Evaluating open-domain conversation models has been an open challenge due to the open-ended nature of conversations. In addition to static evaluations, recent work has started to explore a variety of per-turn and per-dialog interactive evaluation mechanisms and provide advice on the best setup. In this work, we adopt the interactive evaluation framework and further apply to multiple models with a focus on per-turn evaluation techniques. Apart from the widely used setting where participants select the best response among different candidates at each turn, one more novel per-turn evaluation setting is adopted, where participants can select all appropriate responses with different fallback strategies to continue the conversation when no response is selected. We evaluate these settings based on sensitivity and consistency using four GPT2-based models that differ in model sizes or fine-tuning data. To better generalize to any model groups with no prior assumptions on their rankings and control evaluation costs for all setups, we also propose a methodology to estimate the required sample size given a minimum performance gap of interest before running most experiments. Our comprehensive human evaluation results shed light on how to conduct credible human evaluations of open domain dialog systems using the interactive setup, and suggest additional future directions.

Embodied task completion is a challenge where an agent in a simulated environment must predict environment actions to complete tasks based on natural language instructions and egocentric visual observations.We propose a variant of this problem where the agent predicts actions at a higher level of abstraction called a plan, which helps make agent actions more interpretable and can be obtained from the appropriate prompting of large language models.We show that multimodal transformer models can outperform language-only models for this problem but fall significantly short of oracle plans.Since collecting human-human dialogues for embodied environments is expensive and time-consuming, we propose a method to synthetically generate such dialogues, which we then use as training data for plan prediction.We demonstrate that multimodal transformer models can attain strong zero-shot performance from our synthetic data, outperforming language-only models trained on humanhuman data.

Vishakh Padmakumar, Behnam Hedayatnia, Di Jin, Patrick Lange, Seokhwan Kim, Nanyun Peng, Yang Liu, Dilek Hakkani-Tur. Proceedings of the 24th Annual Meeting of the Special Interest Group on Discourse and Dialogue. 2023.

Robots operating in human spaces must be able to engage in natural language interaction, both understanding and executing instructions, and using conversation to resolve ambiguity and correct mistakes. To study this, we introduce TEACh, a dataset of over 3,000 human-human, interactive dialogues to complete household tasks in simulation. A Commander with access to oracle information about a task communicates in natural language with a Follower. The Follower navigates through and interacts with the environment to complete tasks varying in complexity from "Make Coffee" to "Prepare Breakfast", asking questions and getting additional information from the Commander. We propose three benchmarks using TEACh to study embodied intelligence challenges, and we evaluate initial models' abilities in dialogue understanding, language grounding, and task execution.

Embodied agents need to be able to interact in natural language -understanding task descriptions and asking appropriate follow up questions to obtain necessary information to be effective at successfully accomplishing tasks for a wide range of users.In this work, we propose a set of dialog acts for modelling such dialogs and annotate the TEACh dataset that includes over 3,000 situated, task oriented conversations (consisting of 39.5k utterances in total) with dialog acts.TEACh-DA is one of the first large scale dataset of dialog act annotations for embodied task completion.Furthermore, we demonstrate the use of this annotated dataset in training models for tagging the dialog acts of a given utterance, predicting the dialog act of the next response given a dialog history, and use the dialog acts to guide agent's non-dialog behaviour.In particular, our experiments on the TEACh Execution from Dialog History task where the model predicts the sequence of low level actions to be executed in the environment for embodied task completion, demonstrate that dialog acts can improve end task success rate by up to 2 points compared to the system without dialog acts.

We present a novel statistical shape descriptor for arbitrary three-dimensional shapes as a six-dimensional feature for generic classification purposes. Our feature parameterizes the complete geo-metrical relation of the global shape and additionally considers local dissimilarities while being invariant to the shape appearance. Our approach allows the classification of large-scale shapes with only small local dissimilarities. Our feature can be easily quantized and mapped into a histogram, which can be used for efficient and effective classification. We take advantage of GPU processing in order to efficiently compute our invariant local shape descriptor feature even for large-scale shapes. Our synthetic benchmarks show that our approach outperforms state-of-the-art methods for local shape dissimilarity classification. In general, it yields robust and promising recognition rates even for noisy data.

We present a novel approach for approximating objective functions in arbitrary deterministic and stochastic multi-objective blackbox simulations. Usually, simulated-based optimization approaches require pre-defined objective functions for optimization techniques in order to find a local or global minimum of the specified simulation objectives and multi-objective constraints. Due to the increasing complexity of state-of-the-art simulations, such objective functions are not always available, leading to so-called blackbox simulations.

We present a new graph-based approach called GraphPool for the generation, management and distribution of simulation states for 3D simulation applications. Currently, relational databases are often used for this task in simulation applications. In contrast, our approach combines novel wait-free nested hash map techniques with traditional graphs which results in a schema-less, in-memory, highly efficient data management. Our GraphPool stores static and dynamic parts of a simulation model, distributes changes caused by the simulation and logs the simulation run. Even more, the GraphPool supports sophisticated query types of traditional relational databases. As a consequence, our GraphPool overcomes the associated drawbacks of relational database technology for sophisticated 3D simulation applications. Our GraphPool has several advantages compared to other state-of-the-art decentralized methods, such as persistence for simulation state over time, object identification, standardized interfaces for software components as well as a consistent world model for the overall simulation system. We tested our approach in a synthetic benchmark scenario but also in real-world use cases. The results show that it outperforms state-of-the-art relational databases by several orders of magnitude.

We present a novel knowledge discovery approach for automatic feasible design space approximation and parameter optimization in arbitrary multiobjective blackbox simulations. Our approach does not need any supervision of simulation experts. Usually simulation experts conduct simulation experiments for a predetermined system specification by manually reducing the complexity and number of simulation runs by varying input parameters through educated assumptions and according to prior defined goals. This leads to a error-prone trial-and-error approach for determining suitable parameters for successful simulations.In contrast, our approach autonomously discovers unknown relationships in model behavior and approximates the feasible design space. Furthermore, we show how Pareto gradient information can be obtained from this design space approximation for state-of-the-art optimization algorithms. Our approach gains its efficiency from a novel spline-based sampling of the parameter space in combination within novel forest-based simulation dataflow analysis. We have applied our new method to several artificial and real-world scenarios and the results show that our approach is able to discover relationships between parameters and simulation goals. Additionally, the computed multiobjective solutions are close to the Pareto front.