Large language models have achieved impressive few-shot performance on a wide variety of tasks. However, in many settings, users require confidence estimates for model predictions. While traditional classifiers produce scores for each label, language models instead produce scores for the generation which may not be well calibrated. We compare generations across diverse prompts and show that these can be used to create confidence scores. By utilizing more prompts we can get more precise confidence estimates and use response diversity as a proxy for confidence. We evaluate this approach across ten multiple-choice question-answering datasets using three models: T0, FLAN-T5, and GPT-3. In addition to analyzing multiple human written prompts, we automatically generate more prompts using a language model in order to produce finer-grained confidence estimates. Our method produces more calibrated confidence estimates compared to the log probability of the answer to a single prompt. These improvements could benefit users who rely on prediction confidence for integration into a larger system or in decision-making processes.

Grapheme-to-phoneme conversion is an important component in many speech technologies, but until recently there were no multilingual benchmarks for this task. The third iteration of the SIGMORPHON shared task on multilingual grapheme-to-phoneme conversion features many improvements from the previous year’s task (Ashby et al., 2021), including additional languages, three subtasks varying the amount of available resources, extensive quality assurance procedures, and automated error analyses. Three teams submitted a total of fifteen systems, at best achieving relative reductions of word error rate of 14% in the crosslingual subtask and 14% in the very-low resource subtask. The generally consistent result is that cross-lingual transfer substantially helps grapheme-to-phoneme modeling, but not to the same degree as in-language examples.

Location information can support social media analyses by providing geographic context. Some of the most accurate and popular Twitter geolocation systems rely on rule-based methods that examine the user-provided profile location, which fail to handle informal or noisy location names. We propose Geo-Seq2seq, a sequence-to-sequence (seq2seq) model for Twitter user geolocation that rewrites noisy, multilingual user-provided location strings into structured English location names. We train our system on tens of millions of multilingual location string and geotagged-tweet pairs. Compared to leading methods, our model vastly increases coverage (i.e., the number of users we can geolocate) while achieving comparable or superior accuracy. Our error analysis reveals that constrained decoding helps the model produce valid locations according to a location database. Finally, we measure biases across language, country of origin, and time to evaluate fairness, and find that while our model can generalize well to unseen temporal data, performance does vary by language and country.

Metrics for Inter-Annotator Agreement (IAA), like Cohen’s Kappa, are crucial for validating annotated datasets. Although high agreement is often used to show the reliability of annotation procedures, it is insufficient to ensure or reproducibility. While researchers are encouraged to increase annotator agreement, this can lead to specific and tailored annotation guidelines. We hypothesize that this may result in diverging annotations from different groups. To study this, we first propose the Lee et al. Protocol (LEAP), a standardized and codified annotation protocol. LEAP strictly enforces transparency in the annotation process, which ensures reproducibility of annotation guidelines. Using LEAP to annotate a dialog dataset, we empirically show that while research groups may create reliable guidelines by raising agreement, this can cause divergent annotations across different research groups, thus questioning the validity of the annotations. Therefore, we caution NLP researchers against using reliability as a proxy for reproducibility and validity.

Researchers across disciplines use Twitter geolocation tools to filter data for desired locations. These tools have largely been trained and tested on English tweets, often originating in the United States from almost a decade ago. Despite the importance of these tools for data curation, the impact of tweet language, country of origin, and creation date on tool performance remains largely unknown. We explore these issues with Carmen, a popular tool for Twitter geolocation. To support this study we introduce Carmen 2.0, a major update which includes the incorporation of GeoNames, a gazetteer that provides much broader coverage of locations. We evaluate using two new Twitter datasets, one for multilingual, multiyear geolocation evaluation, and another for usage trends over time. We found that language, country origin, and time does impact geolocation tool performance.

The language of Twitter differs significantly from that of other domains commonly included in large language model training. While tweets are typically multilingual and contain informal language, including emoji and hashtags, most pre-trained language models for Twitter are either monolingual, adapted from other domains rather than trained exclusively on Twitter, or are trained on a limited amount of in-domain Twitter data.We introduce Bernice, the first multilingual RoBERTa language model trained from scratch on 2.5 billion tweets with a custom tweet-focused tokenizer. We evaluate on a variety of monolingual and multilingual Twitter benchmarks, finding that our model consistently exceeds or matches the performance of a variety of models adapted to social media data as well as strong multilingual baselines, despite being trained on less data overall.We posit that it is more efficient compute- and data-wise to train completely on in-domain data with a specialized domain-specific tokenizer.

Twitter is commonly used for civil unrest detection and forecasting tasks, but there is a lack of work in evaluating how civil unrest manifests on Twitter across countries and events. We present two in-depth case studies for two specific large-scale events, one in a country with high (English) Twitter usage (Johannesburg riots in South Africa) and one in a country with low Twitter usage (Burayu massacre protests in Ethiopia). We show that while there is event signal during the events, there is little signal leading up to the events. In addition to the case studies, we train Ngram-based models on a larger set of Twitter civil unrest data across time, events, and countries and use machine learning explainability tools (SHAP) to identify important features. The models were able to find words indicative of civil unrest that generalized across countries. The 42 countries span Africa, Middle East, and Southeast Asia and the events range occur between 2014 and 2019.

Narrative generation is an open-ended NLP task in which a model generates a story given a prompt. The task is similar to neural response generation for chatbots; however, innovations in response generation are often not applied to narrative generation, despite the similarity between these tasks. We aim to bridge this gap by applying and evaluating advances in decoding methods for neural response generation to neural narrative generation. In particular, we employ GPT-2 and perform ablations across nucleus sampling thresholds and diverse decoding hyperparameters—specifically, maximum mutual information—analyzing results over multiple criteria with automatic and human evaluation. We find that (1) nucleus sampling is generally best with thresholds between 0.7 and 0.9; (2) a maximum mutual information objective can improve the quality of generated stories; and (3) established automatic metrics do not correlate well with human judgments of narrative quality on any qualitative metric.

We present CUT, a dataset for studying Civil Unrest on Twitter. Our dataset includes 4,381 tweets related to civil unrest, hand-annotated with information related to the study of civil unrest discussion and events. Our dataset is drawn from 42 countries from 2014 to 2019. We present baseline systems trained on this data for the identification of tweets related to civil unrest. We include a discussion of ethical issues related to research on this topic.