1、Al Predicts Pig Emotions From Grunts and Squeals人工智能可以通過咕嚕聲和尖叫聲來預(yù)測(cè)豬的情緒Researchers are using artificial intelligence (Al) as a non-invasive tool in efforts to gain insights into the emotions of animals based on the sounds they make. In a new study published in Scientific Reports this week, a global t

2、eam of researchers created an Al machine learning algorithm to help decode the emotional state of pigs based on their vocalizations.研究人員正在使用人工智能(Al)作為一種非侵入性工具，試圖根據(jù)動(dòng)物發(fā) 出的聲音來洞察它們的情緒。在本周發(fā)表在科學(xué)報(bào)告(Scientific Reports)上 的一項(xiàng)新研究中，一個(gè)全球研究團(tuán)隊(duì)創(chuàng)立了一種人工智能機(jī)器學(xué)習(xí)算法，以根據(jù) 豬的發(fā)聲來幫助解碼它們的情緒狀態(tài)。u Vocal expression of emotions has

3、been observed across species and could provide a non-invasive and reliable means to assess animal emotions/ wrote the study authors affiliated with Harvard University in the United States and other research institutions based in Switzerland, Denmark, the Czech Republic, Germany, Italy, Norway, and F

4、rance.美國(guó)哈佛大學(xué)(Harvard University)和瑞士、丹麥、捷克共和國(guó)、德國(guó)、意 大利、挪威和法國(guó)其他研究機(jī)構(gòu)的研究作者寫道：“在不同物種中都觀察到了情 緒的聲音表達(dá)，這可以為評(píng)估動(dòng)物情緒提供一種非侵入性和可靠的方法?！盇ccording to the American Psychological Association (APA), emotions are complex reaction patterns that include behavioral, physiological, and experiential elements. The APA distingu

5、ishes between emotions and feelings, which are not interchangeable terms. Feelings are self-contained phenomenal experiences/ whereas emotions typically involves feeling but differs from feeling in having an overt or implicit engagement with the world/根據(jù)美國(guó)心理協(xié)會(huì)(American Psychological Association)的說法，

6、情緒是復(fù)雜 的反響模式，包括行為、生理和經(jīng)驗(yàn)因素。美國(guó)心理協(xié)會(huì)區(qū)分了情緒和感覺，它 們是不可互換的術(shù)語。感覺是“獨(dú)立的現(xiàn)象體驗(yàn)”，而情感“通常涉及感覺，但 與感覺不同的是，它與世界有著公開或隱性的接觸”?！?Due to the impact of emotions on vocalization, the analysis of vocal expression of emotions is increasingly being considered as an important non-invasive tool to assess the affective aspects of an

7、imal welfare/ the researchers wrote.研究人員寫道：“由于情緒對(duì)發(fā)聲的影響，對(duì)情緒的聲音表達(dá)的分析越來越 被認(rèn)為是評(píng)估動(dòng)物福利情感方面的重要非侵入性工具?！盕or this study, the researchers analyzed a dataset of commercial pig vocalizations with over 7,400 calls produced by over 400 pigs in 19 context categories from “birth to slaughter/ The pig vocalizations i

8、ncluded high-frequency calls that are common in negative contexts such as screams and squeals and low-frequency calls such as grunts that often occur in neutral or positive situations.在這項(xiàng)研究中，研究人員分析了一個(gè)商業(yè)豬叫聲數(shù)據(jù)集，共有900頭豬在19 個(gè)情境類別一一從“出生到屠宰”一一中發(fā)出7400多個(gè)叫聲。豬的叫聲包括在 消極環(huán)境中常見的高頻叫聲，如大叫和尖叫，以及在中性或積極環(huán)境中常見的低 頻叫聲，如咕嚕

9、聲。The positive context categories include huddling, before and after nursing, enriched, reunion, running to and from an arena, and positive conditioning. The negative context categories are missed nursing, surprise, handling and waiting in a slaughterhouse, barren, novel object, negative conditionin

10、g, fighting, crushing, physical restrain/holding, castration without anesthetics, and isolation.積極的情境類別包括擠成一團(tuán)、護(hù)理前后、充裕、團(tuán)聚、跑進(jìn)跑出競(jìng)技場(chǎng)以 及積極的條件反射。消極的情境類別包括錯(cuò)過護(hù)理、驚訝、在屠宰場(chǎng)處理和等待、 貧瘠、新奇物體、消極的條件反射、打架、擠壓、身體約束/控制、無麻醉劑的閹 割和隔離。The researchers discovered that the pig grunts are shorter in positive situations, high-pit

11、ched squeals happen more frequently in negative situations, and low-pitched grunts are prevalent in positive and negative contexts.研究人員發(fā)現(xiàn)，豬在積極的情況下發(fā)出的咕嚕聲更短，在消極的情況下，高 音調(diào)的尖叫聲更頻繁，而低音調(diào)的咕嚕聲在積極和消極的情況下普遍存在。The scientists tested two ways of classifying the pig vocalizationsa permuted discriminant function an

12、alysis (pDFA) and an Al image classifying neural network, specifically a convolutional neural network (CNN) that uses spectrograms of the vocalization.科學(xué)家們測(cè)試了兩種對(duì)豬發(fā)聲進(jìn)行分類的方法一一排列判別函數(shù)分析 (pDFA)和人工智能圖像分類神經(jīng)網(wǎng)絡(luò)，特別是使用發(fā)聲頻譜圖的卷積神經(jīng)網(wǎng)絡(luò) (CNN)oThey found that the Al neural network classified the valence (positive or

13、negative) with 91.5 percent accuracy, outperforming the pDFA analysis. The researchers credit the Al neural networks ability to make more accurate predictions on its ability to preserve more encoded information from the spectrograms of entire vocalizations.他們發(fā)現(xiàn)，人工智能神經(jīng)網(wǎng)絡(luò)對(duì)價(jià)態(tài)(正或負(fù))的分類準(zhǔn)確率為9L5%,優(yōu) 于排列判別函數(shù)分析。研究人員相信，人工智能神經(jīng)網(wǎng)絡(luò)能夠?qū)ζ鋸?/p>