Some Skeptical Reviews


A test of the effect of reverse speech on priming.

The Journal of Psychology; 5/1/2003; Voss, Carla W.

J. R. Vokey and J. D. Read (1985) indicated that listeners cannot consciously recognize backward messages but that some information can be obtained from reversed speech. If reverse speech has a powerful influence on language processing, as D. J. Oates (1991) has claimed, then one should be able to measure a reliable priming effect from reversed messages. Sixty undergraduate students listened to short messages presented either backward or forward. Immediately after listening to each message, they responded in a lexical decision task to a visually presented word that had either been present or had not been present in the preceding message. No priming effect was found for backward messages, although there was significant priming for forward messages. The results are not consistent with an effect of reverse speech on word processing.

Key words: repetition priming, reverse speech

IN AN ARTICLE IN THE SKEPTICAL INQUIRER, Byrne and Normand (2000) critiqued claims made about the effects of reversed speech. Oates (1991) claimed that backward messages are encoded in normal speech and that this constitutes a second, covert mode of communication in addition to the overt mode of forward speech. This is a recent version of long-standing claims made in the popular media about the effects of backmasking (Vokey & Read, 1985). Oates, however, did not confine his claims to negative messages purportedly heard when popular musical recordings are played backward but articulated a startling theory suggesting that reverse speech is a normal part of human communication. Oates claimed that these backward messages in speech are not rarities but are normally encoded in everyday speech. According to Oates, backward messages can reveal unconscious memories and motives, and this information is communicated unconsciously to listeners.

Byrne and Normand (2000) pointed out that there is potential for harm in some of Oates's claims. For example, Gates (1991) suggested that expert witnesses trained in interpreting reversed messages could testify as to the unconscious motives of a defendant. If Oates's theory is not correct, this could potentially lead juries to wrong verdicts. Byrne and Normand also questioned Oates's stated goal of training individuals to open therapeutic practices. Byrne and Normand argued that psychotherapy based on Oates's theory would be unethical because the theory is so questionable.

It might be argued that Oates's (1991) claims should simply be dismissed, given that they are potentially dangerous and that the claims appear to be implausible. However, we concur with Byrne and Normand (2000) that claims such as these can and should be tested empirically.

Furthermore, the research literature does include some evidence suggesting that information could be obtained from reverse speech. A.L, a backward talker described by Cowan, Leavitt, Massaro, and Kent (1982), was able not only to reverse speech quickly but also to identify correctly words that were presented backwards. This case indicates that it is not beyond the capability of the human brain to both use and recognize backward speech.

The question of whether backward messages can be consciously identified by typical individuals has been addressed by Vokey and Read (1985), who reported a variety of results indicating that messages people hear in reversed recordings are most likely the result of constructive perception rather than of messages actually encoded in the recordings. For example, Vokey and Read reported that college students were unable to identify consciously a reversed message until after they were told what the message was. Also, students could not accurately determine whether reversed messages were questions or statements.

At least some information from backward messages is processed. Vokey and Read (1985) found that college students could easily tell whether the speaker of a backward message was a man or a woman and could tell whether two backward messages were spoken by the same speaker or different speakers. They could also discriminate among languages (English, French, German) at a higher than chance rate. Furthermore, students were slightly but significantly above chance in telling whether a word read to them was or was not present in a backward message they had just heard (55.8% accuracy). However, students could not reliably discriminate meaningful from meaningless sentences or sort sentences into semantic categories.

The claim made by Oates (1991), however, is not that reversed speech is consciously recognized but that one processes it without being aware that one is doing so. Although Gates made broad claims about intentions and memories being encoded backwards in individuals' speech, we focused our investigation not on whether backward messages actually exist in normal speech but on the question of whether these would have an effect on language processing if they did exist.

There is considerable evidence for unconscious processing of normally presented words. As Jacoby and Kelley (1992) stated, "For experimental psychologists, the investigation of unconscious influences has gained new respectability from findings of dissociations between performance on direct and indirect tests of memory and perception" (p. 175). The term unconscious here refers to processing without awareness. Although individuals with amnesia perform poorly on tests of conscious recall, prior exposure to words benefits them as much as it does ordinary individuals on indirect tests, such as reading fragmented versions of previously presented words (Jacoby & Kelley). This suggests that individuals with amnesia are able to process words without being aware that they are doing so.

It is possible that reversed messages could have an effect on memory even if individuals do not consciously recognize the words. There have been several such tests of effects of reversed messages. Vokey and Read (1985) reported that spelling of homophones was not affected by the meaning of a homophone in a previously presented backward message. Swart and Morgan (1992) presented messages in backward recordings of music and found that the backward messages (e.g., "Clean up your room") did not affect Liken-scale attitude ratings for statements such as "Keeping my bedroom clean and neat is important," even when the message was presented six times.

Begg, Needham, and Bookbinder (1993) investigated whether people would be more likely to view a statement as truthful if it was previously presented backward. Although college students could reliably distinguish backward messages they had heard from those they had not heard, their truth ratings of the messages they had heard were not different from messages they had not heard. However, students who heard the same messages played forward did rate the messages they heard as more truthful than messages they had not heard. Thus, forward messages affected truth ratings, but backward messages did not.

It might be argued that these researchers used dependent variables that were not sufficiently sensitive to find an effect of the reversed messages. Measures such as Likert-scale ratings provide a discrete set of possible responses. If backward speech does have an effect, then it may be more likely to appear in a continuous measure such as the time to choose a response rather than' in a measure of which response is chosen. Therefore, in the present study we measured repetition priming to words in a lexical decision task. We used examples of reversed messages provided by Qates (1991) so that it cannot be claimed that the stimuli were artificial or unrepresentative of the types of messages that should affect listeners. If reversed speech does affect word processing, there should be a significant amount of priming from reversed messages in the lexical decision task. As a control, we also measured priming to messages presented in their normal forward form. If we observe priming to forward messages, it would sugges t that a failure to find an effect of backward messages is not the result of a limitation of the method.



Sixty students attending Central Missouri State University volunteered to participate. They received extra credit in psychology courses in exchange for their participation.


Twenty messages that Oates (1991) has claimed exist in reversed speech samples were obtained from the Reverse Speech Enterprise's Web site ( We selected messages that appeared to have some emotional content but that did not include obscene language (see Appendix A). Of the 20 remaining messages, 12 were used for real word trials and 8 were used for pseudoword trials. Each message was digitally recorded by a woman. Reversed versions of each message were created using SoundEdit[TM] 16 software. The messages ranged in duration from 726 to 1,852 ms with a mean duration of 1,380 ms (SD = 285). For 12 of the 20 messages, a content word from the message was selected for use in the lexical decision task. For the remaining 8 messages, a pseudoword was constructed by replacing two letters in the content word for each message. We recorded four additional messages and used them for practice trials.

The number of real word trials was greater than the number of pseudoword trials as a result of the limited number of stimuli that met our criteria of being emotional but that did not include obscene language. It was necessary for the number of real word trials to be a multiple of 4 so that we could create four stimulus lists to counterbalance the items across forward and backward conditions as well as across priming and control trials. To satisfy this constraint, we used 12 of the 20 messages for real word trials, leaving 8 messages for pseudoword trials. Although participants encountered more words than nonwords, the small total number of trials made it unlikely that participants had time to develop a bias in responding across trials.

The four stimulus lists are shown in Appendix B. Each list contained 8 pseudoword trials and 12 trials with real words. Six of the real word trials on each list were priming trials in which the target word for the lexical decision task was contained in the message (priming trials). The other 6 real word trials were control trials in which the target word in the lexical decision task was a word not contained in the message (control trials). For three of the priming trials and three of the control trials, the message was presented in reversed form; for the other half of these trials, the message was presented in forward form.

The same words were used as control trials and priming trials for different participants. Words were counterbalanced across stimulus lists so that each word occurred on all four types of trials (i.e., forward and backward priming and control trials). Each word occurred once on each stimulus list. Thus, each participant encountered each message and each word only once during the course of the experiment. Forward messages were presented on half of the pseudoword trials, and reversed messages were presented on the other half.


After reading and signing an informed consent form, the participants read instructions on a computer screen and placed headphones on their heads before beginning the experiment. Stimuli were presented and responses recorded by using Reaction Time 3.0 software. Four practice trials were followed by the 20 experimental trials. On each trial, the message was presented over the headphones. Three seconds after the beginning of the message, a word or pseudoword was presented in lowercase letters in the center of the screen. Because the messages varied in length, this produced variability in the interval between the end of the message and the appearance of the word or pseudoword.

The participants responded by pressing a key marked YES if they believed the item was a real word or a key marked NO if they believed it was not a real word. We instructed the participants to respond as quickly as they could while still being accurate. There was a 2-s interval between trials during which the word READY appeared at the top of the screen. Trials were presented in random order for each participant, and 15 participants were tested for each of the four stimuli lists. Each participant was debriefed immediately after participating.


We computed priming for each participant by subtracting mean reaction time on priming trials from mean reaction time on control trials. Only times for correct responses were included. Data were excluded for three additional participants who had fewer than two out of three correct trials for any of the four groups of real word trials, yielding a total of 60 participants for all of the results reported below. The mean number of errors out of 20 trials was 0.93 (SD = 1.10).

For forward messages, the mean priming was 97 ms (SD = 201); for backward messages, the mean priming was 10 ms (SD = 359). We conducted single-sample t tests to determine whether the amount of priming was significant for forward or backward messages. Forward message priming was significantly different from zero, t(59) = 3.74, p < .001, [r.sup.2] = .19, but backward message priming was not, t(59) = 0.22, p> .05, [r.sup.2] = .00. Mean reaction times are shown in Table 1.

The number of errors was not significantly correlated with overall reaction times, r(58) = -.02, p> .05. The total number of errors was not significantly correlated with the amount of priming from backward messages, r(58) = -.14, p> .05. However, the participants who made more errors tended to show less priming from forward messages, r(58) = -.36, p < .01.


There was no indication that backward messages caused priming. The amount of backward priming was not statistically significant and the effect size was zero. It is unlikely that the lack of a backward message priming effect was because of low power or to poor sensitivity of the research method, as the same method, using the same stimuli, produced a clear priming effect with forward messages. It is also difficult to argue that the stimuli used were not realistic or emotional enough to elicit a priming effect if one existed, as we used messages that Oates (1991), a proponent of reverse speech, claimed to find in natural speech samples.

The variability of reaction times within conditions was quite large. This is likely a result of the fact that there were only three trials within each condition for each participant. Despite these large standard deviations, there was still a significant priming effect from the forward messages.

It should be noted that a repetition priming paradigm was used in the present study; the same word to which participants responded in the lexical decision task was contained in the backward or forward message. Another approach could be to use a semantic priming paradigm in which participants respond to a word that is semantically related to the word in the message (e.g., wolf and dog). We chose to use the repetition priming approach for several reasons. First, we wanted to use a method that would provide a high probability of detecting an effect of backward speech if such an effect existed. Repetition priming tends to produce large effects, and these effects have been found even when participants claim that they have not recognized the word (Stark & McClelland, 2000). This is important, because Oates (1991) claimed that we process information from reverse speech without being consciously aware that we are doing so.

The method we actually used was cross-modal repetition priming, because the message was presented in auditory form and the stimuli in the lexical decision task were presented visually. Cross-modal repetition priming (with normally presented stimuli) has been shown by Grainger, Van Kang, and Segui (2001). A second reason that we did not use a semantic priming approach is that it would have been difficult to find appropriate semantic primes that were strongly associated with only one word on the limited list of stimuli that we obtained from Oates's website. It is typically assumed that the presentation of a spoken word activates orthographic representations as well as semantic codes (Grainger et al.); thus, any orthographic or semantic processing of the backward messages should have resulted in a priming effect in the lexical decision task. It is therefore unlikely that a semantic priming study would produce any effect of backward messages.

A.L., the individual described by Cowan et al. (1982), was able to identify correctly 11 of 19 backward-spoken words, whereas controls in the Cowan et al. study correctly identified no more than 4 words. Therefore, it is at least possible that the human brain can recognize backward speech. However, the case of A.L. is extraordinary, as shown by the performance of the controls as well as by the results of Vokey and Read (1985). Oates (1991) claimed that backward speech processing is a normal part of human communication; therefore, the ability to process it should not be rare.

Swart and Morgan (1992) and Begg et al. (1993) did not find any influence of backward messages on attitude ratings or on truth judgments, respectively. However, these researchers did not test for effects on word processing in particular nor did they measure response times. Our results provide converging evidence, using a reaction time measure in a lexical decision task, that backward messages do not affect the processing of words. It is important to note that the method we used did not require participants to be aware of recognizing words in the backward messages. The claim that backward speech unconsciously (i.e., without awareness) affects language processing is therefore not consistent with our results.

Of course, it is not possible to prove that the null hypothesis is correct. However, there are now several studies, using different methods, that have not shown any effect of backward speech. Furthermore, we demonstrated the expected effect of forward messages, as did Begg et al. (1993), suggesting that the failure to find an effect of backward messages was not because of a lack of power in the research design. Oates (1991) suggested that the effects of backward speech are at least as powerful as those of normal overt speech. Our results indicate that it is very unlikely that reverse speech has an effect on language processing.



Messages and Target Words Used


                              Priming  Control

Message                        word     word    Pseudoword


Cuddle me.                    cuddle   lie

Hear the lie.                 lie      cuddle

Her name was mocked.          mocked   kill

I know who to kill.           kill     mocked

Sealed the lock.              lock     wolf

Wolf in white van.            wolf     lock

You lady mannequin.           lady     plant

This is a plant.              plant    lady

Man will space walk.          space    shot

He's shot bad.                shot     space

Evil lips are hammering it.   evil     grass

They pulled out the grass.    grass    evil

That's the jive I was given.                    libe

Make me some money.                             tiley

Here I am, rubber duck.                         ruch

Yes I'm not brave.                              grame

Get our funds.                                  tunks

You were smashed.                               spished

I'm with love.                                  lape

The fuss will serve us.                         doss




Stimulus Lists


Message                       List 1            List 2


Cuddle me.               forward priming   backward priming

Hear the lie.            forward priming   backward priming

Her name was mocked.     forward priming   backward priming

I know who to kill.      backward priming  forward priming

Sealed the lock.         backward priming  forward priming

Wolf in white van.       backward priming  forward priming

You lady mannequin.      forward control   backward priming

This is a plant.         forward control   backward priming

Man will space walk.     forward control   backward priming

He's shot bad.           backward control  forward control

Evil lips are hammering  backward control  forward control


They pulled out the      backward control  forward control


That's the jive I was    pseudo            pseudo


Make me some money.      pseudo            pseudo

Here I am, rubber duck.  pseudo            pseudo

Yes I'm not brave.       pseudo            pseudo

Get our funds.           pseudo            pseudo

You were smashed.        pseudo            pseudo

I'm with love.           pseudo            pseudo

The fuss will serve us.  pseudo            pseudo


Message                       List 3            List 4


Cuddle me.               forward control   backward control

Hear the lie.            forward control   backward control

Her name was mocked.     forward control   backward control

I know who to kill.      backward control  forward control

Sealed the lock.         backward control  forward control

Wolf in white van.       backward control  forward control

You lady mannequin.      forward priming   backward priming

This is a plant.         forward priming   backward priming

Man will space walk.     forward priming   backward priming

He's shot bad.           backward priming  forward priming

Evil lips are hammering  backward priming  forward priming


They pulled out the      backward priming  forward priming


That's the jive I was    pseudo            pseudo


Make me some money.      pseudo            pseudo

Here I am, rubber duck.  pseudo            pseudo

Yes I'm not brave.       pseudo            pseudo

Ger our funds.           pseudo            pseudo

You were smashed.        pseudo            pseudo

I'm with love.           pseudo            pseudo

The fuss will serve us.  pseudo            pseudo




Mean Reaction Times (ms) on Lexical Decision Task


                                       Word type


                    Priming               Control                Pseudo


Message type    M       SD          M       SD          M       SD


Forward       1,220     333       1,317     344       1,563     542

Backward      1,302     460       1,312     361       1,617     638

Original manuscript received August 31, 2001

Final revision accepted August 29, 2002


Begg, I. M., Needham, D. R., & Bookbinder, M. (1993). Do backward messages unconsciously affect listeners? No. Canadian Journal of Experimental Psychology, 47, 1-14.

Byrne, T., & Normand, M. (2000, March/April). The demon-haunted sentence: A skeptical analysis of reverse speech. Skeptical Inquirer, 46-49.

Cowan, N., Leavitt, L.A., Massaro, D. W., & Kent, R. D. (1982). A fluent backward talker. Journal of Speech and Hearing Research, 25, 48-53.

Grainger, J., Van Kang, M. N., & Segui, J. (2001). Cross-modal repetition priming of heterographic homophones. Memory & Cognition, 29, 53-61.

Jacoby, L. L., & Kelley, C. M. (1992). A process-dissociation framework for investigating unconscious influences: Freudian slips, projective tests, subliminal perception, and signal detection theory. Current Directions in Psychological Science, 1, 174-179.

Oates, D. J. (1991). Reverse speech: Hidden messages in human communication. Indianapolis, IN: Knowledge Systems.

Stark, C. E. L., & McClelland, J. L. (2000). Repetition priming of words, pseudowords, and nonwords. Journal of Experimental Psychology: Learning, Memory and Cognition, 26, 945-972.

Swart, L. C., & Morgan, C. L. (1992). Effects of subliminal backward-recorded messages on attitudes. Perceptual and Motor Skills, 75, 1107-1113.

Vokey, J. R., & Read, J. D. (1985). Subliminal messages: Between the devil and the media. American Psychologist, 40, 1231-1239.

We thank Joseph J. Ryan for pointing out the Skeptical Inquirer article that led to this research. We also thank the editors of The Journal of Psychology for many insightful suggestions.

Address correspondence to David S. Kreiner, Department of Psychology, Lovinger 1111, Central Missouri State University, Warrensburg, MO 64093-5089; (e-mail).

COPYRIGHT 2003 Heldref Publications



This material is published under license from the publisher through the Gale Group, Farmington Hills, Michigan.
All inquiries regarding rights should be directed to the Gale Group.

HighBeam™ Research, LLC. © Copyright 2004. All rights reserved.





Skeptic (
Altadena, CA); 6/22/2000; ANDERSON, JOHN CHRIS

A Test of the Reverse Speech Hypothesis: Are Listeners Able to Detect the Emotional Content at Backward Speech?

AT THE HEIGHT OF THE SCANDAL involving Monica Lewinsky if President Clinton had said during his Grand Jury testimony, "Kiss the lying ass," it would have caused quite an uproar. According to David John Oates, Clinton said precisely that. Only he said it backward while simultaneously responding to a question about his testimony in a previous deposition. Furthermore, Clinton was not aware he was saying "Kiss the lying ass" because it was placed into his speech stream by his unconscious mind. How is this possible? The mechanism is described by Oates in his book Reverse Speech: Voices from the Unconscious (see also his web page: [1]

Oates claims to have discovered that speech actually contains two signals. The forward message is constructed by the left hemisphere and is produced and perceived by the conscious mind. As that is happening, a backward message is being constructed by the right hemisphere representing thoughts and issues in the unconscious mind. The backward message serves a psychodynamic purpose for the speaker. Much as other defense mechanisms can be used to deal with unpleasant information in the unconscious, reverse speech allows a safe outlet for information that would otherwise overwhelm the person producing it.

According to Oates, the implications of his discovery are truly enormous. We agree that if true, they would be. Backward messages, he says, reveal the truth about people's mental states, their internal conflicts, and even the way they fit into the cosmological structure of the universe. Reverse speech has practical implications (serving as a lie detector for job interviews and police departments) and therapeutic applications (by revealing the true state of the unconscious, it is possible for a therapist to know precisely what treatments are required).

Many of the reversals that Oates finds are actually metaphors that he claims come from the collective unconscious (as described by Jung). It is the job of the reverse speech analyst to interpret those metaphors and explain what they mean. It takes years of practice to become a reverse speech analyst. Part of the training involves how to hear speech in the "gibberish" of the signal. Other training teaches the analyst the metaphors used in reverse speach. Without understanding the metaphors, the statements made in reverse speech make little sense.

The purpose of the research reported here was to test some of the claims made by Oates for the reverse speech hypothesis. One might wonder why a hypothesis that makes such fantastic claims is worthy of serious attention. The reasons are clear. If the hypothesis is correct the implications are as far-reaching as Oates and his followers claim. Businesses, the police, and even the partners of unfaithful spouses will have a new tool at their disposal for ferreting out the truth. Furthermore, therapists will be able to gain new insight into their patients' thought processes, making for more effective treatment. On the other hand, if the hypothesis is wrong, a lot of wasted time and effort can be saved. Clients can be spared the trouble of undergoing an ineffective treatment. Innocent people can avoid being accused, tried, and convicted, as they were in the early 1990s when belief in facilitated communication inspired fantasies of sexual abuse in facilitated conversations. [2]

The reverse speech hypothesis makes two claims that can be tested empirically: 1. Backward messages are being embedded into forward speech by the unconscious. 2. These messages can be perceived and understood by listeners. One could argue that the second claim is not all that unreasonable. Surprisingly, reversing speech does preserve a fair amount of the available phonetic information. Begg, Needham, and Bookbinder demonstrated that sentences played backward can be recognized later. [3] Saberi and Perrott demonstrated that if speech is cut into segments of up to 50 milliseconds (ms), and each segment is then reversed, listeners report perfect intelligibility. [4] There is also evidence to suggest that some aspects of meaning are related to phonological information. For example, the phonetic symbolism effect is that words describing small things (e.g., "teeny") tend to use high, front vowels, and words describing large things (e.g., "huge") tend to use low, back vowels. [5] Perhaps the unconscious can take ad vantage of these sorts of regularities to comprehend backward speech.

It is the claim that listeners can unconsciously perceive the content of reverse speech that we test in the experiments reported below. We begin by reviewing some relevant literature on the nature of unconscious processing. Then, we consider research into listeners' ability to detect backward messages in rock albums. Finally, we employ a methodology that has the best chance of allowing listeners to detect the meaning of reverse speech.


Loftus and Klinger asked the question "Is the unconscious smart or dumb?" [6] They defined several possible criteria for what makes an unconscious process smart: it is complex, it can deal flexibly with a novel situation, or it always does what is best for us. As in most psychodynamic theories, the reverse speech hypothesis would not include an endorsement of the third criterion. There are many ways in which the reverse speech hypothesis' version of the unconscious could be harmful; perhaps the simplest is that by using reverse speech as an outlet, the unconscious is revealing secret information that any listener can detect.

The reverse speech hypothesis does include the claim that the unconscious is smart according to the first two criteria. A process that can make minor tweaks to a forward message and still produce a meaningful utterance every 10-15 seconds would have to be complex, as would a process that could extract the information from the speech stream. The unconscious of the reverse speech hypothesis would also have to be able to deal with novel situations. For example, Oates claims it is possible for two people to carry on a backward, unconscious conversation while they are also having a forward conversation. Carrying on a conversation is certainly an activity that is unique to particular situations and would require a flexible process. Is there evidence that the unconscious is smart in this way?

Greenwald, in his "New Look 3," proposed that unconscious cognition is possible. He described three levels of unconscious processing that have been explored: physical features, single words, and multiword strings. Based on experiments using subliminal activation and selective attention to uncover effects of unconscious processing on activation, memory, and retrieval, Greenwald concluded that the unconscious analysis of physical features has been established and that some effects of single words have been obtained. However, "These studies have provided no evidence that can confidently be interpreted as indicating (attentionless) unconscious analysis at the level of multiword strings." [7] He thus concluded that "Unconscious cognition has been found to be severely limited in its analytic capability."

In other words, the kind of smart unconscious proposed by the reverse speech hypothesis is not supported by research on unconscious processing. In fact, Greenwald issued a challenge to find evidence that unconscious processes can extract the meaning of even a two-word sentence. This is hardly the level of sophistication required if the reverse speech hypothesis were correct, and it would be a severe condemnation of unconscious processing if the two word test failed.

In sum, research into the types of tasks that can be performed unconsciously suggests that the processes proposed by the reverse speech hypothesis are beyond the analytical capabilities of the unconscious. However, one could make the case that all that is really being postulated is an automatic process akin to those that assist in comprehending forward speech (like lexical retrieval). Evidence that the unconscious isn't smart does not, by itself, rule out the reverse speech hypothesis. Whether or not listeners can comprehend the meaning of backward speech is an empirical question.


Research has been conducted to investigate people's ability to detect the meaning of backward speech in the context of backward messages embedded in popular music. This research was inspired in part by religious leaders' claims that satanic messages are hidden in rock music, and by a court case in which it was claimed that two young men committed suicide because the message "do it" was hidden in a rock album. [9] The results of this research can be summed up quite simply: listeners are sensitive to the physical characteristics of backward messages, but they cannot discern the meaning of those messages.

Vokey and Read showed that people can discriminate male from female voices in backward speech (98% correct), can detect a change from one speaker to another (78% correct), and can determine the (forward) language of backward speech (47% correct for sorting amongst three languages). However, with these same backward speech stimuli, listeners could not tell statements from questions, could not tell if two statements had the same or different meaning, and could not tell if the statements made sense. Listeners were also unable to sort backward messages into the categories of nursery rhymes, Christian, satanic, pornographic, or advertising. [10]

Begg, et a1 extended Vokey and Read's results by showing that even if the backward messages are memorable, their meaning does not "'leak' through." They played messages to participants either forward or backward. All participants then attempted to determine if backward statements were old or new. Participants who had first heard the statements played backward were able to tell old from new statements better than participants who had only heard the statements played forward. So, participants were able to extract information about the physical features of the statements (as Vokey and Read found), and they had some memory of what they had heard. [11]

After the recognition task, participants were asked to rate forward versions of each statement as true or false. Begg, et a1 expected an illusory truth effect for participants who had extracted the meaning of the statements when they were first presented. Participants who had heard the forward statements did show an effect; participants who had heard the statements played backward did not. The researchers concluded that participants can extract information about the physical features of backward messages, but they cannot extract information about the meaning of the messages. They left the question open of whether or not listeners would be able to unconsciously perceive the emotional content of backward speech. Is there any evidence for the unconscious perception of emotional information?


Kunst-Wilson and Zajonc demonstrated that affective judgments could be made on the basis of information that participants did not consciously perceive. They prepared a number of distorted octagons that had sides of varying lengths that gave each a distinctive shape. The various kinds of octagons were presented for i millisecond under constrained viewing conditions. Participants were generally unaware of seeing anything more than a flash of light, and could not reliably discriminate later between octagons they had seen before and new, differently proportioned octagons on a subsequent forced choice task, supporting the claim that the initial perception was unconscious. However, when participants were asked to choose which of two shapes they liked better, they preferred shapes they had actually seen before to shapes they had never seen at above chance levels. [12]

A number of studies have shown that in anxious individuals threat-related information produces interference. [13, 14] This may possibly be due to a pre-attentive process. [15, 16] This type of effect is reasonable from an evolutionary perspective. Since the function of anxiety is to alert an organism that some sort of threat is present--even if that threat is presented in a brief exposure, or attention is directed somewhere else when the threat appears--anxiety is the ideal emotion to be perceived outside of conscious awareness.

Ohman makes the case that anxiety detection is hard-wired and can operate in an unconscious fashion. [17] He cites work by LeDoux showing that in rats there is a direct neural connection from auditory nuclei to a "significance evaluator" and "fear effector system" in the amygdala. [18] This pathway bypasses the route that analyzes a stimulus for meaning. Ohman proposes in his model for the generation of anxiety that it can more "plausibly be attributed to some gross and relatively simple features of the stimuli than to a complete analysis of their meaning." In other words, the generation of anxiety is exactly the kind of dumb process that the unconscious should be capable of carrying out.

Given that anxiety can arise due to information that is not consciously perceived, it is possible that anxiety-producing information presented in reverse speech may be unconsciously perceived as well. This notion will be the basis for the experiments reported below.


The same basic procedure was used for both of the experiments. Emotionally arousing information was recorded. This information was then played to participants either forward or backward, and they rated their arousal. We pitted two hypotheses against one another in the experiments. The prediction of the reverse speech hypothesis is that listeners will detect the emotional content of messages, even if they are played backward. In other words, there will be a difference between arousing information and neutral information; the direction in which the material is presented will have no effect.

The altemative hypothesis is that the emotional content of backward speech will not be perceived. If true, then increased arousal will only be found for arousing materials played forward. In other words, we are predicting an interaction. When the material is played forward, arousal will be high for arousing information, but not for neutral information. When the material is played backward, arousal will below for both kinds of information.


The question for the first experiment was this: Will listeners be able to detect the emotional content of arousing words played in reverse? For the first experiment the stimulus materials were lists of words that were either arousing or neutral. These lists of words were played either forward or backward.

The arousing words were infected, polio, mutilated, breathlessness painful, deformed, surgery, hospital, ambulance, wound, blood, burial, burns, choking, heart attack, hurt, germs, sickness, tumor, collapse, corrosive, accident, attack, drowned. The neutral words were depicted, agency, cable, tenant, house, stay, chart, woodpile, lamp, signature, reverse, assembly, camera, terminology vehicle, luggage, napkins, wall, sand, news, pencil, between, end, glass. Both lists of words were adapted from Lundh and Ost. [19] One male experimenter digitally recorded the word lists 16 times each to produce a stimulus long enough to affect arousal. These lists were transferred to cassette tapes. The neutral tape was 11 minutes and 38 seconds long; the arousing tape was 11 minutes and 44 seconds long. Reversed versions were made by reversing each list and transferring the reversed lists to cassette tape.

Arousal was measured using the adjective checklist from Mackay et al. [20] Participants rated 15 adjectives as to whether those adjectives applied to them at that moment. The adjectives (in the order they were presented) were drowsy, tired, vigorous, activated, stimulated, alert, idle, energetic, passive, aroused, active, somnolent, sleepy, lively, sluggish. Scores could range from 0 to 15, with higher scores meaning more arousal.

The experimenters recruited 45 people to participate in the experiment. Most of these participants were college students; all participation was voluntary. The participants were run one at a time. The experimenters told participants that they would be listening to a list of words. The participants were warned that they would rate some adjectives after the list. The experimenters then played the list of words to the participant. After the list, the participant was given the arousal checklist.

The data were analyzed using an independent samples Analysis of Variance (ANOVA) comparing the list type (arousing, neutral) and the direction (forward, backward). The prediction from the reverse speech hypothesis was that the arousal and neutral lists would differ in the amount of arousal produced, but this difference was not significant [F(1,41) = 1.24, MSE = 14.01]. The mean amount of arousal for the arousing list was 5.70 and the mean amount of arousal for the neutral list was 4.46.

As predicted by the alternative hypothesis, there was a significant interaction between the list type and the direction in which the lists were played [F(1,41) = 4.65, MSE = 14.01]. For lists played forward, the arousal score for the arousing list was higher than the arousal score for the neutral list. For lists played backward the arousal scores for the arousing and neutral texts did not differ. These data are illustrated in FIGURE 1.

The results are clear. When arousing words are played forward, they produce arousal. When arousing words are played backward, they do not. In addition to fitting quite well with the prediction from the alternative hypothesis, this result is exactly what was predicted for the apprehension of emotional content from backward speech by Begg et al.

One possible criticism of this study is that it does not reflect the natural task facing listeners. For example, if a listener is unconsciously evaluating the content of a politician's speech, the stimulus is not a list of words, but rather a connected discourse. Is it fair to claim that results from a word list will generalize to real speech? The second experiment was designed to answer this objection.


The stimuli for the second experiment were taken from the novel The Hot Zone [21] The arousing text was 4 minutes and 44 seconds long. It described a man dying from the Ebola virus on an airplane. The passage was extremely unpleasant, and it was chosen to produce a considerable amount of arousal in the participants. The tone of the passage can be gleaned from the following sample:

The airsickness bag fills up to the brim with a substance known as the vomito negro, or the black vomit. The black vomit is not really black; it is a speckled liquid of two colors, black and red, a stew of tarry granules mixed with fresh red arterial blood. It is a hemorrhage, and it smells like a slaughterhouse.

The neutral passage was taken from a part of the book describing a visit to a mountain rainforest. This passage was 4 minutes and 48 seconds long. A sample is:

We traveled in two Land Rovers, Carrie driving one and Robin driving the other... Carrie and Robin's two boys rode with Carrie. We were also accompanied by three men who were members of the MacDonalds' safari staff. Their names are Katana Chege, Herman Andembe, and Morris Mulatya. They are professional safari men, and they do most of the work around the campsite.

Although we have substituted more natural stimuli in the second experiment, the two hypotheses make the same predictions as in the first experiment. If the reverse speech hypothesis is correct, we would expect more arousal for the arousing text than the neutral text regardless of whether it is played forward or backward. For the alternative hypothesis, we would expect increased arousal only for the arousing text played forward.

One male experimenter digitally recorded both excerpts. Forward and backward versions of these were copied onto cassette tapes for presentation to participants. A total of 83 students from psychology courses participated in the experiment, for which they received extra credit. For each session, participants were told that they would listen to someone speaking and then rate some adjectives. The appropriate tape was played, and the adjective checklist was handed out immediately after it finished.

The data were analyzed using an independent samples ANOVA comparing the text type (arousing, neutral) and the direction (forward, backward). Once again, the prediction from the reverse speech hypothesis was that the arousal and neutral texts would differ in the amount of arousal produced. The means for arousing and neutral were 5.88 and 2.64, respectively. This difference was significant [F(1,79) = 15.92, MSE = 13.59]. However, the interpretation of this difference is qualified by the interaction between the two variables, which was significant [F(l,79) = 7.73, MSE = 13.59]. When the text was played forward, the arousal score was lower for the neutral paragraph than for the arousing paragraph. When the text was played backward the arousal scores for the neutral and arousing paragraphs did not differ. These data are illustrated in FIGURE 2. Even though there is an overall difference between arousing and neutral texts, that difference results from when the texts are played forward. As with a list of words, when a text is the stimulus the reverse speech hypothesis cannot account for the results.

The results of the two experiments are remarkably consistent Highly arousing information only produces arousal if it is played in the forward direction. When arousing information is played backward, it is no longer arousing. Given the results of previous research on listeners' ability to unconsciously perceive the meaning of backward speech, our are not surprising. However, these experiments do show that even for arousal--something that is probably hard-wired into the perceptual system and that can arise from stimuli that are not themselves consciously perceived [22]--information in backward speech does not get through to a listener. This is true even though reversing speech preserves most of the phonetic information of forward speech.


How much damage do our results do to the reverse speech hypothesis? At this point, the second claim (that listeners can perceive the meaning of reverse speech) has been proven false. The semantic content does not get through, and neither does the emotional meaning. The first claim (that speakers embed backward messages into speech) has been left relatively intact. However, we have two reasons why we think this part of the hypothesis may also be wrong.

First, a number of studies have shown that expectations play a large role in what listeners hear in backward speech. For example, Vokey and Read listened carefully to recordings of Jabberwocky and the 23rd Psalm played backward. They identified 12 areas in the tapes where strings of tones suggested spoken messages (such as "saw a girl with a weasel in her mouth). Vokey and Read them asked their listeners to try to hear one of the specific messages that they "discovered." Only half of the messages listeners were asked to listen for were actually "in" the backward recording they were hearing (the other half were from the other tape). Listeners were likely to hear the messages for which some evidence was actually present in the tape they were hearing. They did not hear the messages that had been "found" in the recording that was not played. [23]

To test the effect of expectation, Vokey and Read then asked listeners to try to hear other messages "discovered" in the backward recording they were asked to listen to. For example, if subjects had listened for "saw a girl with a weasel in her mouth" from Jabberwocky, they now listened for another phrase that there was evidence for--"I saw Satan." Participants were first asked if they had heard this message on their previous trials with the passage. Most had not, but they could hear it once they had been told to listen for it. Changing the mental set changed what messages people "heard" in backward speech. This type of biased listening is similar to a reverse speech analyst listening to backward speech for a message that he or she "knows" is there. For example, a therapist listening to tapes from a child who has been sexually abused knows to listen for reverse speech messages relating to the abuse. Anything in the backward speech that can be heard as sexual (or that can be interpreted as a metaphor relating to sexual acts) could then be detected. Other "messages" would be ignored.

Thorne and Himelstein further demonstrate how mental set can alter what a person hears in backward recordings. [24] They played rock-and-roll songs backward for listeners. One group was told to record their reactions to the music. A second group was told to write down any words they heard in the music. A third group was told to listen for satanic messages. For the group recording its reactions, 5% of what they thought they heard was satanic; for the group listening for words, 18% of what they thought they heard was satanic; for the group listening for satanic messages, 41% of what they thought they heard was satanic. People's task set determined whether or not they interpreted a message as satanic.

Another factor in "wishful listening" is the belief in the phenomenon itself. Benoit and Thomas showed that listeners who believed in subliminal perception were more likely to "hear" a subliminal message in a jazz composition than were nonbelievers (there was no subliminal message in the music). [25] Moreover, believers' moods changed in the direction suggested by the experimenter as a result of listening to the music (e.g., if they expected the music to induce a positive mood, their mood became more positive). Nonbelievers' moods did not change as a result of listening to the music.

The results of these studies suggest to us that what people "hear" in backward speech is a product of their belief in the phenomenon and their expectations about the content of whatever messages they are listening for. True, something is probably there in the signal that can be interpreted by the listener based on these expectations, but there is no evidence to show that what is in the stimulus is really "in" the stimulus and not merely in the imagination of the listener.

The second reason we doubt the claim that speakers embed messages in reverse speech deals with Oates' claims about the phenomenon itself. For example, Oates claims to have analyzed conversations taking place in reverse speech. These conversations are based on the listener's comprehension of the meaning of the backward message, which is not possible. As an example, consider a conversation Oates had with a reporter. Oates asked the reporter to lie about his age. When analyzing the reversals, Oates claims to find the reporter uttering his true age in reverse speech. Oates' response (also in reverse speech) contains this true age. This is supposed to show that Oates unconsciously perceived the reporter's age and echoed it back. Other backward "conversations" in Oates' book have listeners responding to utterances that are as much as eight words long (e.g., "Might as well keep saying that I'm filthy."). This is clearly beyond the capabilities of an unconscious that cannot meet the two word challenge. We propose th at Oates' ability to analyze reverse speech to produce a result that is not possible demonstrates that it is all wishful listening.

Our discussion of these issues does not conclusively prove that the first claim of the reverse speech hypothesis is wrong. However, it does highlight the fact that supporters of the hypothesis have a great deal of work ahead of them if they want to "prove" that it is true. At this point; the principle of Occam's razor--if two explanations exist for a phenomenon, choose the simpler explanation--can be invoked to rule out the claim that listeners embed backward messages in their speech. Well understood mechanisms already account for speech production. Layering on a second message produced by different processes introduces unnecessary complexity to the theory and a high standard of support is required for such a hypothesis to be credible.


At this point we feel compelled to remind readers why we think this hypothesis is worth testing. Outside of the world of academia, ideas can gain support on the basis of anecdotal reports. Simply appearing on television and repeating an idea to a credulous audience can be sufficient to give that idea momentum and help it work its way into the popular consciousness (Oates claims to have given more than 200 media interviews). For example, the idea that subliminal messages in advertising will exert an undue influence on consumers started with a report of a (possibly) fictitious experiment in a movie theater where participants were told to "eat popcorn." This idea has been expanded upon by Key and is now one of the "facts" about psychology that everyone knows, [26] It was, in part, responsible for the Judas Priest trial regarding backward messages in rock albums.

To the extent that an idea is an interesting diversion, it makes no difference whether its claims are true or false. However, when an idea is used in an applied setting, then its claims should be held up to scrutiny. The reverse speech hypothesis is being pushed as offering new insights into therapy. Psychologists are being encouraged to undergo reverse speech training so that they can incorporate it into their interactions with clients. Dawes makes a strong case that therapeutic interventions in psychology should be based on the results of scientific experiments and not on the claims of a practitioner who feels strongly that the intervention works. [27] In this case, all of the evidence suggests that the reverse speech hypothesis is wrong.

Oates also advocates the use of reverse speech in investigative work. He claims to have been consulted in at least one criminal investigation. In his list of endorsements, Oates cites an attorney and the National Private Investigator's Journal. Before the reverse speech hypothesis is used in applied settings, its validity should be thoroughly established. This is particularly important with respect to the judicial system where the consequences of an error can be so severe. Not only is the well-being of individuals at stake, but legal precedents can also be set. For example, even though Judas Priest was not found liable, the judge's decision that backward messages are not protected by the first amendment has been used in other cases.

In his book Oates proclaimed that "like it or not, the day of Truth has arrived." We agree, and our data show that the truth is that the reverse speech hypothesis is wrong.

Dr. William Langston, Ph.D. is an assistant professor of psychology at Middle Tennessee State University. His area of research specialization is the psychology of language, and his primary teaching responsibility is research methods. He is the author of nine scholarly articles and has made 12 presentations at national or international research conferences. He is also the author of the Research Methods Laboratory Manual (which will be published by Wadsworth this summer and from which this biography has been lifted). John Chris Anderson is an undergraduate psychology major at Middle Tennessee State University.

John Chris Anderson was an undergraduate psychology major at Middle Tennessee State University when the paper was written. He will pursue a Master's degree in Industrial/Organizational psychology in the Fall of 2001.


(1.) Oates, D.J. 1996. Reverse Speech: Voices From the Unconscious. San Diego: ProMotion Publishing.

(2.) Green, G. 1994. "Facilitated Communication: Mental Miracle or Sleight of Hand?" Skeptic, 2:68-76.

(3.) Begg, I. M., D.R. Needham, and M. Bookbinder. 1993. "Do Backward Messages Unconsciously Affect Listeners? No." Canadian Journal of Experimental Psychology. 47:1-14.

(4.) Saberi, K. and D. R Perrott. 1999. "Cognitive Restoration of Reversed Speech." Nature, 398:760.

(5.) Pinker, S. 1994. The Language Instinct. New York HarperPerennial.

(6.) Loftus, B. F. and M. R Klinger. 1992. "Is the Unconscious Smart or Dumb?" American Psychologist. 47: 761-765. See also Bruner, J. 1992. "Another Look at New Look 1." American Psychologist. 47:780-783.

(7.) Greenwald, A. G. 1992. "New look 3: Unconscious Cognition Redaimed." American Psychologist. 47:766-779.

(8.) Litman, R. E. and N. L Farberow. 1993. "Pop-Rock Music as Precipitating Cause in Youth Suicide." Journal of Forensic Sciences. 39:494-499.

(9.) Moore, T. E. 1996. "Scientific Consensus and Expert Testimony: Lessons From the Judas Priest Trial." Skeptical Inquirer. 20:32-38.

(10.) Vokey, J. R. and J. D. Read. 1985." Subliminal Messagen Between the Devil and the Media." Amen can Psychologist. 40:1231-1239.

(11.) Begg, et al, 1.

(12.) Kunst-Wilson, W.R. and R.B. Zajonc. 1980. "Affective Discrimination of Stimuli that Cannot be Recognized." Science 207:557-558.

(13.) MacLeod, C. and A. Mathews. 1988. "Anxiety and the Allocation of Attention to Threat." The Quarterly Journal of Experimental Psychology. 40A: 653-670.

(14.) Mathews, A. and C. MacLeod. 1985. "Selective Processing of Threat Cues in Anxiety States." Behaviour Research and Therapy 23:563-569.

(15.) _____. 1986. "Discrimination of Threat Cues Without Awareness in Anxiety States."

Journal of Abnormal Psychology. 95: 131-138.

(16.) Mathews, A., A. Richards, and M. Eysenck. 1989. "Interpretation of Homophones Related to Threat in Anxiety States." Journal of Abnormal Psychology. 98: 31-34.

(17.) Ohman, A. 1993. "Fear and Anxiety as Emotional Phenomena: Clinical Phenomenology, Evolutionary Perspectives, and Information-Processing Mechanisms." In M. Lewis and J. M. Haviland (Eds.), Handbook of emotions, 511-536. New York Guilford.

(18.) LeDous, J. E. 1990. "Fear Pathways in the Brain: implications for a Theory of the Emotional Brain." In P. F. Brain, S. Parmigiani, R. J. Blanchard, and D. Mainardi (Eds.), Fcar and Defense, 163-177. London: Harwood.

(19.) Lundh, L.G., and L.G. Ost. 1997. "Explicit and Implicit Memory Bias in Social Phobia: The Role of Subdiagnostic Type." Behaviour Research and Therapy 35:305-317.

(20.) Mackay, C., T. Cox, G. Burrows, & T. Lazzerini, 1978. "An Inventory for the Measurement of Self-Reported Stress and Arousal." British Journal of Social and Clinical Psychology 17:283-284.

(21.) Preston, R. 1994. The Hot Zone. New York Anchor Books Doubleday.

(22.) Robles, R, R Smith, C. S. Carver, and A. R. Wellens. 1987. "Influence of Subliminal Visual Images on the Experience of Anxiety." Personality and Social Psvchology Bulletin. 13:399-410.

(23.) Vokey and Read, 1985.

(24.) Thorne, S.B. and P. Himelstein. 1984. "The Role of Suggestion in the Perception of Satanic Messages in Rock-and-Roll Recordings." The Journal of Psychology 116: 245-248.

(25.) Benoit S.C. and R. L. Thomas. 1992. "The Influence of Expectancy in Subliminal Perception Experiments." The Journal of General Psychology. 119: 335-341.

(26.) Key, W.B. 1980. The Clam-Plate Orgy: And Other Subliminals the Media Use to Manipulate Your Behavior. Englewood Cliffs, NJ: Prentice Hall.

(27.) Dawes, R.M. 1994. House of Cards. New York The Free Press.

COPYRIGHT 2000 Skeptics Society & Skeptic Magazine


^top of page

David John Oates
Founder and Developer of Reverse Speech Technologies
PO Box 678, Noarlunga Centre, SA 5168. Australia
phone: 08 83824372 - international: 61 8 83824372

Reverse Speech TM is a trademark owned by David John Oates.
All contents of the web site is Copyright © 2001 by David John Oates