Postdoctoral Researcher in Semantically-Guided Natural Language Generation
Are you interested in language generation, semantics, deep learning, and do you want to address the fundamental limitations of current state-of-the-art language models? Join our research team!
We seek a Postdoctoral researcher for the ERC-sponsored project Next-generation Natural Language Generation (NG-NLG). The project aims to provide a quality breakthrough in natural language generation (NLG) with a combination of symbolic and neural approaches. Current NLG is either one or the other: Symbolic NLG (templates, rules, used in the industry) is costly to develop and lacking in output fluency, while neural NLG (language models, research state of the art) is fluent, but uncontrollable and often produces inaccurate outputs.
NG-NLG aims at the best of both by combining symbolic and neural approaches: symbolic approaches for semantics and text planning, and constrained neural language models for text generation. The symbolic component brings controllability and enables logical reasoning and knowledge grounding, while the neural component is essential for producing natural text. The long-term goal is human level in both accuracy and fluency, easy language and domain portability, better model efficiency, and ultimately wider uptake of NLG technology.
What you'll do
You'll be at the core of the project, closely cooperating with the PI on developing semantic formalisms based on existing knowledge graphs and semantic annotation, which will be suitable for text planning and NLG. Your core duties will include:
- Research on semantic formalisms for NLG
- Implementing research systems for NLG and natural language understanding (semantic parsing)
- Collaborating with other team members
- Experimental data collection, processing, and analysis
- Co-supervising graduate students on project-related topics
What we need from you
- Experience at PhD level in natural language processing (NLP) or related areas
- Experience with experimental data collection and analysis
- Experience with development of machine learning and deep learning models for NLP
- Strong programming skills (Python, Pytorch/Tensorflow)
- Good academic writing skills & a publication record in your area
- We offer a full-time (40 hr/week) fixed-term employment contract at Charles University, planned duration is 36 months.
- The gross salary is 60k CZK (ca. 2,400 EUR) per month (ca. 47k CZK/1,900 EUR after taxes)
- 25 days of paid leave per year is standard
- Job location is Prague, Czechia
Charles University is the top-ranking university in Czechia, attracting strong talents locally and internationally. The Institute of Formal and Applied Linguistics, where the ERC project is located, provides the perfect theoretical, personal, as well as material support. It has a decades-long excellent track record in various areas of NLP research involving both neural and symbolic methods (NLG, syntactic and semantic parsing, MT, theoretical linguistics), with extensive international collaboration in EU projects and beyond. The Institute also has an in-house high-performance computing cluster at its disposal, which ensures sufﬁcient computational power for our experiments.
The PI Ondřej Dušek has a 10-year international research experience in the NLP area, including direct practice with handcrafted and neural NLG methods, neural LMs, semantic formalisms as well as theoretical linguistics. He was one of the ﬁrst to deploy neural models in NLG, organized large NLG evaluation campaigns, and has experience with large-scale experiments with real users (Amazon Alexa Prize). He leads a group of 5 PhD students and cooperates closely with other researchers in the 100-strong Institute of Formal and Applied Linguistics.
How to apply
Please send your application per email to email@example.com and include "Postdoc NG-NLG" in the subject. Interviews will be conducted on a rolling basis (starting at the end of May) and the position may be filled as soon as a suitable candidate has been found.
The deadline for application is 31 July 2022, but the position may close sooner if a candidate is found before this date. You are welcome to contact us beforehand if you're considering applying.
Applications should include the following:
- Your CV, including a list of publications
- A motivation letter (max. 1 page), explaining why you are good fit for this position
- A link to your PhD thesis
- Email contacts of 2+ people who we can ask for reference about you
- Optional: links to code you worked on (ideally Git repositories)