Cennet Oguz's research while affiliated with Deutsches Forschungszentrum für Künstliche Intelligenz and other places

... However, there are no existing high-quality avatar-text datasets to support supervised text-driven 3D avatar generation. As for animating 3D avatars, a few attempts [Ghosh et al. 2021;Tevet et al. 2022] have been made towards text-driven motion generation by leveraging a motion-text dataset. Nonetheless, restricted by the scarcity of paired motion-text data, those fully supervised methods have limited generalizability. ...

... They used the KIT Motion-Language Dataset, reaching PCK scores of 70% when using a threshold of 55 mm. Another work [19] used complex natural language sentences to generate motion through an encoder-decoder structure. They used a hierarchical two-stream model (pose and sentence encoders) along with a pose discriminator. ...

... Recently, have applied Xu and Choi's (2020) re-implementation of Lee et al.'s span-based entity coreference model to resolve the deictic anaphors in the DD track of the CODI-CRAC 2021 shared task after augmenting it with a type prediction model (see Section 4). Not only did they achieve the highest score on each dataset, but they beat the second-best system (Anikina et al., 2021), which is a non-span-based neural approach combined with hand-crafted rules, by a large margin. These results suggest that a spanbased approach to DD resolution holds promise. ...

... Some prevalent methods of few-shot sequence labeling usually focus on token-level metric learning [12,20,26,34,44,50,68], in which the model assigns a label to each query token based on a learned distance metric. For example, Fritzler et al. [20] construct prototype for each class to classify query tokens and Das et al. [12] optimize token-level Gaussian-distributed embeddings to classify query tokens. ...

... Slot filling, as an important part of the task-oriented dialogue system, is mainly used to extract specific information in user utterances. Traditional supervised methods have shown remarkable performance in slot filling tasks (Liu and Lane, 2016;Goo et al., 2018;E et al., 2019;He et al., 2020b;Wu et al., 2020;He et al., 2020a;Oguz and Vu, 2020;Qin et al., 2020), but they require a large amount of domain-specific labeled data. ...

... Indeed, empirical tests of SGNS and PMI demonstrate that SGNS is extremely capable of preserving second-order context overlap-even weighting this higher than first-order context overlap-while PMI is completely incapable of capturing it at all. In a simulation experiment performed by Schlechtweg et al. 34 , the average cosine distance between words with first-and second-order context overlap were 0.11 and 0.00 respectively using SGNS and 0.51 and 1.0 using PMI. While matrix factorization of the PMI matrix is also able to capture such higher-order effects, Levy and Goldberg establish that in practice SGNS arrives at a different result than factorization of the PMI matrix, and that pure factorization does not perform well on many NLP tasks 33 . ...