詞匯的含義如何在大腦中編碼?要回答這一問題似乎需要測(cè)量大腦對(duì)字典中的每個(gè)詞匯的響應(yīng)過程,。如今,,美國(guó)科學(xué)家設(shè)計(jì)出了一個(gè)計(jì)算機(jī)模型,能夠預(yù)測(cè)有關(guān)理解單個(gè)名詞的大腦活動(dòng)模式,。
美國(guó)賓夕法尼亞州匹茲堡市卡內(nèi)基梅隆大學(xué)的Tom M. Mitchell和同事,,首先確定了來自12個(gè)語(yǔ)義類別——包括動(dòng)物、服裝,、工具和交通工具——的60個(gè)具體名詞的語(yǔ)義構(gòu)成,。在一個(gè)總計(jì)達(dá)10000億個(gè)詞匯的文本數(shù)據(jù)庫(kù)中,他們?cè)u(píng)估了這些名詞出現(xiàn)在25個(gè)動(dòng)詞鄰近的頻率,,這些動(dòng)詞不是指示發(fā)動(dòng)機(jī)便是指感官運(yùn)動(dòng),。基于這些動(dòng)詞與名詞聯(lián)合出現(xiàn)的頻率,,每個(gè)動(dòng)詞被賦予了針對(duì)每個(gè)名詞的頻率值,。依據(jù)這種方式,計(jì)算機(jī)創(chuàng)建了“語(yǔ)義識(shí)別標(biāo)志”,,能夠捕捉每個(gè)名詞的含義,。研究人員同時(shí)獲得了來自9名志愿者的有關(guān)60個(gè)名詞的神經(jīng)活動(dòng)模式——這些志愿者被暴露于一部功能性核磁共振(fMRI)掃描儀下,并被要求思考每一個(gè)名詞,。
研究人員隨后研制了一種計(jì)算模式,,能夠掌握與每個(gè)名詞相關(guān)的神經(jīng)和語(yǔ)義特征。這一模型被用來接受58個(gè)名詞的訓(xùn)練,,隨后又用它來根據(jù)剩下兩個(gè)名詞的語(yǔ)義信號(hào)來預(yù)測(cè)它們的神經(jīng)信號(hào),。預(yù)測(cè)得到的神經(jīng)活動(dòng)模式與志愿者實(shí)際針對(duì)這兩個(gè)名詞的fMRI掃描結(jié)果的匹配準(zhǔn)確率達(dá)到了77%,。研究人員隨后再次利用59個(gè)名詞對(duì)這一模型進(jìn)行了訓(xùn)練,并讓其提供與剩下的1個(gè)名詞有關(guān)的fMRI圖像——最終,,這一模型在從一個(gè)有1001個(gè)詞匯組成的詞匯表中找出正確的名詞時(shí),,其準(zhǔn)確率達(dá)到了72%,。
有趣的是,,當(dāng)研究人員評(píng)估與模型有關(guān)的25個(gè)動(dòng)詞的神經(jīng)活動(dòng)時(shí),他們發(fā)現(xiàn),,大腦中與動(dòng)詞“吃”有關(guān)的區(qū)域包括味覺皮層,,而與動(dòng)詞“跑”有關(guān)的區(qū)域包括部分顳葉(與對(duì)生物學(xué)運(yùn)動(dòng)的理解有關(guān))。這意味著對(duì)于能夠編碼一個(gè)動(dòng)詞含義的大腦區(qū)域而言,,當(dāng)一個(gè)人完成了具有動(dòng)詞意義的動(dòng)作時(shí),,這些區(qū)域便會(huì)被激活。研究人員在最近出版的美國(guó)《科學(xué)》雜志上報(bào)告了這一研究成果,。
這一模型使得神經(jīng)科學(xué)家在沒有進(jìn)行上千次的大腦掃描的情況下,,能夠評(píng)估任何具體的名詞是如何在大腦中被編碼的。將名詞的語(yǔ)義特征擴(kuò)展到形容詞將更進(jìn)一步地改進(jìn)這一模型,,特別是其在相同種類的名字中區(qū)分神經(jīng)特征的能力(但這一模型并不善于完成這項(xiàng)任務(wù)),。精煉這一模型同時(shí)使得它能夠預(yù)測(cè)更多抽象概念的神經(jīng)表達(dá)。無(wú)論如何,,僅僅利用25個(gè)動(dòng)詞在語(yǔ)義上定義名詞,,這一模型描述了非常準(zhǔn)確的神經(jīng)特征,表明具體名詞的神經(jīng)表達(dá)至少部分基于事物的感覺和運(yùn)動(dòng)特征,。(生物谷Bioon.com)
生物谷推薦原始出處:
Science,,DOI: 10.1126/science.1152876,Tom M. Mitchell,,Marcel Adam Just
Predicting Human Brain Activity Associated with the Meanings of Nouns
Tom M. Mitchell,1* Svetlana V. Shinkareva,2 Andrew Carlson,1 Kai-Min Chang,3,4 Vicente L. Malave,5 Robert A. Mason,3 Marcel Adam Just3
The question of how the human brain represents conceptual knowledge has been debated in many scientific fields. Brain imaging studies have shown that different spatial patterns of neural activation are associated with thinking about different semantic categories of pictures and words (for example, tools, buildings, and animals). We present a computational model that predicts the functional magnetic resonance imaging (fMRI) neural activation associated with words for which fMRI data are not yet available. This model is trained with a combination of data from a trillion-word text corpus and observed fMRI data associated with viewing several dozen concrete nouns. Once trained, the model predicts fMRI activation for thousands of other concrete nouns in the text corpus, with highly significant accuracies over the 60 nouns for which we currently have fMRI data.
1 Machine Learning Department, School of Computer Science, Carnegie Mellon University, Pittsburgh, PA 15213, USA.
2 Department of Psychology, University of South Carolina, Columbia, SC 29208, USA.
3 Center for Cognitive Brain Imaging, Carnegie Mellon University, Pittsburgh, PA 15213, USA.
4 Language Technologies Institute, School of Computer Science, Carnegie Mellon University, Pittsburgh, PA 15213, USA.
5 Cognitive Science Department, University of California, San Diego, La Jolla, CA 92093, USA.