Papers by Afra Alishahi
Analyzing analytical methods: The case of phonology in neural models of spoken language (2020.acl-main)
Copied to clipboard
| Challenge: | Recent studies have focused on the strengths and weaknesses of various methods for analyzing phonology representations. |
| Approach: | They propose to use diagnostic classifiers and representational similarity analysis to quantify to what extent phonemes and phoneme sequences are encoded. |
| Outcome: | The proposed method is based on two commonly applied techniques . it shows that global-scope methods yield more consistent and interpretable results . |
Quantifying Context Mixing in Transformers (2023.eacl-main)
Copied to clipboard
| Challenge: | Self-attention weights and their transformed variants have been used for analyzing token-to-token interactions in Transformer-based models, but they are not faithful to the models’ decisions as they are only one part of an encoder block. |
| Approach: | They propose a new context mixing score customized for Transformers that provides us with a deeper understanding of how information is mixed at each encoder layer. |
| Outcome: | The proposed score outperforms other methods in linguistically informed rationales, probing, and faithfulness analysis. |
Correlating Neural and Symbolic Representations of Language (P19-1)
Copied to clipboard
| Challenge: | a popular technique for analyzing neural representations involves predicting information of interest from the activation patterns. |
| Approach: | They propose to use Representational Similarity Analysis and Tree Kernels to quantify how strongly activation patterns correspond to symbolic representations. |
| Outcome: | The proposed methods show that they exhibit the expected pattern of results on a synthetic language. |
Learning to Understand Child-directed and Adult-directed Speech (2020.acl-main)
Copied to clipboard
| Challenge: | linguistic properties of child-directed speech differ from adult-directed in many ways . linguistic differences between CDS and ADS are retained, but the acoustic properties are similar. |
| Approach: | They compare the task performance of models trained on adult-directed speech and child-directed language . they propose that CDS is optimized for learnability, but not for comprehension . |
| Outcome: | The proposed model trains on adult-directed speech and child-directed language . the model generalizes better on the training register and on synthesized speech . |
Transformer-specific Interpretability (2024.eacl-tutorials)
Copied to clipboard
| Challenge: | Transformers are dominant play-ers in various scientific fields, but their inner workings remain opaque. |
| Approach: | This tutorial presents a trending approach to interpreting Transformers . it uses specific features of the Transformer architecture to quantify context- mixing interactions . |
| Outcome: | This tutorial aims to show how a new trending approach can be applied to Transformer-based models. |
Homophone Disambiguation Reveals Patterns of Context Mixing in Speech Transformers (2023.emnlp-main)
Copied to clipboard
| Challenge: | 'context mixing' is a feature of Transformers that is used to build up representations of acoustic and linguistic structure in speech models. |
| Approach: | They propose to use a French spelling quirk to probe context mixing in speech models to find out how to translate spoken words into written equivalents. |
| Outcome: | The proposed model incorporates cues to identify correct transcription, whereas encoder-decoder models relegate task to decoder modules. |
Revisiting the Hierarchical Multiscale LSTM (C18-1)
Copied to clipboard
| Challenge: | Hierarchical Multiscale LSTM model learns structure from character input . high complexity of architecture, training and implementations might hinder its applicability . |
| Approach: | They propose to reproduce and ablate hierarchical multiscale LSTM language model and show that simplifying certain aspects of the architecture can improve its performance. |
| Outcome: | The proposed model performs better when simplified and linguistic units are learned by different levels of the model. |
Encoding of lexical tone in self-supervised models of spoken language (2024.naacl-long)
Copied to clipboard
| Challenge: | Existing research on representations of phonetic and phonological information has focused on segmental features such as phonemes. |
| Approach: | They propose to analyze the tone encoding capabilities of self-supervised Spoken Language Models, using Mandarin and Vietnamese as case studies. |
| Outcome: | The proposed models encode lexical tone even when trained on non-tonal languages. |
Learning English with Peppa Pig (2022.tacl-1)
Copied to clipboard
| Challenge: | Current approaches to model or simulate the acquisition of spoken language via grounding in perception are not generalizable to real-life situations that humans or adaptive artificial agents experience. |
| Approach: | They propose to use a dataset based on the children’s cartoon Peppa Pig to train a bi-modal architecture that learns aspects of the visual semantics of spoken language. |
| Outcome: | The proposed model learns to represent speech and visual data in a joint vector space. |