DOCSLIB.ORG

Arxiv:2105.06020V1

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Arxiv:2105.06020V1 Are Larger Pretrained Language Models Uniformly Better? Comparing Performance at the Instance Level Ruiqi Zhong Dhruba Ghosh Dan Klein Jacob Steinhardt Computer Science Division, University of California, Berkeley fruiqi-zhong, djghosh13, klein, [email protected] Abstract Prior works hint at differing answers. Hendrycks Larger language models have higher accu- et al.(2020) and Desai and Durrett(2020) find racy on average, but are they better on ev- that larger pretrained models consistently improve ery single instance (datapoint)? Some work out-of-distribution performance, which implies that suggests larger models have higher out-of- they might be uniformly better at a finer level. distribution robustness, while other work sug- Henighan et al.(2020) claim that larger pretrained gests they have lower accuracy on rare sub- image models have lower downstream classifica- groups. To understand these differences, we tion loss for the majority of instances, and they investigate these models at the level of indi- vidual instances. However, one major chal- predict this trend to be true for other data modal- lenge is that individual predictions are highly ities (e.g. text). On the other hand, Sagawa et al. sensitive to noise in the randomness in train- (2020) find that larger non-pretrained models per- ing. We develop statistically rigorous meth- form worse on rare subgroups; if this result gener- ods to address this, and after accounting for alizes to pretrained language models, larger models pretraining and finetuning noise, we find that will not be uniformly better. Despite all the in- our BERT-LARGE is worse than BERT-MINI direct evidence, it is still inconclusive how many on at least 1−4% of instances across MNLI, SST-2, and QQP, compared to the overall ac- instances larger pretrained models perform worse curacy improvement of 2−10%. We also on. find that finetuning noise increases with model A na¨ıve solution is to finetune a larger model, size, and that instance-level accuracy has mo- compare it to a smaller one, and find instances mentum: improvement from BERT-MINI to where the larger model is worse. However, this BERT-MEDIUM correlates with improvement approach is flawed, since model predictions are from BERT-MEDIUM to BERT-LARGE . Our noisy at the instance level. On MNLI in-domain findings suggest that instance-level predictions development set, even the same architecture with provide a rich source of information; we there- fore recommend that researchers supplement different finetuning seeds leads to different pre- model weights with model predictions. dictions on ∼8% of the instances. This is due to under-specification (D’Amour et al., 2020), where 1 Introduction there are multiple different solutions that can mini- Historically, large deep learning models (Peters mize the training loss. Since the accuracy improve- arXiv:2105.06020v1 [cs.CL] 13 May 2021 et al., 2018; Devlin et al., 2019; Lewis et al., 2020; ment from our BERT-BASE1 to BERT-LARGE is Raffel et al., 2019) have improved the state of 2%, most signals across different model sizes will the art on a wide range of tasks and leaderboards be dominated by noise due to random seeds. (Schwartz et al., 2014; Rajpurkar et al., 2016; Wang To account for the noise in pretraining and fine- et al., 2018), and empirical scaling laws predict tuning, we define instance accuracy as “how often that larger models will continue to increase per- a model correctly predicts an instance” (Figure1 formance (Kaplan et al., 2020). However, little is left) in expectation across pretraining and finetun- understood about such improvement at the instance ing seeds. We estimate this quantity by pretraining (datapoint) level. Are larger models uniformly bet- 10 models with different seeds, finetuning 5 times ter? In other words, are larger pretrained models for each pretrained models (Figure1 middle), and better at every instance, or are they better at some 1This is not the original release by Devlin et al.(2019); we instances, but worse at others? pretrained models ourselves. averaging across them. (L4/H256, 4 Layers with hidden dimension 256), However, this estimate is still inexact, and we SMALL (L4/H512), MEDIUM (L8/H512), BASE might falsely observe smaller models to be better (L12/H768), and LARGE (L24/H1024). For each at some instances by chance. Hence, we propose architecture we pre-trained models with 10 differ- a random baseline to estimate the fraction of false ent random seeds and fine-tuned each of them 5 discoveries (Section3, Figure1 right) and formally times (50 total) on each task; see Figure1 middle. upper-bound the false discoveries in Section4. Our Since pretraining is computationally expensive, we method provides a better upper bound than the clas- reduced the context size during pretraining from sical Benjamini-Hochberg procedure with Fisher’s 512 to 128 and compensated by increasing train- exact test. ing steps from 1M to 2M. AppendixA includes Using the 50 models for each size and our im- more details about pretraining and finetuning and proved statistical tool, we find that, on the MNLI their computational cost, and AppendixB verifies in-domain development set, the accuracy “decays” that our cost-saving changes do not affect accuracy from BERT-LARGE to BERT-MINI on at least ∼4% qualitatively. of the instances, which is significant given that the improvement in overall accuracy is 10%. These Notation. We use i to index an instance in the decaying instances contain more controversial or evaluation set, s for model sizes, P for pretraining wrong labels, but also correct ones (Section 4.2). seeds and F for finetuning seeds. c is a random Therefore, larger pretrained language models are variable of value 0 or 1 to indicate whether the not uniformly better. prediction is correct. Given the pretraining seed P s We make other interesting discoveries at the in- and the finetuning seed F , ci = 1 if the model of stance level. Section5 finds that instance-level size s is correct on instance i, 0 otherwise. To keep accuracy has momentum: improvement from MINI the notation uncluttered, we sometimes omit these to MEDIUM correlates with improvement from superscripts or subscripts if they can be inferred MEDIUM to LARGE . Additionally, Section6 at- from context. tributes variance of model predictions to pretrain- Unless otherwise noted, we present results on ing and finetuning random seeds, and finds that the MNLI in-domain development set in the main finetuning seeds cause more variance for larger paper. models. Our findings suggest that instance-level predictions provide a rich source of information; 3 Comparing Instance Accuracy we therefore recommend that researchers supple- ment model weights with model predictions. In this To find the instances where larger models are worse, spirit, we release all the pretrained models, model a na¨ıve approach is to finetune a larger pretrained predictions, and code here: https://github.com/ model, compare it to a smaller one, and find in- ruiqi-zhong/acl2021-instance-level. stances where the larger is incorrect but the smaller 2 Data, Models, and Predictions is correct. Under this approach, BERT-LARGE is worse than BERT-BASE on 4.5% of the instances To investigate model behavior, we considered dif- and better on 7%, giving an overall accuracy im- ferent sizes of the BERT architecture and fine- provement of 2.5%. tuned them on Quora Question Pairs (QQP2), However, this result is misleading: even if we Multi-Genre Natural Language Inference (MNLI; compare two BERT-BASE model with different Williams et al.(2020)), and the Stanford Sen- finetuning seeds, their predictions differ on 8% of timent Treebank (SST-2; Socher et al.(2013)). the instances, while their accuracies differ only by To account for pretraining and finetuning noise, 0.1%; Table1 reports this baseline randomness we averaged over multiple random initializations across model sizes. Changing the pretraining seed and training data order, and thus needed to pre- also changes around 2% additional predictions be- train our own models rather than downloading yond finetuning. off the internet. Following Turc et al.(2019) we Table1 also reports the standard deviation of trained 5 architectures of increasing size: MINI overall accuracy, which is about 40 times smaller. 2https://www.quora.com/q/quoradata/First-Quora- Such stability starkly contrasts with the noisiness at Dataset-Release-Question-Pairs the instance level, which poses a unique challenge. Instances Instance 1 Instance i Average across seeds Model Finetuning Seeds Instance F(*)1 F(*)2 Seed 1 Seed 2 Seed 3 Sizes Accuracy Pretrain P1 X ✓ X X Instance 1 X ✓ ✓ 66% MINI -ing P2 ✓ X … ✓ X Instance 2 X X X 0% Seeds P3 X ✓ ✓ ✓ Instance 3 ✓ X X 33% … … … Instance 4 ✓ ✓ ✓ 100% ✓ ✓ X ✓ LARGE ✓ X … ✓ X ✓ ✓ X X Figure 1: Left: Each column represents the same architecture trained with a different seed. We calculate accuracy for each instance (row) by averaging across seeds (column), while it is usually calculated for each model by averaging across instances. Middle: A visual layout of the model predictions we obtain, which is a binary-valued tensor with 4 axes: model size s, instance i, pretraining seeds P and finetuning seeds F . Right: for each instance, we calculate the accuracy gain from MINI to LARGE and plot the histogram in blue, along with a random baseline in red. Since the blue distribution has a bigger left tail, smaller models are better at some instances. ^ s DiffFTune DiffPTrain Stdall accuracy estimates Acci , i.e. MINI 7.2% 10.7% 0.2% s s SMALL 7.2% 10.7% 0.3% s1 ^ ^ 2 ^ 1 s2 ∆Acci := Acci − Acci : (3) MEDIUM 8.0% 10.7% 0.3% BASE 8.5% 10.6% 0.2% BASE ^ We histogram LARGE∆Acci in Figure2 (b). We LARGE 8.6% 10.1% 0.2% observe a unimodal distribution centered near 0, with tails on both sides.
Load more

Recommended publications
  • Lyrical Ballads
    LYRICAL BALLADS Also available from Routledge: A SHORT HISTORY OF ENGLISH LITERATURE Second Edition Harry Blamires ELEVEN BRITISH POETS* An Anthology Edited by Michael Schmidt WILLIAM WORDSWORTH Selected Poetry and Prose Edited by Jennifer Breen SHELLEY Selected Poetry and Prose Edited by Alasdair Macrae * Not available from Routledge in the USA Lyrical Ballads WORDSWORTH AND COLERIDGE The text of the 1798 edition with the additional 1800 poems and the Prefaces edited with introduction, notes and appendices by R.L.BRETT and A.R.JONES LONDON and NEW YORK First published as a University Paperback 1968 Routledge is an imprint of the Taylor & Francis Group This edition published in the Taylor & Francis e-Library, 2005. “To purchase your own copy of this or any of Taylor & Francis or Routledge’s collection of thousands of eBooks please go to www.eBookstore.tandf.co.uk.” Second edition published 1991 by Routledge 11 New Fetter Lane, London EC4P 4EE Simultaneously published in the USA and Canada by Routledge 29 West 35th Street, New York, NY 10001 Introduction and Notes © 1963, 1991 R.L.Brett and A.R.Jones All rights reserved. No part of this book may be reprinted or reproduced or utilized in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. British Library Cataloguing in Publication Data Wordsworth, William 1770–1850 Lyrical ballads: the text of the 1978 edition with the additional 1800 poems and the prefaces.
    [Show full text]
  • 1 MISS MORRIS and the STRANGER By
    MISS MORRIS AND THE STRANGER better eyelashes than mine. I write quite seriously. There is one woman who is above the common weakness of by Wilkie Collins vanity--and she holds the present pen. I. So I gave my lost stranger a lesson in politeness. The lesson took the form of a trap. I asked him if he would WHEN I first saw him, he was lost in one of the Dead like me to show him the way to the inn. He was still Cities of England--situated on the South Coast, and called annoyed at losing himself. As I had anticipated, he Sandwich. bluntly answered: "Yes." Shall I describe Sandwich? I think not. Let us own the "When you were a boy, and you wanted something," I truth; descriptions of places, however nicely they may be said, "did your mother teach you to say 'Please'?" written, are always more or less dull. Being a woman, I naturally hate dullness. Perhaps some description of He positively blushed. "She did," he admitted; "and she Sandwich may drop out, as it were, from my report of our taught me to say 'Beg your pardon' when I was rude. I'll conversation when we first met as strangers in the street. say it now: 'Beg your pardon.'" He began irritably. "I've lost myself," he said. This curious apology increased my belief in his redeeming qualities. I led the way to the inn. He followed "People who don't know the town often do that," I me in silence. No woman who respects herself can endure remarked.
    [Show full text]
  • Samuel Baker the University of Texas at Austin [email protected]
    Samuel Baker The University of Texas at Austin [email protected] 5/13/2009 Scott’s Stoic Characters: Ethics, Sentiment, and Irony in The Antiquary, Guy Mannering, and “The Author of Waverley” Published in Modern Language Quarterly (2009) 70 (4): 443–471: Cite only from published version. https://doi-org.ezproxy.lib.utexas.edu/10.1215/00267929-2009-011 Whomever one reads on the question, it seems that Walter Scott created “the classical form of the historical novel” by standing at a locus of contradiction. According to the most influential such accounts, Scott “portrayed objectively the ruination of past social formations, despite all his human sympathy for, and artistic sensitivity to, the splendid, heroic qualities which they contained”; he could accomplish such feats of mediation because he was “two men … both the prudent Briton and the passionate Scot."1 These accounts of Scott’s contradictory investments–by Georg Lukács and David Daiches, respectively–find an emblem in the title Virginia Woolf gave her essay on Scott's late-life journal. Referring to how Scott installed modern lighting at his pseudo- medieval estate, Woolf calls her piece “Gas at Abbotsford.”2 Such formulations became 1 Georg Lukács, The Historical Novel [1937], Hannah and Stanley Mitchell, tr. (Lincoln, Neb.: Univ. of Nebraska Press, 1983), pp.54-55; David Daiches, “Scott’s Achievement as a Novelist” [1951], in Literary Essays (Chicago: Univ. of Chicago Press, 1967), 92. 2 Virginia Woolf, “Gas at Abbotsford” [1940], in Collected Essays, 4 vols. (New York: Harcourt Brace, 1967), 1:128-138. touchstones for Scott criticism some decades ago.3 Many critics since have adopted the concept of a split Scott that, whatever their other differences, Lukács, Daiches, and Woolf share.
    [Show full text]
  • Commonplace Book of Elizabeth G
    Dickinson College Archives & Special Collections http://archives.dickinson.edu/ Documents Online Title: Commonplace Book of Elizabeth G. Fergusson Date: 1770-1787 Location: MC 2006.3, B1, F1 Contact: Archives & Special Collections Waidner-Spahr Library Dickinson College P.O. Box 1773 Carlisle, PA 17013 717-245-1399 [email protected] [Inside cover] the 13, 1787 Poems [three words erased and written over]written between the Years 1770 and 1787 [original writing underneath shows 1772 and 1777] at Graeme Park by Laura by Mrs Fergusson a few of them by Betsy Graeme but all Written at Graeme park [Full page pasted over top of inside front cover] [one word illegible] 98 Extract from [M Addison?] [top half of words covered] in praise of Poetry “ I have always been of opinion that Virtue Sinks deepest in the heart of man when it is recommended by the powerful Charms of poetry. The most active principle in our Mind is the imagination: To it a good Part makes his Court immediatly, and by this Fairly Takes Care to Gain it first, Our Passions And Inclinations come over next. And our Reason Surenders it Soly at pleasure in the End Thus the whole Souls is insensibly betrad into morality. There is a certain Elevation of Soul a Sedate imagining and a noble turn of virtue that raises the Hero from the plain honest man, To which Verse can only raise as the Bold Metaphors And Sounding members peculiar To the Parts, Rouse up all our Sleeping faculties And alarm all the power of the Soul like Virgils Exclent Tumpet [Page Break] Contents Ode to Spring; Ode to Summer; Ode to autumn.
    [Show full text]
  • Ephesians: a Kingdom Perspective
    Ephesians: a kingdom perspective Allen J Browne Ephesians: a kingdom perspective Ephesians 1 Copyright 2020 © Allen J Browne, Perth, Western Australia (unpublished) About the author: https://allenbrowne.blog/about/ All rights reserved. Permission is granted to distribute the contents of this book in part or in whole for personal or educational use, but not for profit. This permission is conditional on a) giving attribution to the author, and b) not altering the content or misrepresenting the author’s meaning. Scripture quotations: • NIV: The Holy Bible, New International Version, Copyright © 1973, 1978, 1984, 2011 by Biblica, Inc. • ESV: English Standard Version Copyright © 2001 by Crossway Bibles, • NLT: New Living Translation, Copyright © 1996, 2004, 2015 by Tyndale House Foundation. • Translations and paraphrases by the author. Cover photo: Amphitheatre at Ephesus, Copyright © Allen J Browne, 2014. Version: 2020-05 (last saved 2020-05-21) Page 2 Ephesians: a kingdom perspective Ephesians 1 Contents Foreword 5 Part 1: The good news message (Ephesians 1 – 3) 6 Ephesians 1 6 The significance of kingdom in New Testament letters (Eph 1:1) ..................................... 6 The destiny God has planned for us (Eph 1:4-10) ............................................................. 8 Kingdom or Church? (Eph 1:18-23) ................................................................................. 10 Excursus: why “church”? .................................................................................................. 11 Ephesians
    [Show full text]
  • Carmelized Civilization Traein: a Curated Collection of Poems and Prose Edited by Hilary Downey, John F Sherry Jr and John Schouten
    Carmelized Civilization Traein: A curated collection of poems and prose edited by Hilary Downey, John F Sherry Jr and John Schouten Downey, H., Sherry, J. F., & Schouten, J. W. (Eds.) (2020). Carmelized Civilization Traein: A curated collection of poems and prose edited by Hilary Downey, John F Sherry Jr and John Schouten. 1-75. Consumer Culture Theory , Leicester, United Kingdom. Document Version: Publisher's PDF, also known as Version of record Queen's University Belfast - Research Portal: Link to publication record in Queen's University Belfast Research Portal Publisher rights Copyright 2020 the Authors. This work is made available online in accordance with the publisher’s policies. Please refer to any applicable terms of use of the publisher. General rights Copyright for the publications made accessible via the Queen's University Belfast Research Portal is retained by the author(s) and / or other copyright owners and it is a condition of accessing these publications that users recognise and abide by the legal requirements associated with these rights. Take down policy The Research Portal is Queen's institutional repository that provides access to Queen's research output. Every effort has been made to ensure that content in the Research Portal does not infringe any person's rights, or applicable UK laws. If you discover content in the Research Portal that you believe breaches copyright or violates any law, please contact [email protected]. Download date:07. Oct. 2021 New Street and S. Martins Church by Miss E. S. Paget Image from Glimpses of Ancient Leicester by Mrs T. Fielding Johnson Foreword Written by Hilary Downey This year has brought unimaginable changes to work and lifestyle, beyond our comprehension.
    [Show full text]
  • The Prelude As Spiritual Autobiography
    This dissertation has been 61—5131 microfilmed exactly as received WENZEL, Elizabeth Brown, 1927- THE PRELUDE AS SPIRITUAL AUTOBIOGRAPHY. The Ohio State University, Ph.D., 1961 Language and Literature, general University Microfilms, Inc., Ann Arbor, Michigan Copyright by izab eth Brovm Wenzel 1962 THE PRELUDE AS SPIRITUAL AUTOBIOGRAPHY DISSERTATION Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School o f The Ohio S ta te U n iv ersity By Elizabeth Brown Wenzel, B. A., M. A. ***** The Ohio State University 1961 Approved by ^ Adviser v Department of English CONTENTS Introduction............................................................ 1 I "These Autobiographical times of ours" . 10 II "We see but darkly Even when we look • . behind us" 55 III "A linked lay of Truth" .................................................96 IV "Each man is a memory to himself". • • 128 V "Him s e lfe . .a true Poem" .... 172 VI "Ce qu'on dit de soi est toujours poesie" 212 Bibliography .................................................................................... 2J0 Autobiography .................................................................................... 2^5 ii INTRODUCTION 11 The Prelude is not an autobiography," v/rites the most recent of Wordsworth's biographers, but one hundred and forty-one pages la t e r , th e same w riter d e cla r e s th a t in th e poem, "we fin d [Wordsworth] writing what was in effect his spiritual autobiography." Mrs. Mary Moorman's seeming contradiction reflects a confusion that exists in the use of the term "autobiography" both in Wordsworth criticism and in general. Scholars have used the term more and less generically as a category for The Prelude, from de Selincourt's in ju n c tio n th a t "The Prelude i s a great poem, but i t i s a lso the frank autobiography of a great man.
    [Show full text]
  • Childe Harold's Pilgrimage Canto Iv
    CHILDE HAROLD’S PILGRIMAGE CANTO IV Look at the end for Appendix 1: Hobhouse’s four stanzas “in the Childe’s style” and Appendix 2: Gibbon, Chapter 71. Background Byron arrived in Venice on November 10th 1816, and stayed while Hobhouse travelled with members his family to Naples. Unwillingly – for he was most attached to his Venetian mistress, Mariana Segati – Byron went south on April 17th 1817. He paid a short visit to Florence on April 22nd, and then proceeded to Rome, where, with Hobhouse, he stayed between April 29th and May 20th. 1 He returned to Venice on May 28th. He started Childe Harold IV on June 26th, and had finished the first draft by July 29th. He worked on the poem throughout the autumn, stopping only to rough-out Beppo , a poem so diametrically opposed to Childe Harold in matter and idiom that it might have come from another pen. Hobhouse left Venice on January 7th 1818, and Byron wrote to Murray My dear Mr Murray, You’re in damned hurry To set up this ultimate Canto, But (if they don’t rob us) You’ll see M r Hobhouse Will bring it safe in his portmanteau. 2 – The poem was published on April 28th 1818. Influence The fourth and last canto of Byron’s poem shows his holiday with Shelley (palpable for much of Canto III) to be over, and the baleful influence of Hobhouse to have returned. Claire Claremont wrote to her ex-lover on January 12th 1818, after the poem had been dispatched, with Hobhouse, to London.
    [Show full text]
  • Jackson Ryder Jones 42 the Map of My Palm Is a Road Straight Into the Ocean 43 Enjambment 44
    Acknowledgments My eternal gratitude to Professor Major Jackson for his advice, encouragement, kindness, and perhaps most of all his endless patience. Writing this was not any easy task and without his guidance it would have been an impossible one. And a special thanks to my mother for the beautiful cover photograph and for the years of sage gardening advice. 2 Contents Introduction 5 I. My Mother’s Garden 14 Psychiatrist’s Notes (The Way) 15 Fragments 16 Manic 17 Depressive 19 Abilify 20 Seroquel 20 Clozapine 21 Lithium 21 Home 22 Scientific America 23 Psychiatrist’s Notes (Family History) 24 II. The Lies I Tell 26 Fragments 27 Journal, August 18 28 Explaining Death 29 Wanting 30 Waiting 31 Psychiatrist’s Notes (The Numbers) 32 In Response to the Psychiatrist Asking Why I Didn’t Leave a Note 33 3 And It Isn’t That She Didn’t Love Me 34 Psychiatrist’s Notes (Motivation) 35 III. This Weather 37 According to the Intake Nurse 38 When 39 Mis Hijas 40 Ward B 41 Jackson Ryder Jones 42 The Map of My Palm Is a Road Straight Into the Ocean 43 Enjambment 44 IV. Mythology Tells Us 46 Navel Gazing 47 Fragments 48 July 13 50 Transcendence 51 About Her 52 Psychiatrist’s Notes (Flight) 53 4 Introduction Structure & Inspiration This project was born out of a fascination with nontraditional forms of storytelling. While reading works like Don’t Let Me Be Lonely (2004) by Claudia Rankine, The Autobiography of My Mother (1995) by Jamaica Kincaid, and The Beauty of The Husband (2001) by Anne Carson, I was struck by their lack of traditional narrative frameworks.
    [Show full text]
  • THE TRANSFORMATION of the HUMAN PERSON THROUGH CONTEMPLATION: an ANALYSIS of JOHN CASSIAN's CONFERENCES Thesis Submitted to Th
    THE TRANSFORMATION OF THE HUMAN PERSON THROUGH CONTEMPLATION: AN ANALYSIS OF JOHN CASSIAN’S CONFERENCES Thesis Submitted to The College of Arts and Sciences of the UNIVERSITY OF DAYTON In Partial Fulfillment of the Requirements for The Degree of Master of Arts in Theological Studies By Lindsay Horn Dayton, Ohio August 2021 THE TRANSFORMATION OF THE HUMAN PERSON THROUGH CONTEMPLATION: AN ANALYSIS OF JOHN CASSIAN’S CONFERENCES Name: Horn, Lindsay Rose APPROVED BY: ___________________________________ Sandra Yocum, Ph.D. Faculty Advisor ___________________________________ Meghan Henning, Ph.D. Faculty Reader ___________________________________ Jana Bennett, Ph.D. Faculty Reader ___________________________________ Jana Bennett, Ph.D. Department Chair ii ABSTRACT THE TRANSFORMATION OF THE HUMAN PERSON THROUGH CONTEMPLATION: AN ANALYSIS OF JOHN CASSIAN’S CONFERENCES Name: Horn, Lindsay Rose University of Dayton Advisor: Dr. Sandra Yocum This thesis examines the desert mothers’ and fathers’ view of contemplation as a journey toward theosis, or communion with God. To accomplish this, “contemplation” is analyzed within the fifth century document, The Conferences of John Cassian¸ where John Cassian, the document’s author, has conversation with various desert abbas on the monastic life. The argument presented within this thesis is that ascetic practices of renunciation and purification serve as a catalyst for the contemplative life where the contemplator comes face-to-face with God. In the process of experiencing purity of heart, the contemplator transforms into the likeness of God. Also needed in the process of transformation is the cultivation of silence, for it is within silence the contemplator recognizes God’s nearness. The conclusion of this thesis ties together the themes of theological anthropology, renunciation, purity of heart, and contemplation in the story of Martha and Mary Luke 10:38-42.
    [Show full text]
  • Heart of the Insect| [Poems]
    University of Montana ScholarWorks at University of Montana Graduate Student Theses, Dissertations, & Professional Papers Graduate School 1980 Heart of the insect| [Poems] Laurie Blauner The University of Montana Follow this and additional works at: https://scholarworks.umt.edu/etd Let us know how access to this document benefits ou.y Recommended Citation Blauner, Laurie, "Heart of the insect| [Poems]" (1980). Graduate Student Theses, Dissertations, & Professional Papers. 2523. https://scholarworks.umt.edu/etd/2523 This Thesis is brought to you for free and open access by the Graduate School at ScholarWorks at University of Montana. It has been accepted for inclusion in Graduate Student Theses, Dissertations, & Professional Papers by an authorized administrator of ScholarWorks at University of Montana. For more information, please contact [email protected]. COPYRIGHT ACT OF 1976 THIS IS AN UNPUBLISHED MANUSCRIPT IN WHICH COPYRIGHT SUB­ SISTS. ANY FURTHER REPRINTING OF ITS CONTENTS MUST BE APPROVED BY THE AUTHOR, MANSFIELD LIBRARY UNIVERSITY OF MONTANA DATE: FLF P. 11 1980 Heart of the Insect by Laurie Blauner B.A., Sarah Lawrence College, December 1974 Presented in partial fulfillment of the requirements for the degree of Master of Fine Arts UNIVERSITY OF MONTANA December, 1980 Approved by: Chch'rman, Board of Examiners )e<fri, Graduate Scrool f gd Date UMI Number: EP35760 All rights reserved INFORMATION TO ALL USERS The quality of this reproduction is dependent upon the quality of the copy submitted. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if material had to be removed, a note will indicate the deletion.
    [Show full text]
  • E.M. Forster's Short Stories
    View metadata, citation and similar papers at core.ac.uk brought to you by CORE provided by ScholarWorks at Central Washington University Central Washington University ScholarWorks@CWU All Master's Theses Master's Theses 1968 E.M. Forster’s Short Stories Joan Meredith Kerns Central Washington University Follow this and additional works at: https://digitalcommons.cwu.edu/etd Part of the English Language and Literature Commons Recommended Citation Kerns, Joan Meredith, "E.M. Forster’s Short Stories" (1968). All Master's Theses. 845. https://digitalcommons.cwu.edu/etd/845 This Thesis is brought to you for free and open access by the Master's Theses at ScholarWorks@CWU. It has been accepted for inclusion in All Master's Theses by an authorized administrator of ScholarWorks@CWU. For more information, please contact [email protected]. , l E. M. FORSTER'S SHORT STORIES A Thesis Presented to the Graduate Faculty Central Washington State College In Partial Fulfillment of the Requirements for the Degree Master of Arts by Joan Meredith Kerns Augu.st 1968 .,. N011331100 1Vt3Y APPROVED FOR THE GRADUATE FACULTY ________________________________ H. L. Anshutz, COMMITTEE CHAIRMAN _________________________________ Anthony Canedo _________________________________ Frank M. Collins TABLE OF CONTENTS CHAPTER PAGE I. INTRODUCTION .AND REVIEW OF CRITICISM • • • • • • • l Introduction • • • • • • • • • • • • • • • • • • l Review of Criticism • • • • • • • • • • • • • • 2 II. VOLUME ONE OF THE SHORT STORIES 0 • • • • • • • • ll "The Celestial Omnibus" • • • • • • • • • • • • ll 11 "The Story of a Panic • o • • • • • • • • • • • 20 "The Curate's Friend" • • • • 0 • • • • • • • • 28 "Other Kingdom" • • • • • • • • • • • • • • • • 33 "The Other Side of the Hedge" • • • • • • • • • 38 "The Road from Colonus" 0 • • • 0 • • • • • • • 43 III. VOLUME TWO: -THE ETERNAL MOMENT ---------AND OTHER STORIES 49 "The Point of It".
    [Show full text]