We explore replacing the declarative memory system of theACT-R cognitive architecture with a distributional semanticsmodel. ACT-R is a widely used cognitive architecture, butscales poorly to big data applications and lacks a robust modelfor learning association strengths between stimuli. Distribu-tional semantics models can process millions of data pointsto infer semantic similarities from language data or to in-fer product recommendations from patterns of user prefer-ences. We demonstrate that a distributional semantics modelcan account for the primacy and recency effects in free recall,the fan effect in recognition, and human performance on it-erated decisions with initially unknown payoffs. The modelwe propose provides a flexible, scalable alternative to ACT-R’s declarative memory at a level of description that bridgessymbolic, quantum, and neural models of cognition.

Computational models of distributional semantics can analyze a corpus to derive representations of word meanings in terms of each word’s relationship to all other words in the corpus. While these models are sensitive to topic (e.g., tiger and stripes) and synonymy (e.g., soar and fly), the models have limited sensitivity to part of speech (e.g., book and shirt are both nouns). By augmenting a holographic model of semantic memory with additional levels of representations, we present evidence that sensitivity to syntax is supported by exploiting associations between words at varying degrees of separation. We find that sensitivity to associations at three degrees of separation reinforces the relationships between words that share part-of-speech and improves the ability of the model to construct grammatical sentences. Our model provides evidence that semantics and syntax exist on a continuum and emerge from a unitary cognitive system.

In this report, we introduce the Gemini 1.5 family of models, representing the next generation of highly compute-efficient multimodal models capable of recalling and reasoning over fine-grained information from millions of tokens of context, including multiple long documents and hours of video and audio. The family includes two new models: (1) an updated Gemini 1.5 Pro, which exceeds the February version on the great majority of capabilities and benchmarks; (2) Gemini 1.5 Flash, a more lightweight variant designed for efficiency with minimal regression in quality. Gemini 1.5 models achieve near-perfect recall on long-context retrieval tasks across modalities, improve the state-of-the-art in long-document QA, long-video QA and long-context ASR, and match or surpass Gemini 1.0 Ultra's state-of-the-art performance across a broad set of benchmarks. Studying the limits of Gemini 1.5's long-context ability, we find continued improvement in next-token prediction and near-perfect retrieval (>99 as Claude 3.0 (200k) and GPT-4 Turbo (128k). Finally, we highlight real-world use cases, such as Gemini 1.5 collaborating with professionals on completing their tasks achieving 26 to 75 categories, as well as surprising new capabilities of large language models at the frontier; when given a grammar manual for Kalamang, a language with fewer than 200 speakers worldwide, the model learns to translate English to Kalamang at a similar level to a person who learned from the same content.

Knowledge-grounded dialogue generation is a challenging task because it requires satisfying two fundamental yet often competing constraints: being responsive in a manner that is specific to what the conversation partner has said while also being attributable to an underlying source document. In this work, we bring this trade-off between these two objectives (specificity and attribution) to light and ask the question: Can explicit content planning before the response generation help the model to address this challenge? To answer this question, we design a framework called PLEDGE, which allows us to experiment with various plan variables explored in prior work, supporting both metric-agnostic and metric-aware approaches. While content planning shows promise, our results on whether it can actually help to navigate this trade-off are mixed -- planning mechanisms that are metric-aware (use automatic metrics during training) are better at automatic evaluations but underperform in human judgment compared to metric-agnostic mechanisms. We discuss how this may be caused by over-fitting to automatic metrics and the need for future work to better calibrate these metrics towards human judgment. We hope the observations from our analysis will inform future work that aims to apply content planning in this context.

Temperature sampling is a conventional approach to diversify large language model predictions. As temperature increases, the prediction becomes diverse but also vulnerable to hallucinations -- generating tokens that are sensible but not factual. One common approach to mitigate hallucinations is to provide source/grounding documents and the model is trained to produce predictions that bind to and are attributable to the provided source. It appears that there is a trade-off between diversity and attribution. To mitigate any such trade-off, we propose to relax the constraint of having a fixed temperature over decoding steps, and a mechanism to guide the dynamic temperature according to its relevance to the source through KL-divergence. Our experiments justifies the trade-off, and shows that our sampling algorithm outperforms the conventional top-k and top-p algorithms in conversational question-answering and summarization tasks.

Large neural models have brought a new challenge to natural language generation (NLG): It has become imperative to ensure the safety and reliability of the output of models that generate freely. To this end, we present an evaluation framework, Attributable to Identified Sources (AIS), stipulating that NLG output pertaining to the external world is to be verified against an independent, provided source. We define AIS and a two-stage annotation pipeline for allowing annotators to evaluate model output according to annotation guidelines. We successfully validate this approach on generation datasets spanning three tasks (two conversational QA datasets, a summarization dataset, and a table-to-text dataset). We provide full annotation guidelines in the appendices and publicly release the annotated data at https://github.com/google-research-datasets/AIS.

Humans tend to follow the Uniform Information Density (UID) principle by distributing information evenly in utterances. We study if decoding algorithms implicitly follow this UID principle, and under what conditions adherence to UID might be desirable for dialogue generation. We generate responses using different decoding algorithms with GPT-2 on the Persona-Chat dataset and collect human judgments on their quality using Amazon Mechanical Turk. We find that (i) surprisingly, model-generated responses follow the UID principle to a greater extent than human responses, and (ii) decoding algorithms that promote UID do not generate higher-quality responses. Instead, when we control for surprisal, non-uniformity of information density correlates with the quality of responses with very low/high surprisal. Our findings indicate that encouraging non-uniform responses is a potential solution to the ``likelihood trap'' problem (quality degradation in very high-likelihood text). Our dataset containing multiple candidate responses per dialog history along with human-annotated quality ratings is available at https://huggingface.co/datasets/saranya132/dialog_uid_gpt2.

Despite recent progress, it has been difficult to prevent semantic hallucinations in generative Large Language Models. One common solution to this is augmenting LLMs with a retrieval system and making sure that the generated output is attributable to the retrieved information. Given this new added constraint, it is plausible to expect that the overall quality of the output will be affected, for example, in terms of fluency. Can scaling language models help? Here we examine the relationship between fluency and attribution in LLMs prompted with retrieved evidence in knowledge-heavy dialog settings. Our experiments were implemented with a set of auto-metrics that are aligned with human preferences. They were used to evaluate a large set of generations, produced under varying parameters of LLMs and supplied context. We show that larger models tend to do much better in both fluency and attribution, and that (naively) using top-k retrieval versus top-1 retrieval improves attribution but hurts fluency. We next propose a recipe that could allow smaller models to both close the gap with larger models and preserve the benefits of top-k retrieval while avoiding its drawbacks.

This report introduces a new family of multimodal models, Gemini, that exhibit remarkable capabilities across image, audio, video, and text understanding. The Gemini family consists of Ultra, Pro, and Nano sizes, suitable for applications ranging from complex reasoning tasks to on-device memory-constrained use-cases. Evaluation on a broad range of benchmarks shows that our most-capable Gemini Ultra model advances the state of the art in 30 of 32 of these benchmarks - notably being the first model to achieve human-expert performance on the well-studied exam benchmark MMLU, and improving the state of the art in every one of the 20 multimodal benchmarks we examined. We believe that the new capabilities of the Gemini family in cross-modal reasoning and language understanding will enable a wide variety of use cases. We discuss our approach toward post-training and deploying Gemini models responsibly to users through services including Gemini, Gemini Advanced, Google AI Studio, and Cloud Vertex AI.

AI researchers have posited Dungeons and Dragons (D&D) as a challenge problem to test systems on various language-related capabilities. In this paper, we frame D&D specifically as a dialogue system challenge, where the tasks are to both generate the next conversational turn in the game and predict the state of the game given the dialogue history. We create a gameplay dataset consisting of nearly 900 games, with a total of 7,000 players, 800,000 dialogue turns, 500,000 dice rolls, and 58 million words. We automatically annotate the data with partial state information about the game play. We train a large language model (LM) to generate the next game turn, conditioning it on different information. The LM can respond as a particular character or as the player who runs the game—i.e., the Dungeon Master (DM). It is trained to produce dialogue that is either in-character (roleplaying in the fictional world) or out-of-character (discussing rules or strategy). We perform a human evaluation to determine what factors make the generated output plausible and interesting. We further perform an automatic evaluation to determine how well the model can predict the game state given the history and examine how well tracking the game state improves its ability to produce plausible conversational output.

AbstractKnowledge-grounded dialogue systems powered by large language models often generate responses that, while fluent, are not attributable to a relevant source of information. Progress towards models that do not exhibit this issue requires evaluation metrics that can quantify its prevalence. To this end, we introduce the Benchmark for Evaluation of Grounded INteraction (Begin), comprising 12k dialogue turns generated by neural dialogue systems trained on three knowledge-grounded dialogue corpora. We collect human annotations assessing the extent to which the models’ responses can be attributed to the given background information. We then use Begin to analyze eight evaluation metrics. We find that these metrics rely on spurious correlations, do not reliably distinguish attributable abstractive responses from unattributable ones, and perform substantially worse when the knowledge source is longer. Our findings underscore the need for more sophisticated and robust evaluation metrics for knowledge-grounded dialogue. We make Begin publicly available at https://github.com/google/BEGIN-dataset.

Compared to standard retrieval tasks, passage retrieval for conversational question answering (CQA) poses new challenges in understanding the current user question, as each question needs to be interpreted within the dialogue context. Moreover, it can be expensive to re-train well-established retrievers such as search engines that are originally developed for non-conversational queries. To facilitate their use, we develop a query rewriting model CONQRR that rewrites a conversational question in the context into a standalone question. It is trained with a novel reward function to directly optimize towards retrieval using reinforcement learning and can be adapted to any off-the-shelf retriever. CONQRR achieves state-of-the-art results on a recent open-domain CQA dataset containing conversations from three different sources, and is effective for two different off-the-shelf retrievers. Our extensive analysis also shows the robustness of CONQRR to out-of-domain dialogues as well as to zero query rewriting supervision.

Author(s): Tang, Zixin; Putnam, Michael; Reitter, David T | Abstract: We investigated how the language background (L1) of bilinguals influences the representation and use of the second language (L2) through computational models. With the essays part from The International Corpus Network of Asian Learners of English (ICNALE), we compared variables indicating syntactic complexity in their L2 production to predict L1. We then trained neural language models based on BERT to predict the L1 of these English learners. Results showed the systematic influence of L1 syntax properties on English learners' L2 production, which further confirmed integrations of syntactic knowledge across languages in bilingual speakers. Results also found neural models can learn to represent and detect such L1 impacts, while multilingually trained models have no advantage in doing so.

Knowledge-grounded dialogue systems are intended to convey information that is based on evidence provided in a given source text. We discuss the challenges of training a generative neural dialogue model for such systems that is controlled to stay faithful to the evidence. Existing datasets contain a mix of conversational responses that are faithful to selected evidence as well as more subjective or chit-chat style responses. We propose different evaluation measures to disentangle these different styles of responses by quantifying the informativeness and objectivity. At training time, additional inputs based on these evaluation measures are given to the dialogue model. At generation time, these additional inputs act as stylistic controls that encourage the model to generate responses that are faithful to the provided evidence. We also investigate the usage of additional controls at decoding time using resampling techniques. In addition to automatic metrics, we perform a human evaluation study where raters judge the output of these controlled generation models to be generally more objective and faithful to the evidence compared to baseline dialogue systems.

With the essays part from The International Corpus Network of Asian Learners of English (ICNALE) and the TOEFL11 corpus, we fine-tuned neural language models based on BERT to predict English learners’ native languages. Results showed neural models can learn to represent and detect such native language impacts, but multilingually trained models have no advantage in doing so.

Native speakers can judge whether a sentence is an acceptable instance of their language. Acceptability provides a means of evaluating whether computational language models are processing language in a human-like manner. We test the ability of language models, simple language features, and word embeddings to predict native speakers’ judgments of acceptability on English essays written by non-native speakers. We find that much sentence acceptability variance can be captured by a combination of misspellings, word order, and word similarity (r = 0.494). While predictive neural models fit acceptability judgments well (r = 0.527), we find that a 4-gram model is just as good (r = 0.528). Thanks to incorporating misspellings, our 4-gram model surpasses both the previous unsupervised state-of-the art (r = 0.472), and the average native speaker (r = 0.46), demonstrating that acceptability is well captured by n-gram statistics and simple language features.

Computational models of distributional semantics (a.k.a. word embeddings) represent a word's meaning in terms of its relationships with all other words. We examine what grammatical information is encoded in distributional models and investigate the role of indirect associations. Distributional models are sensitive to associations between words at one degree of separation, such as 'tiger' and 'stripes', or two degrees of separation, such as 'soar' and 'fly'. By recursively adding higher levels of representations to a computational, holographic model of semantic memory, we construct a distributional model sensitive to associations between words at arbitrary degrees of separation. We find that word associations at four degrees of separation increase the similarity assigned by the model to English words that share part-of-speech or syntactic type. Word associations at four degrees of separation also improve the ability of the model to construct grammatical English sentences. Our model proposes that human memory uses indirect associations to learn part-of-speech and that the basic associative mechanisms of memory and learning support knowledge of both semantics and grammatical structure.

Why do bilinguals switch languages within a sentence? The present observational study asks whether word surprisal and word entropy predict code-switching in bilingual written conversation. We describe and model a new dataset of Chinese-English text with 1476 clean code-switched sentences, translated back into Chinese. The model includes known control variables together with word surprisal and word entropy. We found that word surprisal, but not entropy, is a significant predictor that explains code-switching above and beyond other well-known predictors. We also found sentence length to be a significant predictor, which has been related to sentence complexity. We propose high cognitive effort as a reason for code-switching, as it leaves fewer resources for inhibition of the alternative language. We also corroborate previous findings, but this time using a computational model of surprisal, a new language pair, and doing so for written language.

We demonstrate that the key components of cognitive architectures (declarative and procedural memory) and their key capabilities (learning, memory retrieval, probability judgment, and utility estimation) can be implemented as algebraic operations on vectors and tensors in a high-dimensional space using a distributional semantics model. High-dimensional vector spaces underlie the success of modern machine learning techniques based on deep learning. However, while neural networks have an impressive ability to process data to find patterns, they do not typically model high-level cognition, and it is often unclear how they work. Symbolic cognitive architectures can capture the complexities of high-level cognition and provide human-readable, explainable models, but scale poorly to naturalistic, non-symbolic, or big data. Vector-symbolic architectures, where symbols are represented as vectors, bridge the gap between the two approaches. We posit that cognitive architectures, if implemented in a vector-space model, represent a useful, explanatory model of the internal representations of otherwise opaque neural architectures. Our proposed model, Holographic Declarative Memory (HDM), is a vector-space model based on distributional semantics. HDM accounts for primacy and recency effects in free recall, the fan effect in recognition, probability judgments, and human performance on an iterated decision task. HDM provides a flexible, scalable alternative to symbolic cognitive architectures at a level of description that bridges symbolic, quantum, and neural models of cognition.