A wide body of evidence shows that human language processing difficulty is predicted by the information-theoretic measure surprisal, a word’s negative log probability in context. However, it is still unclear how to best estimate these probabilities needed for predicting human processing difficulty – while a long-standing belief held that models with lower perplexity would provide more accurate estimates of word predictability, and therefore lead to better reading time predictions, recent work has shown that for very large models, psycholinguistic predictive power decreases. One reason could be that language models might be more confident of their predictions than humans, because they have had exposure to several magnitudes more data. In this paper, we test what effect temperature-scaling of large language model (LLM) predictions has on surprisal estimates and their predictive power of reading times of English texts. Firstly, we show that calibration of large language models typically improves with model size, i.e. poorer calibration cannot account for poorer fit to reading times. Secondly, we find that temperature-scaling probabilities lead to a systematically better fit to reading times (up to 89% improvement in delta log likelihood), across several reading time corpora. Finally, we show that this improvement in fit is chiefly driven by words that are composed of multiple subword tokens.

Hallucinations in text generation occur when the system produces text that is not grounded in the input. In this work, we tackle the problem of hallucinations in neural chart summarization. Our analysis shows that the target side of chart summarization training datasets often contains additional information, leading to hallucinations. We propose a natural language inference (NLI) based method to preprocess the training data and show through human evaluation that our method significantly reduces hallucinations. We also found that shortening long-distance dependencies in the input sequence and adding chart-related information like title and legends improves the overall performance.

We present Barch, a new English dataset of human-written summaries describing bar charts. This dataset contains 47 charts based on a selection of 18 topics. Each chart is associated with one of the four intended messages expressed in the chart title. Using crowdsourcing, we collected around 20 summaries per chart, or one thousand in total. The text of the summaries is aligned with the chart data as well as with analytical inferences about the data drawn by humans. Our datasets is one of the first to explore the effect of intended messages on the data descriptions in chart summaries. Additionally, it lends itself well to the task of training data-driven systems for chart-to-text generation. We provide results on the performance of state-of-the-art neural generation models trained on this dataset and discuss the strengths and shortcomings of different models.

Scripts capture commonsense knowledge about everyday activities and their participants. Script knowledge proved useful in a number of NLP tasks, such as referent prediction, discourse classification, and story generation. A crucial step for the exploitation of script knowledge is script parsing, the task of tagging a text with the events and participants from a certain activity. This task is challenging: it requires information both about the ways events and participants are usually uttered in surface language as well as the order in which they occur in the world. We show how to do accurate script parsing with a hierarchical sequence model and transfer learning. Our model improves the state of the art of event parsing by over 16 points F-score and, for the first time, accurately tags script participants.

We present results of the first gender classification experiments on Slovene text to our knowledge. Inspired by the TwiSty corpus and experiments (Verhoeven et al., 2016), we employed the Janes corpus (Erjavec et al., 2016) and its gender annotations to perform gender classification experiments on Twitter text comparing a token-based and a lemma-based approach. We find that the token-based approach (92.6% accuracy), containing gender markings related to the author, outperforms the lemma-based approach by about 5%. Especially in the lemmatized version, we also observe stylistic and content-based differences in writing between men (e.g. more profane language, numerals and beer mentions) and women (e.g. more pronouns, emoticons and character flooding). Many of our findings corroborate previous research on other languages.