Szilvia Németh

I am assisting in the building of an annotated dataset (corpus) of the Komi-Permyak language for linguistic research. My tasks are: - the planning of the structure of the corpus, - the creation of user manuals for the of FLEX software, - supporting the linguists’ work using Python scripts: including the collection of texts from various Internet sources, the digitisation of paper-based texts, and the building of a morphology analyser of Komi-Permyak, - creating the project website (using Bootstrap 5).

Examined semantic networks and the grammaticalization of posture verbs in various languages. Tasks: • data collection (from various printed and digital sources) and labelling data in Excel, • performing descriptive data analysis with Python (Pandas), documenting the research with Jupyter Notebook, and visualising the data using Excel and Matplotlib, • publishing the results in academic journals.

Examined how information is delivered in spontaneous speech through intonation units in Mansi. • Performed descriptive data analysis: searched for patterns and trends in complex, multivariable data sets and visualized the data with Excel and Python (Pandas, NumPy, Matplotlib). • Documented the research with Jupyter Notebook, and published the findings in academic journals, gave presentations at scientific conferences and popular science events. • Led BA and MA courses.

I collaborated with the research group of the HUN-REN Hungarian Research Centre for Linguistics and conducting further research on posture verbs.

Belfry.io was a multiplatform tool that was able to identify hate speech automatically in user-generated comments (using NLP tools and AI), in order to help moderate comment streams. • Conducted customer interviews to better understand customer needs. • Supervised a team of five colleagues: worked closely with developers, machine learning engineers, and a UX/UI expert. • Wrote manuals for end users of Belfry.io, and trained them to use the software. • Gave investor pitch decks and meetup talks.

Belfry.io was a multiplatform tool that was able to identify hate speech automatically in user-generated comments, in order to help moderate comment streams. • Identify data sets with high accuracy for training machine learning models, and set a plan for collecting the data. • Created gold-standard data sets for evaluating machine learning models. • Conducted SQL queries in BigQuery database to create ontologies and controlled vocabularies. • Adapted NLP tools (e.g., HunSpell command-line tool).

The goal of the project was to create databases of 3 endangered languages that are difficult to research in the absence of audio material. • Designed the structure of the databases: selected the IT tools used and elaborated the annotation principles. • Trained annotators, wrote manuals, and supervised the database construction process (project management). • Collected data using both structured and unstructured interviews, and questionnaire surveys. • Performed data cleaning and descriptive data analysis.

Examined the acoustic-phonetic features of alcoholic influence using automatic tools to identify drunkenness based on speech production. • Designed complex experiments and data collection methods: conducted structured interviews with 51 people, and collecting data via online questionnaire surveys. • Transcribed the interviews, annotated the audio using Praat, and extracted the mathematical data. • Gave presentations at scientific conferences, university courses, and scientific dissemination programs.

The Typological Database of the Ugric Languages displays the significant features of four languages. Tasks: • managing and coordinating the IT tasks of the website containing the database with aspects of the research group, working closely with the web developer, • participating in the coordination of the database and coordinating this work (training of annotators, writing manuals), • collecting data via questionnaire surveys, • researching and writing the contents of the database for the Mansi language.