Yannick Frommherz
Research Associate
at the Chair of Applied Linguistics
Office hours during the lecture period
By arrangement at https://tu-dresden.zoom.us/j/65007057407?pwd=NlNWL0t6YXNCbHZpWlY4cWtSRTNUZz09
Office hours during the lecture-free period
By arrangement at https://tu-dresden.zoom.us/j/65007057407?pwd=NlNWL0t6YXNCbHZpWlY4cWtSRTNUZz09
Vita
2014 - 2017 |
Bachelor's degree in German Studies and Geography at the University of Basel (Switzerland), Bachelor's thesis on topic shifting in political television discussions |
2017 - 2019 |
Master's degree in Linguistics at the University of Lund (Sweden), Master's thesis on verbs of thought in German and Swedish against the background of a universality hypothesis |
2020 - 2021 |
Research Associate in the field of Natural Language Processing (NLP) and data management at the Fraunhofer Institute for Integrated Circuits in Erlangen, Germany |
since November 2021 | Research Associate in the project Experimentierraum Digitale Medienkompetenz (ExDiMed) at the Chair of Applied Linguistics |
Dissertation project
Alignment describes the psycholinguistic phenomenon that interlocutors align with each other on different linguistic levels within an interaction, e.g. on the lexical level: If speaker A introduces the expression x to reference an entity, it is likely that her interlocutor B will take up the same expression x to refer to the same entity (Pickering & Garrod, 2004; Szmrecanyi, 2005; Koulouri et al., 2016). This tendency has been shown for both human-human interaction (Branigan et al., 2000) and human-machine interaction, including for lexis (Branigan & Pearson, 2016; Huiyang & Min, 2022), syntax (Heyselaar et al., 2017) and prosody (Raveh et al, 2019), as well as for various settings such as interactions with chatbots (Lotze, 2016), robots (Fischer, 2016) and voice assistants (Linnemann & Jucks, 2018).
The majority of research on alignment in human-machine interaction is of an experimental nature, whereby the so-called picture-naming-matching task is predominantly used, usually as a wizard-of-oz setup. Test subjects and the machine (or a human disguised as a machine, i.e. a wizard) alternately exchange descriptions of pictures, which the other has to assign to one of several pictures. The machine's description conceals linguistic elements (e.g. a passive construction) to which the respondent then (unconsciously) aligns or does not align in her description (Pearson et al., 2006, Heyselaar et al., 2017). While this design is well controllable, it emulates a scenario that is far removed from real human-machine interaction, which reduces the ecological validity of such findings. On the other hand, a corpus-based approach cannot directly demonstrate alignment processes in the brain, but can only show the persistence of linguistic structures (Lotze, 2016). At the same time, it offers the advantage that (reasonably) authentic interactions between humans and actual machines can be evaluated. Nevertheless, there is hardly any corpus-based research on alignment in human-machine interaction. A laudable exception is Lotze (2016), who analyzed the topic in interactions between humans and chatbots and found extensive evidence for lexical and syntactic persistence. What is still completely lacking is corpus-based research on persistence in interactions between humans and voice assistants, which are becoming increasingly widespread in the form of Siri and Alexa (Byrne et al., 2019; Yuan et al., 2020). This dissertation project aims to fill this research gap by analyzing several German-language corpora with interactions between humans and Alexa for persistence.
Publications
2023
-
Keyness in song lyrics: Challenges of highly clumpy data , 1 May 2023, 36, 1, p. 21-38, 18 p.Electronic (full-text) versionResearch output: Contribution to journal > Research article
2022
-
Digitale Kompetenzen für Geistes- und Sozialwissenschaftler:innen. Vorzüge eines Blended Learning-Formats für die Vermittlung von Programmierkenntnissen. , 5 Jul 2022, 2, 1, 12 p.Electronic (full-text) versionResearch output: Contribution to journal > Research article
2021
-
Thinking Things in German versus Swedish. A Cross-Linguistic comparison of Verbs of Thinking in Two Genetically Close Languages. , 11 Nov 2021, 76, 2, p. 464-506, 43 p.Electronic (full-text) versionResearch output: Contribution to journal > Research article
-
Crowdsourcing Ecologically-Valid Dialogue Data for German , 21 Jun 2021, 3, 686050Electronic (full-text) versionResearch output: Contribution to journal > Research article
Presentations
12/2023 Alignment in human-language assistant interaction (Artificial Friday. Linguistic perspectives on artificial intelligence, December 01, 2023) https://artificial-friday.de/yannick-frommherz-dresden/#info
09/2023 Human-language assistant interaction between naturalness and error-proneness (Annual Conference of the Society for Applied Linguistics, Section Media Communication, September 20 - 22, 2023, Johannes Gutenberg University Mainz) https://www.gal2023.de/
03/2023 "Hello Humanities!" - A Modular Python Programming Course Targeting the Specific Needs and Requirements of Humanities Students (Conference: Programming an Data Infrastructure in Digital Humanities, March 27 - 29, 2023, Universidade de Évora, Portugal) https://indico.hpc.uevora.pt/event/36/
11/2022 Thinking Things in German versus Swedish. A Cross-Linguistic comparison of Verbs of Thinking in Two Genetically Close Languages (Invited speaker at the Grammar Seminar of the Language and Literature Center at Lund University, Sweden) https://www.sol.lu.se/sol/kalendarium/evenemang/grammatikseminarietthe-grammar-seminar-yannick-frommherz-dresden-thinking-things-german-versus
Teaching
Summer semester 2023 | Programming for humanities scholars and social scientists |
winter semester 2022/23 |
Learning programming and applying it to the history of language (with Robert Schuppe) |
Research interests
-
Digital linguistics
-
Interactional linguistics
-
Human-machine interaction