all stimuli, by 9 months of age they could no longer discriminate among other-race faces or non-native tones. However, infants could discriminate among same-race stimuli both at 6 and at 9 months of age. These results demonstrate that the same infants undergo perceptual narrowing for both other-race faces and non-native speech tones between the ages of 6 and 9 months. This parallel development of perceptual narrowing occurring in both the face and speech perception modalities over the same period of time lends support to the domain-general theory of perceptual narrowing in face and speech perception. KW - face perception KW - speech perception KW - longitudinal KW - infant KW - perceptual KW - narrowing Y1 - 2021 U6 - https://doi.org/10.1016/j.infbeh.2021.101607 SN - 0163-6383 SN - 1879-0453 VL - 64 PB - Elsevier CY - New York ER - TY - JOUR A1 - Tran, Thuan T1 - Non-canonical word order and temporal reference in Vietnamese JF - Linguistics : an interdisciplinary journal of the language sciences N2 - The paper revisits Duffield's (2007) (Duffield, Nigel. 2007. Aspects of Vietnamese clausal structure: Separating tense from assertion. Linguistics 45(4). 765-814) analysis of the correlation between the position of a 'when'-phrase and the temporal reference of a bare sentence in Vietnamese. Bare sentences in Vietnamese, based on (Smith, Carlota S. & Mary S. Erbaugh. 2005. Temporal interpretation in Mandarin Chinese. Linguistics 43(4). 713-756), are argued to obtain their temporal interpretation from their aspectual composition, and the default temporal reference: bounded events are located in the past, unbounded events at present. It is shown that the correlation so observed in when-questions is superficial, and is tied to the syntax and semantics of temporal modification and the requirement that temporal adverbials denoting future time is base generated in sentence-initial position, and past time adverbials in sentence-final position. A 'when'-phrase, being temporally underspecified, obtains its temporal value from its base position. However, the correlation between word order and temporal reference in argument wh-questions and declaratives is factual, depending on whether the predicate-argument configuration allows for a telic interpretation or not. To be specific, it is dependent on whether the application of Generic Modification (Snyder, William. 2012. Parameter theory and motion predicates. In Violeta Demonte & Louise McNally (eds.), Telicity, change, and state. Acrosscategorial view of event structure, 279-299. Oxford: Oxford University Press) or accomplishment composition is realized. Canonical declaratives, and argument wh-questions, with telicity inducing material, license GM or accomplishment composition, yielding bounded events, hence past; by contrast, their noncanonical counterparts block GM or accomplishment composition, giving rise to unbounded event descriptions, hence non-past. KW - Vietnamese KW - accomplishment composition KW - temporal reference KW - generic KW - modification KW - temporal modification Y1 - 2021 U6 - https://doi.org/10.1515/ling-2020-0256 SN - 0024-3949 SN - 1613-396X VL - 59 IS - 1 SP - 1 EP - 34 PB - De Gruyter Mouton CY - Berlin ER -