nykek5i 发表于 2024-8-18 23:52:42

AI诗人的十四行诗,能赛过莎翁的吗?

<img src="https://mmbiz.qpic.cn/mmbiz_png/SE4iaQdbe1bXGcibOmJCk9OMZ16f6x2d1KiakATIBuDJsMBGX72MB0cYYdPmlEOUfNXjQ16SI6hVRStJIGSVX1t5Q/640?wx_fmt=png&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1&amp;wx_co=1" style="width: 50%; margin-bottom: 20px;">
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">人工智能创作简史</strong></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">这是<strong style="color: blue;">莎士比亚</strong>第18首十四行诗中的一个诗节:</p><img src="https://mmbiz.qpic.cn/mmbiz_jpg/SE4iaQdbe1bXGcibOmJCk9OMZ16f6x2d1KERLh7ZzwqiadeW9cKicLiaiba6D1hkFFUMgBtE0C5p0XnVAicJX3XMd0xsg/640?wx_fmt=jpeg&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1&amp;wx_co=1" style="width: 50%; margin-bottom: 20px;">
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">Shall I compare thee to a summers day?</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">thou art more lovely and more temperate:</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">rough winds do shake the darling buds of May,</p>and summers lease hath all too short a date
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">我可能把你和夏天相比拟?</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">你比夏天更可爱更温和:</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">暴风</span>会把五月的花苞吹落地,</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">夏天<span style="color: black;">亦</span>嫌太短促,匆匆而过*</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">*译者注</strong></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">&nbsp;本文中<span style="color: black;">显现</span>的莎士比亚十四行诗中文译作除非<span style="color: black;">尤其</span>说明,均为梁实秋译本。</p><img src="https://mmbiz.qpic.cn/mmbiz_jpg/SE4iaQdbe1bXGcibOmJCk9OMZ16f6x2d1KPE6UXq10Mg7BBu9DibjKxGSZl2lx9INrL0KL8wNZEztbnBoR7azpv8Q/640?wx_fmt=jpeg&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1&amp;wx_co=1" style="width: 50%; margin-bottom: 20px;">
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">-</span>&nbsp;John Burningham<span style="color: black;">&nbsp;-</span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">这是由<strong style="color: blue;">人工智能程序Deep-speare</strong>创作的十四行诗:</p><img src="https://mmbiz.qpic.cn/mmbiz_jpg/SE4iaQdbe1bXGcibOmJCk9OMZ16f6x2d1KvvcYkyA5B0Q9G77cCJicAPicUoJtXQIrcBgEDFekdpf7PemWVW9Cj8hw/640?wx_fmt=jpeg&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1&amp;wx_co=1" style="width: 50%; margin-bottom: 20px;">
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">Yet in a circle pallid as it flow,</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">by this bright sun, that with his light display,</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">rolld from the sands, and half the buds of snow,</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">and calmly on him shall infold away</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">它流淌,在苍白的旋涡中</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">路过明媚的阳光,他的光辉闪耀</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">最终</span>滚落在细沙上,雪般的花蕾含苞待放</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">轻覆其身,悄然而逝</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">仔细阅读Deep-speare写的诗,你可能会察觉到一丝荒谬,<span style="color: black;">不外</span>它倒<strong style="color: blue;">确实遵循章法</strong>,用英文老师的话<span style="color: black;">来讲</span><span style="color: black;">便是</span>,乍看之下这首诗歌<strong style="color: blue;">每一行的节奏、韵律和基本语法都还挺像<span style="color: black;">那样</span>一回事</strong>。正如该<span style="color: black;">科研</span>团队<span style="color: black;">发掘</span>的那样,当向全世界展示AI创作的诗歌时,不少人都被糊弄过去了——<strong style="color: blue;">大<span style="color: black;">都数</span>读者<span style="color: black;">没法</span>区<span style="color: black;">掰开</span>AI生成的诗和人类作者写出的作品</strong>。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">该团队由三位<span style="color: black;">设备</span>学习<span style="color: black;">科研</span>人员和一名文学学者<span style="color: black;">构成</span>,<span style="color: black;">运用</span><strong style="color: blue;">古腾堡工程在线图书馆</strong>(Project Gutenberg)收录的2700首十四行诗来训练AI诗人。该AI诗人利用被<span style="color: black;">叫作</span>为深度学习的AI<span style="color: black;">办法</span>,对训练数据库中的诗歌进行挖掘,<span style="color: black;">持续</span>地“冥思苦想”出与样本匹配的诗句,自己学会作诗的。<span style="color: black;">她们</span>并<span style="color: black;">无</span>像以往类似的项目<span style="color: black;">同样</span>,给AI诗人输入韵律、发音方面的字典等资源,然而,<strong style="color: blue;">Deep-speare靠着摸索,自己学会了十四行诗写作的三大要素:节奏、韵律,和自然语言的基本原理(即<span style="color: black;">怎样</span>将词汇流畅地衔接在<span style="color: black;">一块</span>)</strong>。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">*译者注</strong></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">古腾堡工程(Project Gutenberg),是一项志愿工作,致力于将文化作品的数字化和归档,以“鼓励创作和发行电子书”为其宗旨。该工程肇始于1971年,是最早的数字图书馆。<span style="color: black;">截止</span>2012年7月,古腾堡工程声<span style="color: black;">叫作</span>超过40,000件馆藏。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">该团队的<strong style="color: blue;"><span style="color: black;">目的</span></strong>是<strong style="color: blue;">探索深度学习在生成自然语言上能做到何种程度,以及<span style="color: black;">怎样</span>充分利用诗歌的有趣特质</strong>。尤其是像十四行诗<span style="color: black;">这般</span>的诗歌,其节奏和韵律有着刻板和规整的模式,想<span style="color: black;">晓得</span><span style="color: black;">是不是</span><span style="color: black;">能够</span><strong style="color: blue;"><span style="color: black;">经过</span>设计系统体系</strong>,让Deep-speare自主学习此类模式。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">这一切的<span style="color: black;">奋斗</span>离不开计算创造力<span style="color: black;">行业</span>的蓬勃发展</strong>。一幅由AI创作出的<strong style="color: blue;">肖像画</strong>曾在纽约佳士得(Christie’s)拍卖行以43.2万美元的价格成交;AI作曲项目DeepBach创作的巴赫风格<strong style="color: blue;">音乐作品</strong>足以让人信以为真;而在雕塑和舞蹈的<span style="color: black;">行业</span>,<span style="color: black;">亦</span>能看见AI的身影。除此之外,在<strong style="color: blue;">语言和文学</strong>方面,来自OpenAI实验室的GPT-2文本生成系统凭实力证明了<strong style="color: blue;">人工智能<span style="color: black;">能够</span>仅<span style="color: black;">经过</span>开头句,就能生产出相当流畅的文本段落</strong>。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">*译者注</strong></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">计算创造力(computatinal creativity, <span style="color: black;">亦</span><span style="color: black;">叫作</span>为人工智能创造力,<span style="color: black;">设备</span>创造力或创造力计算学)是一项跨学科的<span style="color: black;">科研</span><span style="color: black;">行业</span>,<span style="color: black;">位置于</span>人工智能,认知心理学,哲学和艺术<span style="color: black;">行业</span>的交汇处。</p><img src="https://mmbiz.qpic.cn/mmbiz_jpg/SE4iaQdbe1bXGcibOmJCk9OMZ16f6x2d1Kn6301dZOrSZRnven0vyDNUw21eScQia6kbMjDHXIOFiasJENicib2XZBXA/640?wx_fmt=jpeg&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1&amp;wx_co=1" style="width: 50%; margin-bottom: 20px;">
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">-</span>&nbsp;Fran Horne<span style="color: black;">&nbsp;-</span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">在过去的十年里,正是<strong style="color: blue;">深度学习的蓬勃发展</strong>才使得这些关于计算创造力的<strong style="color: blue;">实验<span style="color: black;">作为</span>可能</strong>。深度学习<span style="color: black;">拥有</span><strong style="color: blue;">创作追求的几个关键<span style="color: black;">优良</span></strong>:对初学者<span style="color: black;">来讲</span>,它非常<strong style="color: blue;">灵活</strong>,<span style="color: black;">况且</span><span style="color: black;">亦</span><strong style="color: blue;">能相对容易地训练出能执行<span style="color: black;">各样</span>任务的深度学习系统</strong>(<span style="color: black;">咱们</span><span style="color: black;">叫作</span>之为模型)。这些模型<span style="color: black;">善于</span><span style="color: black;">发掘</span>模式,并从中进行总结归纳——有时候<span style="color: black;">乃至</span>会产生令人惊奇的结果,而这种<span style="color: black;">优良</span><span style="color: black;">亦</span>被<span style="color: black;">叫作</span>为“<strong style="color: blue;">偶发创造力</strong>”。<span style="color: black;">同期</span>,<strong style="color: blue;">深度学习算法内在<span style="color: black;">原因</span>的随机性会<span style="color: black;">引起</span>输出结果的千变万化</strong>。这种结果上的千变万化<span style="color: black;">倘若</span>能够被人类协作者耐心地筛出,<span style="color: black;">那样</span>这种可变性将<strong style="color: blue;">非常适合创造性应用</strong>。不仅如此,深度算法还能<strong style="color: blue;">相对容易地<span style="color: black;">创立</span>可处理<span style="color: black;">区别</span>类型数据的模型</strong>,<span style="color: black;">包含</span>文本、语音、文字和视频等。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">一首诗是<span style="color: black;">怎样</span>被写出来的?</strong></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">关于自然语言处理模型</strong></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">十四行诗最<span style="color: black;">重点</span>有两个特点:14行的长度,和由两部分<span style="color: black;">构成</span>的“论证”结构。</strong>后者指的是诗人<span style="color: black;">通常</span>会在诗歌的前半部分提出一个问题,<span style="color: black;">而后</span>用结尾的部分来解答这个疑惑。在十六世纪,英国诗人采用“<strong style="color: blue;">五步抑扬格</strong>”(iambia pentameter)的节奏方式来创作诗歌,即一行诗歌<span style="color: black;">持有</span>10个音节(5个音步)的轻重音节奏。<span style="color: black;">一般</span>,一首十四行诗由三个用来提出“问题”的四行诗(quatrain)和末尾的对偶句(couplet)<span style="color: black;">形成</span>,诗歌的韵律常为「ABAB CDCD &nbsp;EFEF GG」。在莎士比亚手里,此种韵律形式<span style="color: black;">运用</span>到了最纯熟的地步,以至于今天<span style="color: black;">咱们</span>都<span style="color: black;">叫作</span>其为“莎士比亚体之十四行诗”。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">在Deep-speare项目组中,<span style="color: black;">咱们</span>试图生成莎士比亚十四行诗前半部分的“发问”段落中的一段四行诗。<span style="color: black;">因此呢</span>,<strong style="color: blue;">相比于简单复制诗歌的十四行形式<span style="color: black;">或</span>结尾的两行“论证”的诗句,<span style="color: black;">咱们</span>更专注于生成遵循“五步抑扬格”和韵律的文本</strong>。<span style="color: black;">咱们</span><span style="color: black;">亦</span>许会在<span style="color: black;">将来</span>的某天继续<span style="color: black;">解决</span>这个更难的挑战,但<span style="color: black;">日前</span><span style="color: black;">咱们</span>更<span style="color: black;">必须</span>证明的是AI诗人具备产生一段单独的四行诗的能力。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">Deep-speare采用了<strong style="color: blue;">三种自然语言处理模型</strong>进行创作,它们分别是<span style="color: black;">经过</span><span style="color: black;">评定</span>备选单词的概率<span style="color: black;">选取</span>合适单词的<strong style="color: blue;">语言模型</strong>、<span style="color: black;">评定</span>每行诗歌的节奏的<strong style="color: blue;">节奏模型</strong>,以及<span style="color: black;">保证</span>每行诗歌都遵循韵律的<strong style="color: blue;">韵律模型</strong>。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">AI诗人会<strong style="color: blue;">随机<span style="color: black;">选取</span>一种经典韵律</strong>来创作诗节。<span style="color: black;">例如</span>下图这个例子,它就采用了“ABBA”的押韵方式,即在一个四行的诗节<span style="color: black;">其中</span>,首尾两行押韵,中间两行押韵。在<span style="color: black;">把握</span>这个模版之后,它以一个令人惊讶的方式<span style="color: black;">起始</span>生成这首诗歌——从最后一行的最后一个单词<span style="color: black;">起始</span>,从右到左生成符合规律的文本。</p><img src="https://mmbiz.qpic.cn/mmbiz_jpg/SE4iaQdbe1bXGcibOmJCk9OMZ16f6x2d1KCLI8NKprIPqWiapJqVYbQ0cQGN4yf7l9HFpj0SMbsictI2DrGuxrtOCA/640?wx_fmt=jpeg&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1&amp;wx_co=1" style="width: 50%; margin-bottom: 20px;">
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;"><span style="color: black;">1. </span></strong>在上图的例子“from pardon to him, who will stand and wait”中,Deep-speare<span style="color: black;">首要</span>从句末的“wait”这个单词<span style="color: black;">起始</span>生成文本,利用<span style="color: black;">以上</span><span style="color: black;">说到</span>的概率处理模型,<span style="color: black;">针对</span>词语的<span style="color: black;">选取</span>,它先<span style="color: black;">思虑</span>了所有英文词汇出<span style="color: black;">此刻</span>该位置的可能性,选出其中排名前五的候选单词。然而,<strong style="color: blue;">模型并不会直接<span style="color: black;">运用</span>概率最大的单词,而是从五个单词里面随机<span style="color: black;">选取</span>一个</strong>。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;"><span style="color: black;">2. </span></strong>以退为进,AI诗人<strong style="color: blue;"><span style="color: black;">经过</span>对<span style="color: black;">每一个</span>词倒推来进行文本生成</strong>,而每一个候选单词的概率得分<span style="color: black;">表示</span>了单词之间相邻关系和出<span style="color: black;">此刻</span>同一句子中的概率(如下图所示)。</p><img src="https://mmbiz.qpic.cn/mmbiz_jpg/SE4iaQdbe1bXGcibOmJCk9OMZ16f6x2d1K67bCxdB7rFDqYWRn40WtykXCUe1dNO6vjUzp1ibcbWfTC6EFH2Yo3wA/640?wx_fmt=jpeg&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1&amp;wx_co=1" style="width: 50%; margin-bottom: 20px;">
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;"><span style="color: black;">3.</span></strong> 利用<span style="color: black;">这般</span>的<span style="color: black;">办法</span>,AI诗人生产了许多候选诗句,接下来<strong style="color: blue;">节奏模型会给它们打分</strong>,找出符合“五步抑扬格”节奏规律的诗句。</p><img src="https://mmbiz.qpic.cn/mmbiz_jpg/SE4iaQdbe1bXGcibOmJCk9OMZ16f6x2d1KVo3MvTyexDibMTx49AnibyI2lqgbLXhzmbSGUZuybTKGCW1tncRVS98g/640?wx_fmt=jpeg&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1&amp;wx_co=1" style="width: 50%; margin-bottom: 20px;">
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">Shall I behold him in his cloudy state</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">for just but tempteth me to stay and pray</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">a cry: if it will drag me, find no way</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">怎么忍心看他愁云惨雾伤悲</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">耿耿于怀的我驻足祈祷安慰</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">只怕眼泪来袭 &nbsp; 我<span style="color: black;">没法</span>应对</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;"><span style="color: black;">4. </span></strong>以退为进、从后往前,AI诗人<strong style="color: blue;"><span style="color: black;">持续</span>重复<span style="color: black;">这般</span>的<span style="color: black;">过程</span></strong>,以从最后一句到<span style="color: black;">第1</span>句的方式,继续生成文本。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;"><span style="color: black;">5. </span></strong>在寻找合适的单词来给第二句和<span style="color: black;">第1</span>句诗结尾时(“pray” 和 “state”),AI诗人会<strong style="color: blue;">给候选单词的“押韵度”评分</strong>,找到与“way”和 “wait”押韵的辞藻。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">总的<span style="color: black;">来讲</span>,<span style="color: black;">咱们</span>的系统由<strong style="color: blue;">三个部分</strong><span style="color: black;">构成</span>:<strong style="color: blue;">一个学会了“五步抑扬格”的节奏模型,一个学会了词汇押韵的模型,以及一个学会了词与词之间的经典搭配的语言模型</strong>,其中<strong style="color: blue;">语言模型</strong>是能够逐字生成十四行诗的<strong style="color: blue;">最<span style="color: black;">重点</span>部分</strong>。</p><img src="https://mmbiz.qpic.cn/mmbiz_jpg/SE4iaQdbe1bXGcibOmJCk9OMZ16f6x2d1K6ygCic9AxXFIzIFWrTGE8FFBHr4b6RmoO1Q7fTfLkicrZBqibHg3yPOnw/640?wx_fmt=jpeg&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1&amp;wx_co=1" style="width: 50%; margin-bottom: 20px;">
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">-</span>Vera Molnar<span style="color: black;">&nbsp;-</span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">语言模型会<span style="color: black;">经过</span><strong style="color: blue;">概率评分</strong>的<span style="color: black;">办法</span>,对任意语句进行排名,判断<span style="color: black;">那些</span>语句在某种语言之中是正确的 (在<span style="color: black;">咱们</span>的例子中是英语)。<strong style="color: blue;">经过适当训练的语言模型会赋予流利的句子较高的概率分,给无<span style="color: black;">道理</span>的句子较低的概率分</strong>。<span style="color: black;">思虑</span>到语言生产和理解方式<span style="color: black;">通常</span>是逐词进行,这个原理<span style="color: black;">实质</span>上<span style="color: black;">准许</span><span style="color: black;">咱们</span>把更<span style="color: black;">繁杂</span>的、句子层面上的问题分解成单词层面的简单问题。<span style="color: black;">因此呢</span>,<strong style="color: blue;">语言模型的工作<span style="color: black;">便是</span><span style="color: black;">经过</span>部分句子预测下一个单词是什么</strong>。为了进行这种预测,模型会<span style="color: black;">查询</span>所有可能的单词并给予它们概率分数,而这些分数取决于<span style="color: black;">日前</span>句子中已有的词汇。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">语料库&nbsp;</strong></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">一般</span>,自然语言模型会<span style="color: black;">经过</span>语料库的单词和句子判断文本的<span style="color: black;">显现</span>概率,而语料库的内容<span style="color: black;">能够</span>来自维基百科词条,Reddit中的讨论,或者专门用于训练语言处理的语料库。从文字库中,人工智能<span style="color: black;">能够</span>学习<span style="color: black;">那些</span>单词是最常<span style="color: black;">一块</span><span style="color: black;">显现</span>的。而在<span style="color: black;">咱们</span>的Deep-speare项目中,AI诗人先是<strong style="color: blue;">从在线图书馆Project Gutenberg收录的<span style="color: black;">所有</span>诗歌中学会了语言运用的基本课程</strong>,<span style="color: black;">而后</span>再<strong style="color: blue;"><span style="color: black;">经过</span><span style="color: black;">包括</span>了36.7万个单词的2700首莎士比亚十四行诗来进一步打磨它的十四行诗写作能力。</strong></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">人们对下一个单词的“惊奇”程度<span style="color: black;">能够</span>被用于衡量一个语言模型的质量。<span style="color: black;">倘若</span>下一个单词被赋予了很高的概率分数,这说明这个单词的<span style="color: black;">显现</span>比较符合规律,<span style="color: black;">因此</span>并不会让人感到惊讶;但<span style="color: black;">倘若</span>被赋予了很低的概率分数,<span style="color: black;">那样</span>人们在读到这个单词的时候,会感到非常不自然。<strong style="color: blue;">这种惊奇程度在训练模型的过程中被当成重要的信号。</strong><span style="color: black;">因此呢</span>,当<span style="color: black;">咱们</span><strong style="color: blue;">每次都<span style="color: black;">经过</span><span style="color: black;">海量</span>文本来处理每一个词汇,<span style="color: black;">况且</span>模型不会对连在<span style="color: black;">一块</span>的词汇感到惊讶</strong>,<span style="color: black;">那样</span><span style="color: black;">咱们</span>便<span style="color: black;">能够</span>认为,这个模型<span style="color: black;">已然</span>很大程度上<span style="color: black;">把握</span>了语言的<span style="color: black;">繁杂</span>性了。而这就<span style="color: black;">包含</span>能够正确运用“San ­Francisco”<span style="color: black;">这般</span>由多单词<span style="color: black;">形成</span>的词组而不拆开它们、遵循<span style="color: black;">必定</span>句式和语法结构,以及判断较为<span style="color: black;">繁杂</span>的语义和<span style="color: black;">规律</span>信息(例如,咖啡常常用“浓”和“淡”,而不会用“强壮”和“轻便”来修饰)等原则。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">一旦语言模型被训练好了,无中生有地生成单句或多条诗句就<span style="color: black;">再也不</span>是难事了</strong>。</p><img src="https://mmbiz.qpic.cn/mmbiz_jpg/SE4iaQdbe1bXGcibOmJCk9OMZ16f6x2d1Kazibo2ssx674akCial5qoOSN8ibHarTAroz3EAKqwZqfLHKAM7ic1tU9hA/640?wx_fmt=jpeg&amp;tp=webp&amp;wxfrom=5&amp;wx_lazy=1&amp;wx_co=1" style="width: 50%; margin-bottom: 20px;">
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">-</span>&nbsp;Tom Phillips<span style="color: black;">&nbsp;-</span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">接下来,<span style="color: black;">咱们</span>让<strong style="color: blue;">节奏模型</strong>遵守每行诗歌要有10个以轻音重音的模式结合音节的规律,<span style="color: black;">而后</span>,它会<strong style="color: blue;"><span style="color: black;">检测</span></strong><span style="color: black;">每一个</span>单词的字母和句子的标点符号,<span style="color: black;">经过</span>字母决定音节该<span style="color: black;">怎样</span>分配,哪一个音节要归类为重音。<span style="color: black;">例如</span>“summer”一词有两个音节,“sum”是重音,“mer”是轻音,当Deep-speare在写四行诗时,<strong style="color: blue;">语言模型会生成候选诗句,节奏模型会从中选出符合“五步抑扬格”的一句,<span style="color: black;">而后</span>再重复这一过程,生成下一行诗句</strong>。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">韵律模型</strong><span style="color: black;">亦</span>是从过去的十四行诗集中含英咀华学会的,但它只会看<strong style="color: blue;">每行最后一个单词的字母</strong>。在训练过程中,<span style="color: black;">咱们</span>告诉模型,在四行诗中的每句诗的最后一个词都<span style="color: black;">必须</span><strong style="color: blue;">押韵</strong>,<span style="color: black;">而后</span><span style="color: black;">咱们</span>让它找出<span style="color: black;">哪些</span>词中最<span style="color: black;">类似</span>的词,越是<span style="color: black;">类似</span>的单词,就越有可能押韵。以莎士比亚的诗歌为例子,“day”和“May”的押韵分数很高,“temperate”和“date”<span style="color: black;">亦</span>是如此。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">一旦Deep-speare被训练好并准备创作,<span style="color: black;">咱们</span>会给它<strong style="color: blue;"><span style="color: black;">供给</span>莎士比亚十四行诗中三个最经典的韵律模版来从中<span style="color: black;">选取</span>:AABB, ABBA, ABAB</strong>。在写作过程中,语言模型会先随机<span style="color: black;">选择</span>其中一个模板,再逐词地产生诗句,当写到了应该押韵的单词时,它会为押韵模型<span style="color: black;">供给</span>多个候选单词。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">下面有<strong style="color: blue;">两个例子</strong><span style="color: black;">能够</span>很好地解释Deep-speare生产文本的过程。<strong style="color: blue;"><span style="color: black;">第1</span>篇文本<span style="color: black;">是由于</span>稍微训练过的模型所创作的</strong>,它初步<span style="color: black;">把握</span>了韵律,但尚未找到节奏,且诗歌读起来并不是很通顺。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">by complex grief’s petty nurse. had wise upon</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">along</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">came all me’s beauty, except a nymph of song</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">to be in the prospect, he th of forms i join</p>and long in the hears and must can god to run
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">由<span style="color: black;">繁杂</span>的<span style="color: black;">哀痛</span>的<span style="color: black;">漂亮</span>小护士</p>

    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">来吧我所有的<span style="color: black;">漂亮</span>,除了一首歌</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">在光明的<span style="color: black;">将来</span>里,我加入的形式</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">长期</span>聆听,<span style="color: black;">必要</span>让上帝奔跑</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">第二篇文本则<span style="color: black;">是由于</span>训练臻于完善的模型生成的</strong>,相较<span style="color: black;">第1</span>篇<span style="color: black;">能够</span>看出,它取得了很大的进步。它ABBA的韵律是正确的,遵循了“五步抑扬格”,语言不仅流畅,还颇有诗意!</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">shall i behold him in his cloudy state</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">for just but tempteth me to stop and pray</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">a cry: if it will drag me, find no way</p>from pardon to him, who will stand and wait
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">怎么忍心看他愁云惨雾伤悲</p>

    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">耿耿于怀的我驻足祈祷安慰</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">只怕眼泪,来袭我<span style="color: black;">没法</span>应对</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">谁得宽恕,谁又将驻足等待</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">AI诗人写的诗到底好<span style="color: black;">欠好</span>?</strong></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">在检验Deep-speare的输出结果时,<span style="color: black;">咱们</span><strong style="color: blue;"><span style="color: black;">首要</span>要<span style="color: black;">保证</span>它<span style="color: black;">无</span>复制语料库里的原句</strong>。<span style="color: black;">咱们</span><span style="color: black;">发掘</span>,它作品中的诗句并未和训练数据有很大重叠,<span style="color: black;">因此呢</span>,<span style="color: black;">咱们</span>相信AI诗人不是单靠复制粘贴产生作品的,<strong style="color: blue;">它的诗歌<span style="color: black;">能够</span>说都是原创的</strong>。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">不外</span> ,原创并不是质量好的代名词,为了检验作品质量,<span style="color: black;">咱们</span>请了<strong style="color: blue;">人类评委</strong>进行分析,<span style="color: black;">她们</span>来自两个<span style="color: black;">区别</span>的背景。<strong style="color: blue;"><span style="color: black;">第1</span>组评委是亚马逊Mechanical Turk平台雇佣的众包工人</strong>,<span style="color: black;">她们</span>只会<span style="color: black;">基本</span>的英语,<span style="color: black;">无</span>诗歌方面的专业知识。<span style="color: black;">咱们</span>向<span style="color: black;">她们</span><span style="color: black;">同期</span>展示AI诗人和人类诗人写的十四行诗,并让<span style="color: black;">她们</span>指出哪一首是人类写的。</p><img src="data:image/svg+xml,%3C%3Fxml version=1.0 encoding=UTF-8%3F%3E%3Csvg width=1px height=1px viewBox=0 0 1 1 version=1.1 xmlns=http://www.w3.org/2000/svg xmlns:xlink=http://www.w3.org/1999/xlink%3E%3Ctitle%3E%3C/title%3E%3Cg stroke=none stroke-width=1 fill=none fill-rule=evenodd fill-opacity=0%3E%3Cg transform=translate(-249.000000, -126.000000) fill=%23FFFFFF%3E%3Crect x=249 y=126 width=1 height=1%3E%3C/rect%3E%3C/g%3E%3C/g%3E%3C/svg%3E" style="width: 50%; margin-bottom: 20px;">
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">-&nbsp;</span>Camilo Huinca&nbsp;<span style="color: black;">-</span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">但<span style="color: black;">第1</span>次的检验结果令<span style="color: black;">咱们</span>感到大失所望,<span style="color: black;">由于</span>工<strong style="color: blue;">人们<span style="color: black;">能够</span>近乎完美指出哪首是人类诗歌</strong>!看上去<span style="color: black;">设备</span>学习的结果显然不符合人们的鉴赏标准。那这么一来,<span style="color: black;">咱们</span>AI诗人的<span style="color: black;">科研</span>之路是不是就到此为止了?</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">接着,<span style="color: black;">咱们</span>思考了这近乎百分百识别率<span style="color: black;">背面</span>的<span style="color: black;">原由</span>或许是——<span style="color: black;">第1</span>组评委<strong style="color: blue;">作<span style="color: black;">坏处</span></strong>了。<span style="color: black;">咱们</span>的语料库来自古腾堡工程在线图书馆,因而文本都是<span style="color: black;">能够</span>被搜到的,于是<span style="color: black;">咱们</span>在想,会不会是<span style="color: black;">她们</span>复制了待判断的诗歌,再到网上去搜是谁写的?抱着这种猜测,<span style="color: black;">咱们</span>的<span style="color: black;">科研</span>人员<span style="color: black;">亦</span>依葫芦画瓢,测试了一番,事实证明<span style="color: black;">咱们</span>是对的——人类诗歌作品总是<span style="color: black;">能够</span>搜到<span style="color: black;">有些</span>结果,达到百分百鉴别率是轻而易举的事情。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">为了防止评委们耍小技巧作<span style="color: black;">坏处</span>,<span style="color: black;">咱们</span>把所有的待鉴别诗歌都转换为图像,再让<span style="color: black;">她们</span>指出哪首是人类诗歌作品。看!<strong style="color: blue;">这次<span style="color: black;">她们</span>的正确率从近乎百分百下降到了百分之五十</strong>,这说明<span style="color: black;">她们</span><span style="color: black;">没法</span>准确地区分人类和<span style="color: black;">设备</span>人的诗歌作品。尽管<span style="color: black;">咱们</span>还是<span style="color: black;">没法</span>阻止所有人不去手动输入诗歌再谷歌一下结果,但手动<span style="color: black;">查询</span>确实<span style="color: black;">必须</span>花费不少时间。总的<span style="color: black;">来讲</span>,这次正确率的下降说明了<strong style="color: blue;">AI诗人的作品在某种程度上,确实<span style="color: black;">能够</span>以假乱真</strong>。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">咱们</span>的<strong style="color: blue;">第二位评委是多伦多大学文学助理教授亚当·哈蒙德</strong>(Adam Hammond)。与<span style="color: black;">第1</span>组评委的测试过程<span style="color: black;">区别</span>,第二次质量检验<span style="color: black;">再也不</span>是猜测游戏。相反,哈蒙德将会收到混合着人工和<span style="color: black;">设备</span>创作的诗歌,并<strong style="color: blue;">从韵律、节奏、可读性和情感影响力这四个维度对它们进行打分</strong>。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">他给了Deep-speare十四行诗很高的韵律分和节奏分,<span style="color: black;">实质</span>上,<strong style="color: blue;">与人类写作的十四行诗相比,它们在韵律和节奏上的评分更高</strong>。对这个结果,哈蒙德<span style="color: black;">亦</span>并不感到惊讶,他认为人类诗人经常打破规律来取得<span style="color: black;">有些</span>诗歌效果。反倒是在可读性和情感影响力上,AI诗人则<span style="color: black;">显著</span>逊色了不少,文学专家<span style="color: black;">能够</span>轻易<span style="color: black;">经过</span>这两个方面指出哪首是<span style="color: black;">设备</span>的创作。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">AI写诗的能力是幻想?</strong></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">Deep-speare项目最有趣的<span style="color: black;">地区</span>之一是它<span style="color: black;">导致</span>的轰动。<span style="color: black;">咱们</span>在2018年计算语言学学术会议上做完报告之后,世界各地的<span style="color: black;">资讯</span><span style="color: black;">媒介</span><span style="color: black;">报告</span>了这个成果。<span style="color: black;">非常多</span><span style="color: black;">文案</span>都引用了以下的诗节,以证明AI诗人Deep-speare能够创造像人类创造的诗歌:</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">With joyous gambols gay and still array,</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">no longer when he ’twas, while in his day</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">at first to pass in all delightful ways</p>around him, charming, and of all his days.
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">当哈蒙德在BBC广播电台接受采访的时候,主持人朗读了这个段落并问他的看法,哈蒙德则反问主持人<span style="color: black;">是不是</span>有<span style="color: black;">重视</span>到诗歌中<span style="color: black;">显著</span>的语法错误“he ’twas”来<span style="color: black;">做为</span>“he it was”(不成立的表达)的缩写,主持人<span style="color: black;">暗示</span>她并没<span style="color: black;">重视</span>到。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">社会<span style="color: black;">专家</span>雪莉·特克尔(Sherry Turkle)把人们<span style="color: black;">忽略</span>人工智能的<span style="color: black;">显著</span>错误,却依然感慨其成就的现象<span style="color: black;">叫作</span>为“伊莉莎效应”(the Eliza effect)</strong>。换句话说,人们<span style="color: black;">能够</span>过度<span style="color: black;">诠释</span><span style="color: black;">设备</span>产生的结果,<span style="color: black;">乃至</span>读出来<span style="color: black;">本来</span>不存在的意义。这个现象最早<span style="color: black;">能够</span>追溯到十九世纪六十年代,在麻省理工大学,计算机<span style="color: black;">专家</span>约瑟夫·维森鲍姆(Joseph Weizenbaum)<span style="color: black;">研发</span>了<span style="color: black;">第1</span>个聊天<span style="color: black;">设备</span>人伊莉莎(Eliza),它会模仿心理治疗师的说话方式。尽管这个程序相当粗糙,还有很大的局限性,但维森鲍姆惊讶地<span style="color: black;">发掘</span>用户轻易地就被<span style="color: black;">设备</span>人“<span style="color: black;">诈骗</span>”了。在七十年代,<span style="color: black;">做为</span>维森鲍姆同事的特克尔<span style="color: black;">发掘</span>,<span style="color: black;">乃至</span><span style="color: black;">哪些</span>明知伊莉莎程序有缺陷的<span style="color: black;">科研</span>生<span style="color: black;">亦</span>会向<span style="color: black;">设备</span>抛出问题,并期待它以近似正<span style="color: black;">一般人</span>类的方式去回答。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">特克尔把伊莉莎效应<span style="color: black;">叫作</span>为“<strong style="color: blue;">数字幻想中的人类同谋</strong>”,而这看起来<span style="color: black;">亦</span><span style="color: black;">能够</span>解答人们对Deep-speare诗歌作品的赞叹反应。<strong style="color: blue;">公众太<span style="color: black;">期盼</span>这些十四行诗<span style="color: black;">能够</span>用于证明人工智能的力量,以至于忽略了<span style="color: black;">哪些</span>与之相悖的证据</strong>。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">这种蓄意的误解可能会对AI诗人真正的能力发展<span style="color: black;">导致</span>阻碍。<span style="color: black;">咱们</span>还在<span style="color: black;">连续</span>进行AI诗人的项目,<strong style="color: blue;"><span style="color: black;">目的</span>之一<span style="color: black;">便是</span><span style="color: black;">提高</span><span style="color: black;">咱们</span>AI诗人作品的可读性和情感影响力</strong>。而<span style="color: black;">想要</span><span style="color: black;">提高</span>整体的流畅度,<strong style="color: blue;">其中一个策略是利用大范围的语料库(例如<span style="color: black;">全部</span>维基百科)去“预训练”语言模型</strong>,让它<span style="color: black;">能够</span>在一段较长的叙述中更好<span style="color: black;">把握</span>词汇<span style="color: black;">显现</span>的规律。在这个<span style="color: black;">基本</span>上,<span style="color: black;">咱们</span>再进一步对它进行十四行诗文本生成的训练。</p><img src="data:image/svg+xml,%3C%3Fxml version=1.0 encoding=UTF-8%3F%3E%3Csvg width=1px height=1px viewBox=0 0 1 1 version=1.1 xmlns=http://www.w3.org/2000/svg xmlns:xlink=http://www.w3.org/1999/xlink%3E%3Ctitle%3E%3C/title%3E%3Cg stroke=none stroke-width=1 fill=none fill-rule=evenodd fill-opacity=0%3E%3Cg transform=translate(-249.000000, -126.000000) fill=%23FFFFFF%3E%3Crect x=249 y=126 width=1 height=1%3E%3C/rect%3E%3C/g%3E%3C/g%3E%3C/svg%3E" style="width: 50%; margin-bottom: 20px;">
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">-</span>&nbsp;Tomasz Woźniakowski<span style="color: black;">&nbsp;-</span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;"><span style="color: black;">咱们</span><span style="color: black;">一样</span><span style="color: black;">亦</span>在思考,人类诗人是<span style="color: black;">怎样</span>创作诗歌的</strong>:<span style="color: black;">她们</span>总不会在桌子旁正襟危坐,思考着“我<span style="color: black;">第1</span>个单词应该是什么”,<span style="color: black;">而后</span>冥思苦想,思考下一个单词该接什么好。相反,<span style="color: black;">她们</span>都是心中先构思出一个主题<span style="color: black;">或</span>一段故事,再<span style="color: black;">经过</span>单词和句子表达所思所想。其实在Deep-speare项目中,<span style="color: black;">咱们</span><span style="color: black;">已然</span>做到了让AI诗人<span style="color: black;">按照</span>给定的主题来生成<span style="color: black;">关联</span>的文本,<span style="color: black;">例如</span>让它创作主题是“爱”或“失去”的诗句。让<span style="color: black;">设备</span><span style="color: black;">根据</span><span style="color: black;">必定</span>的<strong style="color: blue;">主题</strong>,<strong style="color: blue;"><span style="color: black;">亦</span>许会<span style="color: black;">提高</span>诗歌的流畅度,但与此<span style="color: black;">同期</span>,它的词汇<span style="color: black;">选取</span>将不会<span style="color: black;">那样</span>丰富</strong>,<span style="color: black;">由于</span>它总要先学习<span style="color: black;">那些</span>词汇符合某些主题。<strong style="color: blue;">之后,<span style="color: black;">咱们</span>计划<span style="color: black;">运用</span><span style="color: black;">更加多</span>分层的语言模型进行实验,先为诗歌生成高级叙事,再在这个框架下生成每一个词汇</strong>——<strong style="color: blue;">就如人类思考的路径<span style="color: black;">同样</span></strong>。</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">确实,这是一个宏伟的<span style="color: black;">目的</span>,但<span style="color: black;">咱们</span><span style="color: black;">期盼</span>Deep-speare项目<span style="color: black;">能够</span>在将来达到<span style="color: black;">这般</span>的标准,就算<span style="color: black;">不可</span>变成真正的AI莎士比亚,那<span style="color: black;">亦</span>要尽力<span style="color: black;">作为</span>莎士比亚在《情女怨》曾描述过的样子:</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">He had the dialect and different skill,</p>Catching all passions in his craft of will.
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">“他有丰富的词汇和无数技巧,</p>

    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">随心所欲让所有人为之倾倒。”</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">作者:Jey Han Lau, Trevor Cohn, Timothy Baldwin and Adam Hammond<span style="color: black;">&nbsp;|&nbsp;</span>封面:Karolis Strautniekas</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">译者:Yingying<span style="color: black;"> | </span>审校:里昂</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">排版:济一</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">原文:</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">https://spectrum.ieee.org/artificial-intelligence/machine-learning/this-ai-poet-mastered-rhythm-rhyme-and-natural-language-to-write-like-shakespeare</p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;"><span style="color: black;">源自</span>:神经现实</strong></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><strong style="color: blue;">编辑:小林绿子</strong></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">近期热门<span style="color: black;">文案</span>Top10</span></p><span style="color: black;">↓ 点击标题<span style="color: black;">就可</span>查看 ↓</span>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">1. </span><a style="color: black;"><span style="color: black;">套娃吗?你先看这个岛中湖中岛中湖中岛</span></a></p><span style="color: black;">
      <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">2. </span><a style="color: black;"><span style="color: black;">都靠这位天才<span style="color: black;">专家</span>20岁时的论文,你<span style="color: black;">才可</span>用手机拍照发<span style="color: black;">伴侣</span>圈</span></a></p>
      <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">3. </span><a style="color: black;"><span style="color: black;">朝天空开枪,子弹掉下来还有杀伤力吗?| No.206</span></a></p>
      <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">4. </span><a style="color: black;"><span style="color: black;">乐高还能悬浮在半空中?上百万人已看懵!</span></a></p>
      <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">5. </span><a style="color: black;"><span style="color: black;">古装片的射箭动作把<span style="color: black;">理学</span>学家看笑了,导演咱能<span style="color: black;">不可</span>专业点?</span></a></p>
      <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">6. </span><a style="color: black;"><span style="color: black;">即使被它淹没<span style="color: black;">亦</span>不会窒息,这是什么神奇液体?</span></a></p>
      <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">7.&nbsp;</span><a style="color: black;"><span style="color: black;">数学课上捡了个橡皮,勾股定理就看不懂了</span></a></p>
      <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">8. </span><a style="color: black;"><span style="color: black;"><span style="color: black;">理学</span>学写给你的情书</span></a></p>
      <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">9. </span><a style="color: black;"><span style="color: black;"><span style="color: black;">独一</span>两次<span style="color: black;">得到</span>诺贝<span style="color: black;">理学</span>学奖的人,</span></a>你却不<span style="color: black;">必定</span>认识他</p>
      <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">10. </span><a style="color: black;"><span style="color: black;">妈妈问我的桌子<span style="color: black;">为何</span>这么乱!</span></a></p>
    </span><a style="color: black;"><span style="color: black;">点此查看以往<span style="color: black;">所有</span>热门<span style="color: black;">文案</span></span></a>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><img src="data:image/svg+xml,%3C%3Fxml version=1.0 encoding=UTF-8%3F%3E%3Csvg width=1px height=1px viewBox=0 0 1 1 version=1.1 xmlns=http://www.w3.org/2000/svg xmlns:xlink=http://www.w3.org/1999/xlink%3E%3Ctitle%3E%3C/title%3E%3Cg stroke=none stroke-width=1 fill=none fill-rule=evenodd fill-opacity=0%3E%3Cg transform=translate(-249.000000, -126.000000) fill=%23FFFFFF%3E%3Crect x=249 y=126 width=1 height=1%3E%3C/rect%3E%3C/g%3E%3C/g%3E%3C/svg%3E" style="width: 50%; margin-bottom: 20px;"></p>




星☆雨 发表于 2024-8-22 13:43:31

可以发布外链的网站 http://www.fok120.com/

流星的美 发表于 2024-8-28 19:06:59

我完全同意你的看法,期待我们能深入探讨这个问题。

ejn2tr4200 发表于 2024-8-30 02:32:07

我完全同意你的观点,说得太对了。

情迷布拉格 发表于 2024-9-3 10:59:48

你的话深深触动了我,仿佛说出了我心里的声音。

听听海 发表于 2024-9-5 13:20:39

楼主发的这篇帖子,我觉得非常有道理。

流星的美 发表于 2024-9-8 23:25:07

交流如星光璀璨,点亮思想夜空。

wrjc1hod 发表于 2024-10-1 14:42:44

楼主的文章非常有意义,提升了我的知识水平。

wrjc1hod 发表于 2024-10-15 18:34:34

楼主继续加油啊!外链论坛加油!

4zhvml8 发表于 2024-10-30 04:33:49

论坛外链网http://www.fok120.com/
页: [1] 2
查看完整版本: AI诗人的十四行诗,能赛过莎翁的吗?