Can advanced multi-modal models effectively tackle complex web-based tasks? Such tasks are often found on crowdsourcing platforms, where crowdworkers engage in challenging micro-tasks within web-based environments.Building on this idea, we present TurkingBench, a benchmark consisting of tasks presented as web pages with textual instructions and multi-modal contexts. Unlike previous approaches that rely on artificially synthesized web pages, our benchmark uses natural HTML pages originally designed for crowdsourcing workers to perform various annotation tasks. Each task’s HTML instructions are instantiated with different values derived from crowdsourcing tasks, creating diverse instances. This benchmark includes 32.2K instances spread across 158 tasks.To support the evaluation of TurkingBench, we have developed a framework that links chatbot responses to actions on web pages (e.g., modifying a text box, selecting a radio button). We assess the performance of cutting-edge private and open-source models, including language-only and vision-language models (such as GPT4 and InternVL), on this benchmark. Our results show that while these models outperform random chance, there is still significant room for improvement. We hope that this benchmark will drive progress in the evaluation and development of web-based agents.

To trust the fluent generations of large language models (LLMs), humans must be able to _verify_ their correctness against trusted, external sources. Recent efforts, such as providing citations via retrieved documents or post-hoc provenance, enhance verifiability but provide no guarantees on their correctness. To address these limitations, we tackle the verifiability goal with a different philosophy: _trivializing the verification process by developing models that quote verbatim statements from trusted sources in their pre-training data._We propose Quote-Tuning, which demonstrates the feasibility of aligning models to quote. The core of Quote-Tuning is a fast membership inference function that efficiently verifies text against trusted corpora. We leverage this tool to design a reward function to quantify quotes in model responses, and curate datasets for preference learning. Experiments show that Quote-Tuning significantly increases verbatim quotes from high-quality documents by up to 130% relative to base models while maintaining response quality. Quote-Tuning is applicable in different tasks, generalizes to out-of-domain data and diverse model families, and provides additional benefits to truthfulness. Our method not only serves as a hassle-free method to increase quoting but also opens up avenues for improving LLM trustworthiness through better verifiability.

Recent watermarked generation algorithms inject detectable signatures during language generation to facilitate post-hoc detection. While token-level watermarks are vulnerable to paraphrase attacks, SemStamp (Hou et al., 2023) applies watermark on the semantic representation of sentences and demonstrates promising robustness. SemStamp employs locality-sensitive hashing (LSH) to partition the semantic space with arbitrary hyperplanes, which results in a suboptimal tradeoff between robustness and speed. We propose k-SemStamp, a simple yet effective enhancement of SemStamp, utilizing k-means clustering as an alternative of LSH to partition the embedding space with awareness of inherent semantic structure. Experimental results indicate that k-SemStamp saliently improves its robustness and sampling efficiency while preserving the generation quality, advancing a more effective tool for machine-generated text detection.

As the influence of large language models (LLMs) spans across global communities, their safety challenges in multilingual settings become paramount for alignment research. This paper examines the variations in safety challenges faced by LLMs across different languages and discusses approaches to alleviating such concerns. By comparing how state-of-the-art LLMs respond to the same set of malicious prompts written in higher- vs. lower-resource languages,we observe that (1) LLMs tend to generate unsafe responses much more often when a malicious prompt is written in a lower-resource language, and (2) LLMs tend to generate more irrelevant responses to malicious prompts in lower-resource languages. To understand where the discrepancy can be attributed, we study the effect of instruction tuning with reinforcement learning from human feedback (RLHF) or supervised finetuning (SFT) on the HH-RLHF dataset. Surprisingly, while training with high-resource languages improves model alignment, training in lower-resource languages yields minimal improvement. This suggests that the bottleneck of cross-lingual alignment is rooted in the pretraining stage. Our findings highlight the challenges in cross-lingual LLM safety, and we hope they inform future research in this direction.

Existing watermarked generation algorithms employ token-level designs and therefore, are vulnerable to paraphrase attacks. To address this issue, we introduce watermarking on the semantic representation of sentences. We propose SemStamp, a robust sentence-level semantic watermarking algorithm that uses locality-sensitive hashing (LSH) to partition the semantic space of sentences. The algorithm encodes and LSH-hashes a candidate sentence generated by a language model, and conducts rejection sampling until the sampled sentence falls in watermarked partitions in the semantic embedding space. To test the paraphrastic robustness of watermarking algorithms, we propose a “bigram paraphrase” attack that produces paraphrases with small bigram overlap with the original sentence. This attack is shown to be effective against existing token-level watermark algorithms, while posing only minor degradations to SemStamp. Experimental results show that our novel semantic watermark algorithm is not only more robust than the previous state-of-the-art method on various paraphrasers and domains, but also better at preserving the quality of generation.

In this work, we explore a useful but often neglected methodology for robustness analysis of text generation evaluation metrics: stress tests with synthetic data. Basically, we design and synthesize a wide range of potential errors and check whether they result in a commensurate drop in the metric scores. We examine a range of recently proposed evaluation metrics based on pretrained language models, for the tasks of open-ended generation, translation, and summarization. Our experiments reveal interesting insensitivities, biases, or even loopholes in existing metrics. For example, we find that BERTScore is confused by truncation errors in summarization, and MAUVE (built on top of GPT-2) is insensitive to errors at the beginning or middle of generations. Further, we investigate the reasons behind these blind spots and suggest practical workarounds for a more reliable evaluation of text generation. We have released our code and data at https://github.com/cloudygoose/blindspot_nlg.

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.

The rampant proliferation of large language models, fluent enough to generate text indistinguishable from human-written language, gives unprecedented importance to the detection of machine-generated text. This work is motivated by an important research question: How will the detectors of machine-generated text perform on outputs of a new generator, that the detectors were not trained on? We begin by collecting generation data from a wide range of LLMs, and train neural detectors on data from each generator and test its performance on held-out generators. While none of the detectors can generalize to all generators, we observe a consistent and interesting pattern that the detectors trained on data from a medium-size LLM can zero-shot generalize to the larger version. As a concrete application, we demonstrate that robust detectors can be built on an ensemble of training data from medium-sized models.

Existing work on controlled text generation (CTG) assumes a control interface of categorical attributes. In this work, we propose a natural language (NL) interface, where we craft a PCFG to embed the control attributes into natural language commands, and propose variants of existing CTG models that take commands as input. In our experiments, we design tailored setups to test the model’s generalization abilities. We find our PCFG-based command generation approach is effective for handling unseen commands compared to fix-set templates. Further, our proposed NL models can effectively generalize to unseen attributes (a new ability enabled by the NL interface), as well as unseen attribute combinations. Interestingly, in model comparisons, the simple conditional generation approach, enhanced with our proposed NL interface, is shown to be a strong baseline in those challenging settings.

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.

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.