Reto Gubelmann

Hongji Li, Andrianos Michail, Reto Gubelmann et al.

Controllable and transparent text generation has been a long-standing goal in NLP. Almost as long-standing is a general idea for addressing this challenge: Parsing text to a symbolic representation, and generating from it. However, earlier approaches were hindered by parsing and generation insufficiencies. Using modern parsers and a safety supervision mechanism, we show how close current methods come to this goal. Concretely, we propose the Sentence Smith framework for English, which has three steps: 1. Parsing a sentence into a semantic graph. 2. Applying human-designed semantic manipulation rules. 3. Generating text from the manipulated graph. A final entailment check (4.) verifies the validity of the applied transformation. To demonstrate our framework's utility, we use it to induce hard negative text pairs that challenge text embedding models. Since the controllable generation makes it possible to clearly isolate different types of semantic shifts, we can evaluate text embedding models in a fine-grained way, also addressing an issue in current benchmarking where linguistic phenomena remain opaque. Human validation confirms that our transparent generation process produces texts of good quality. Notably, our way of generation is very resource-efficient, since it relies only on smaller neural networks.

Reto Gubelmann, Siegfried Handschuh

In this article, we explore the shallow heuristics used by transformer-based pre-trained language models (PLMs) that are fine-tuned for natural language inference (NLI). To do so, we construct or own dataset based on syllogistic, and we evaluate a number of models' performance on our dataset. We find evidence that the models rely heavily on certain shallow heuristics, picking up on symmetries and asymmetries between premise and hypothesis. We suggest that the lack of generalization observable in our study, which is becoming a topic of lively debate in the field, means that the PLMs are currently not learning NLI, but rather spurious heuristics.

Reto Gubelmann, Peter Hongler, Siegfried Handschuh

In this article, we explore the potential of transformer-based language models (LMs) to correctly represent normative statements in the legal domain, taking tax law as our use case. In our experiment, we use a variety of LMs as bases for both word- and sentence-based clusterers that are then evaluated on a small, expert-compiled test-set, consisting of real-world samples from tax law research literature that can be clearly assigned to one of four normative theories. The results of the experiment show that clusterers based on sentence-BERT-embeddings deliver the most promising results. Based on this main experiment, we make first attempts at using the best performing models in a bootstrapping loop to build classifiers that map normative claims on one of these four normative theories.