Columbia Games Corpus

<h3>Introduction</h3></br>
<p>Columbia Games Corpus was developed by the <a href="http://www.cs.columbia.edu/speech/index.cgi">Spoken Language Group, Columbia University </a> and the <a href="https://linguistics.northwestern.edu/">Department of Linguistics, Northwestern University</a>. It consists of approximately 10 hours of spontaneous English conversation along with corresponding orthographic transcripts and annotation. Speech recordings are comprised of two subjects playing a series of computer games requiring verbal communication to achieve joint goals of identifying and moving images on the screen to reach a combined number of points.</p></br>
<p>Each player used a separate laptop computer and could not see the screen of the other player. Participants played two games: the Cards Game and the Objects Game. In the Cards Game, one participant described a card and depending on the task in the game, the second participant searched for the described card or tried to match it from cards shown on their screen. In the Objects Game, each player's screen displayed 5-7 objects, one of which was the target object. One player described the target object's location on their screen, and the other player tried to move that object to the same position on their screen.</p></br>
<h3>Data</h3></br>
<p>Over 12 sessions conducted in 2004, 13 subjects (six female, seven male) participated in the collection. Sessions contained an average of 45 minutes of dialogue.</p></br>
<p>Each recording has corresponding manually time-aligned orthographic transcripts, affirmative cue words discourse annotation, and turn-taking annotation. Annotation guidelines are included in this release. Task files for each game are also included for each recording.</p></br>
<p>Audio data was recorded at a sample rate of 48kHz with 16-bit precision, and later converted to 16kHz, single channel FLAC compressed WAV. All text data is encoded in UTF-8.</p></br>
<h3>Samples</h3></br>
<p>Please view these samples:</p></br>
<ul></br>
<li><a href="desc/addenda/LDC2021S02.flac">Audio (FLAC)</a></li></br>
<li><a href="desc/addenda/LDC2021S02.turns">Turns (TXT)</a></li></br>
<li><a href="desc/addenda/LDC2021S02.words">Words (TXT)</a></li></br>
<li><a href="desc/addenda/LDC2021S02.dm">Discourse Markers (TXT)</a></li></br>
</ul></br>
<h3>Updates</h3></br>
<p>None at this time.</p></br>
Portions © 2021 The Trustees of Columbia University, © 2021 Trustees of the University of Pennsylvania

<h3>引言</h3></br><p>哥伦比亚游戏语料库（Columbia Games Corpus）由哥伦比亚大学口语语言研究组（Spoken Language Group, Columbia University）与西北大学语言学系（Department of Linguistics, Northwestern University）联合开发。该语料库包含约10小时的自发英语对话数据，配套对应正字法转录文本与标注信息。语音录制内容来自两名受试者通过电脑游戏展开的互动对话，双方需通过言语交流协作完成屏幕上的图像识别与移动任务，以达成累计得分的共同目标。</p></br><p>每名受试者使用独立笔记本电脑，且无法查看对方的屏幕。参与者需完成两款游戏：卡牌游戏与物品游戏。在卡牌游戏中，一名受试者描述一张卡牌，根据游戏任务要求，另一名受试者需搜寻对应卡牌，或从屏幕显示的卡牌中匹配目标卡牌。在物品游戏中，每名受试者的屏幕会显示5至7个物品，其中一个为目标物品。一名受试者描述目标物品在其屏幕上的位置，另一名受试者需将己方屏幕上的对应物品移动至相同位置。</p></br><h3>数据概况</h3></br><p>该语料库的采集工作于2004年开展，共完成12轮会话，招募13名受试者（女性6名，男性7名）。每轮会话的平均对话时长为45分钟。</p></br><p>每份录音均配套经过人工时间对齐的正字法转录文本、确认性提示词话语标注与轮次标注。本发布包中包含标注指南，同时为每份录音提供对应游戏的任务文件。</p></br><p>音频数据以48kHz采样率、16位精度录制，后续转换为16kHz单声道FLAC压缩WAV格式。所有文本数据均采用UTF-8编码。</p></br><h3>示例</h3></br><p>请查看以下示例：</p></br><ul></br><li><a href="desc/addenda/LDC2021S02.flac">音频（FLAC）</a></li></br><li><a href="desc/addenda/LDC2021S02.turns">轮次信息（TXT）</a></li></br><li><a href="desc/addenda/LDC2021S02.words">词汇文本（TXT）</a></li></br><li><a href="desc/addenda/LDC2021S02.dm">话语标记（TXT）</a></li></br></ul></br><h3>更新记录</h3></br><p>暂无更新记录。</p></br><p>部分内容 © 2021 哥伦比亚大学理事会，© 2021 宾夕法尼亚大学理事会</p>

搜集汇总

背景与挑战

背景概述

Columbia Games Corpus 是一个由哥伦比亚大学和西北大学开发的英语对话数据集，包含约10小时的自发性对话录音、正交转录和标注，数据采集于2004年，涉及13名参与者在计算机游戏（卡片游戏和对象游戏）中的互动。该数据集特点包括多模态数据（音频采样率16kHz、文本UTF-8编码）、游戏化对话场景以促进自然语言交流，以及详细的标注如肯定提示词和轮换标注，适用于话语分析和韵律研究。