concurring:
I disagree with the majority as to the question whether Agent Hernandez spoke Spanish well enough that he could reasonably be viewed in his role as out-of-court translator as a “conduit” for Romo-Cha-vez, thereby precluding application of the hearsay rule with regard to the statements he made to Agent Simboli. In the context of the rest of the record in this case, however, the admission of Agent Hernandez’s translation was harmless. I therefore concur in the result as to the Nazemi-an issue. See United States v. Nazemian, 948 F.2d 522 (9th Cir.1991). I also concur in the remainder of the majority opinion, albeit with the same caveat regarding the survival of Nazemian after Crawford v. Washington, 541 U.S. 36, 124 S.Ct. 1354, 158 L.Ed.2d 177 (2004), as I expressed in United States v. Hieng, 679 F.3d 1131 (9th Cir.2012) (Berzon, J., concurring).1
I.
The record establishes that Hernandez’s grasp of Spanish was quite weak — much weaker than the majority opinion makes it out to be.
For example, although the opinion says that Hernandez “grew up ... speaking Spanish,” Hernandez actually testified that he “grew up listening to Spanish.” In other words, Hernandez’s parents spoke to him in Spanish. Many people grow up with relatives speaking to them in another language, which they can therefore understand to a degree but not necessarily speak. See Friedemann Pulvermüller & John H. Schumann, Neurobiological Mechanisms of Language Acquisition, 44 Language Learning 681, 685 (1994); see also Lily Wong Filmore, When Learning a Second Language Means Losing the First, 6 Early Childhood Res. Q. 323, 324 (1991) (observing that “[f]ew American-born children of immigrant parents are fully proficient in the[ir] ethnic language” and that “[o]nce these children learn English, they tend not to maintain or to develop the languages spoken at home, even if it is the only one their parents know”); Merrill Swain, The Output Hypothesis and Beyond: Mediating Acquisition through Collaborative Dialogue, in Sociocultural The*963ory and Second Language Learning 97 (James P. Lantolf ed., 2000) (discussing studies showing that the “output” of language through speaking and writing is critical to the acquisition of a foreign language).
Further, the opinion leaves the impression that Hernandez studied Spanish extensively; in fact, he did so only from fifth to seventh grade and then, again, for two years in high school, which he completed in the 1980s. Unfortunately, language teaching and learning in this country are such that this degree of education in a language is extremely unlikely to lead to any degree of fluency. See, e.g., Center for Applied Second Language Studies, Univ. of Or., What Proficiency Level Do High School Students Achieve? 2 (2010) (reporting that, after four years of high school level foreign language instruction in the United States, fewer than four percent of students achieve an “intermediate mid”2 level of speaking proficiency).
Similarly, the majority opinion embroiders the record, at least by implication, when it says that Hernandez spoke Spanish at home with his first and second wives. Hernandez did testify that both of his wives spoke Spanish, and that he speaks Spanish daily in his personal life. But he also testified that both of his wives spoke English too, and that he could “communicate very easily” with them in English. It is thus unclear whether Hernandez’s “daily” use of Spanish at home consisted of a few words or phrases a day, or whether he spoke Spanish extensively with his wives. His demonstration of his Spanish abilities at the trial, discussed below, suggests the former.
No more convincing as to Hernandez’s fluency in Spanish is the majority’s statement that Hernandez conducted interviews in Spanish on a regular basis. In fact, Hernandez testified that he had conducted only around 20 investigative interviews for border agents, of the type at issue here, in the past three years. Hernandez’s main use of Spanish at work consisted instead of taking sworn statements, including “biographical questions and such,” while processing non-U.S. citizens at the passport control station at the border. Given the disparate contexts of passport control and criminal interrogation, it is difficult to see how Hernandez’s experience in taking sworn statements adequately equipped him with the Spanish language skills necessary to interpret Agent Simboli’s interrogation of Romo-Chavez. Moreover, there is no indication in the record as to the accuracy with which Hernandez either translated interrogations or took sworn statements.
Hernandez’s trial testimony confirmed that his command of Spanish was far from dependable. Quite aside from the request that he recite from memory the Spanish version of Miranda warnings — a task that did not necessarily depend on language proficiency,3 as the majority notes — Her*964nandez was asked to translate in court a short (around 60-word) “waiver of rights” from the ICE Miranda form that he had gone over with Romo-Chavez. His translation was then compared with that of the court interpreter. The two translations were generally consistent, but were inconsistent in some ways that could matter when reporting what a potential criminal defendant said about his activities.
For example, the court interpreter translated the phrase “me han leido” as “they have read to me,” whereas Hernandez translated it is “I have read.” Obviously, such a transposition of subject and object could matter mightily when a suspect is giving his story in response to questioning. Similarly, Hernandez translated a phrase as “I sign this document,” whereas the interpreter translated it as “I have signed this document.” As with the subjectobject error, this kind of verb tense mistake is one that someone with a good grasp of Spanish should not be making. Moreover, it is again the kind of error that could result in a suspect being represented as having told one story whereas in fact he told another, with a material discrepancy between the two.
Nor were these errors momentary lapses. Hernandez testified that he didn’t know “the grammar rules for application to the Spanish language.” He also indicated that he relied on his wife for help with tenses when trying to complete sentences. For instance, she would correct him if he confused the phrase “I wanted to go” with “I went.”
While Hernandez testified that he “more commonly” made mistakes in verb tense, the cross-examination revealed that his Spanish vocabulary also had significant limitations. At one point, for example, Romo-Chavez’s attorney asked Hernandez to translate four sentences from Spanish to English, which the court interpreter subsequently translated as, “I am Juan Domin-quez. I live there near Phoenix. I’m going on a trip for pleasure and business. But in this business I had an accomplice that already gave a statement.” Hernandez was able to translate only the first two sentences. Explaining his inability to translate the third and fourth sentences, Hernandez stated, “I have never heard those words before.” Translation is an exacting task, for which professional translators train for many years. See generally Bureau of Labor Statistics, Interpreters and Translators, in Occupational Outlook Handbook (2010-11 ed.), http://www.bls. govt'oco/pdf/ocosl75.pdf; Developing Professional-Level Language Proficiency (Betty Lou Leaver & Boris Shekhtman eds., 2002). Even fully competent translators and interpreters disagree about the proper transformation of one language into another; that is why there are over ten translations of War and Peace, for example, listed for sale by Amazon. While we need not hold that an out-of-court translator must be qualified to be a court interpreter — or a translator of novels' — to meet the Nazemian standard, some high degree of reliability is necessary. The very concept of a “conduit” — especially where the defendant had nothing to do with choosing the translator — suggests accuracy, if not elegance, in converting what the defendant said in one language to another. Where the interpreter’s background and tested proficiency does not confirm the capacity for such accuracy, the entire premise on which Nazemian stands — shaky though it may be with regard to the Confrontation Clause after Crawford, 541 U.S. 36, 124 S.Ct. 1354, see Hieng, 679 F.3d at 1149 (Berzon, J., concurring) — collapses.
*965Romo-Chavez testified that he understood only 30-40% of what Hernandez said to him, which would have consisted primarily of Miranda warnings and questions. Of course, there is ample reason to think this estimate self-serving. But even if Hernandez’s accuracy, or comprehensibility, was much higher, there could easily have been mistakes in translation concerning grammar and verb tense that mattered.
For example, with respect to the question of where Romo-Chavez planned to exchange the two shirts, Simboli testified that Hernandez told him that Romo-Cha-vez first said Phoenix but then changed his answer to the Tucson Mall. However, what Romo-Chavez actually said may have been that he would have gone to the Tucson Mall had he not had other business in Phoenix. Alternatively, he might have said that he had been to the Tucson Mall before (which, as Romo-Chavez testified during trial, he had been) but was going near Phoenix this time.4 Of course, we don’t know that he said either one, but the sorts of verb tense errors we know Hernandez did make could lead to such mistakes.
Similarly, even a subtle misuse of vocabulary could have mattered here. For example, Simboli testified that, in response to a question he did not quite remember, Romo-Chavez told him, “Everything in the car is mine.” In contrast, Romo-Chavez testified that he told Simboli, through Hernandez, only that the clothes and documents in the car were his. Whereas Sim-boli may have intended to ask whether Romo-Chavez acknowledged ownership of everything in the entire car (that is, including the gas tank), the question that Hernandez posed to Romo-Chavez, given Hernandez’s word use difficulties, may actually have been whether he owned everything inside the car — that is, the interior cabin. Such a discrepancy could constitute the difference between a seeming confession and an innocuous admission.
I explained in my concurrence in Hieng that Nazemian should be reconsidered en banc in an appropriate case with regard to its Confrontation Clause holding. See 679 F.3d at 1149 (Berzon, J., concurring). But, as this case illustrates, Nazemian will retain importance as a hearsay precedent even if it is determined that the Confrontation Clause requires that the interpreter be available for cross-examination. Here, Hernandez was available for cross-examination and was extensively cross-examined. He did not, however, remember any details of this particular interpretation session, as is understandable and as will often be the case. As the majority holds, the Confrontation Clause is satisfied by the opportunity to cross examine, even if the person examined does not remember enough to be useful. Maj. Op. at 1141. So, in this species of case, it is often compliance with the Nazemian requisites— particularly, enough language competence that the interpreter’s statements can be viewed as those of the suspect (or other interviewee) — that will be the protection against inaccurate translation prejudicial to the defendant.
Obviously, precisely because the defendant does not know English, he is in no position to judge the accuracy of the translation at the time it is occurring. Where, as here, his defense at trial is, in part, that he did not say during the out-of-court interview what the translator says he said, some meaningful assessment of the language competence of the translator is essential. Hernandez was not sufficiently *966competent that he could be viewed as a “conduit” such that his statements could be taken as those of Romo-Chavez, eliminating what would otherwise be a hearsay problem. Indeed, Hernandez conceded at trial that he did not consider himself to be fluent in Spanish.
One additional note: It does not seem to me to be too much to ask to require that the Department of Homeland Security (DHS) have competent, trained interpreters available at the border to interview suspects (and others), at least in Spanish. Or, better, DHS could record all the interrogations, so that the actual interchange can be available to the parties and the jury (with the help of certified interpreters). Alternatively, law enforcement can wait to interview arrested suspects until there is a competent intermediary available. Here and in many other instances, what was found in the car was sufficient to establish probable cause and therefore to arrest Romo-Chavez; there was no exigent reason to interview him before a competent interpreter could be found.
Errors in translation do occur, and can be critical. See, e.g., Grigoryan v. Mukasey, 277 Fed.Appx. 742, 744 (9th Cir.2008) (granting a motion to reopen an alien’s removal proceedings, where a hearing transcript incorrectly stated that “[the alien] testified to being attacked because her mother was a ‘cook’ rather than a ‘Turk,’ ” an error that went to “the heart of her claim” of having been persecuted based on a protected ground). This court has recognized that “an incorrect or incomplete translation is the functional equivalent of no translation.” Perez-Lastor v. INS, 208 F.3d 773, 778 (9th Cir.2000). We should not bend the hearsay rules out of shape so as to treat statements by one person as if they were statements by another, unless the equivalence is proven. Here, it was not.
II.
In the end, however, I would hold the error in admitting Hernandez’s hearsay statements in this case to be harmless error, for three reasons.
First, Agent Simboli’s testimony of what Romo-Chavez said, via Hernandez’s translation, did not contain any specific admission from Romo-Chavez that he knew of the drugs in the gas tank. Instead, the government used Simboli’s testimony only to attack through contradictions the story Romo-Chavez told at trial. And much of what Agent Simboli reported that Romo-Chavez told him — or did not tell him — via Hernandez was similar to what two other border agents, Agents Tipling and Aldrich, who separately spoke to Romo-Chavez, reported. Romo-Chavez does not challenge Agent Tipling’s or Agent Aldrich’s reports of what he said in Spanish as hearsay. Both officers, like Agent Simboli, for whom Hernandez translated, testified to statements that contradicted Romo-Chavez’s story at trial.
Moreover, Romo-Chavez’s story at trial changed drastically during cross-examination, as he was confronted with evidence concerning his ongoing relationship with Vargas-Diaz, the individual who, according to Romo-Chavez’s testimony, must have placed the drugs in the car he drove across the border. For example, Romo-Chavez initially testified that he ran into Vargas-Diaz only one time, by coincidence, in No-gales. Later, however, he admitted to meeting Vargas-Diaz in Nogales between six to eight times. Similarly, Romo-Cha-vez at first claimed to have never called Vargas-Diaz, but subsequently said that he had called him about three times. At the outset, Romo-Chavez also said that the two ran into each other by coincidence at the Tucson Mall, but he later admitted that Vargas-Diaz knew that he would be *967there, because the men had spoken beforehand by phone.
The government also introduced extensive evidence that Romo-Chavez and Vargas-Diaz drove the same two cars across the border at different times. Romo-Cha-vez made a total of about thirteen trips across the border in those two cars, while Vargas-Diaz made a total of about twelve trips. In addition, the government presented evidence that the two men crossed the border simultaneously approximately thirteen times, evidence inconsistent with Romo-Chavez’s representations that he knew Vargas-Diaz very casually and had no role in any cross-border transactions in which he engaged.
Given these aspects of the record, as well as the basic fact that drug smugglers are unlikely to entrust hundreds of thousands of dollars of drugs to an unknowing driver, cf. United States v. Toro-Barboza, 673 F.3d 1136, 1144-45 (9th Cir.2012), the jury likely would have disbelieved Romo-Chavez’s version of events even without the report of Simboli’s interrogation. If they did, then there was ample evidence from which the jury could have inferred that Romo-Chavez was working with, rather than duped by, Vargas-Diaz, whom he identified as the source of the car and of the drugs.
Finally, Romo-Chavez had ample opportunity to cross-examine Hernandez as to his language competence. Given the lack of competence Hernandez both displayed and admitted to, the likelihood that the jury put much credence in Simboli’s report, as opposed to all the other evidence, is quite low.
Critically, the Nazemian error was an evidentiary one only, not of constitutional dimensions. So the government’s harmless error burden is only to show it is more likely than not that the same result would have been reached absent Agent Simboli’s report of his interrogation of Romo-Cha-vez. See United States v. Gonzalez-Flores, 418 F.3d 1093, 1099 (9th Cir.2005). In combination, the strength of the government’s circumstantial evidence, the many other reasons that the jury had for disbelieving Romo-Chavez’s testimony, and the effective cross-examination of Hernandez regarding his Spanish competence, more than meet the government’s burden of demonstrating that the evidentiary error was harmless.
Conclusion
For the reasons I have explained, I concur only in the result with regard to the Nazemian issue.
The district court characterized the competence of the translator as “marginal.” I believe even that begrudging characterization to be clearly erroneous, for the reasons I have given. But even if not clearly erroneous, the district court’s assessment of Hernandez’s ability as an interpreter is surely one as to which another district judge could have ruled otherwise without committing clear error. Thus, if DHS continues to provide translators who are untrained, untested, and of, at best, “marginal” competence, and also continues not to record interrogations involving an interpreter, it will be unnecessarily risking the validity of the resulting convictions.
. In Hieng, I state in my concurring opinion that Nazemian is not so " 'clearly irreconcilable' " with Crawford, as to permit a three-judge panel to overrule Nazemian. 679 F.3d at 1145 (Berzon, J., concurring) (quoting Miller v. Gammie, 335 F.3d 889, 900 (9th Cir.2003) (en banc)). In particular, Nazemian places the translator’s version of a statement beyond the scope of Confrontation Clause analysis altogether, a holding that would, if correct, seem to survive the fundamental alteration in Confrontation Clause analysis wrought by Crawford. Nonetheless, as I explain in my concurring opinion in Hieng, I am of the view that Nazemian''s holding does ultimately rest on a pre-Crawford understanding of the unity between hearsay concepts and Confrontation Clause analysis. See id. at 1149 (Berzon, J., concurring). The notion that a translator's out-of-court version of a testimonial statement need not be subject to cross-examination seems in great tension with Melendez-Diaz v. Massachusetts, 557 U.S. 305, 129 S.Ct. 2527, 174 L.Ed.2d 314 (2009) and Bullcoming v. New Mexico,-U.S.-, 131 S.Ct. 2705, 180 L.Ed.2d 610 (2011), which hold that laboratory reports may not be admitted without testimony by the persons who conducted the laboratory tests. Id.
. The American Council on the Teaching of Foreign Languages (ACTFL) has characterized foreign language speakers at the "intermediate mid” level as those capable of "handl[ing] successfully a variety of uncomplicated communicative tasks in straightforward social situations,” whose "[c]onversation is generally limited to those predictable and concrete exchanges necessary for survival in the target culture,” who "have difficulty linking ideas, manipulating time and aspect, and using communicative strategies” when called on to perform functions or handle topics at the advanced level, and who "are generally understood by sympathetic interlocutors accustomed to dealing with non-natives.” Am. Council on the Teaching of Foreign Lang., ACTFL Proficiency Guidelines 7 (2012).
. Nonetheless, Hernandez did make mistakes that reflected on his Spanish abilities while attempting to perform this task. For example, he translated the word "questions” as "preguntos” rather than "preguntas,” thereby mistaking the gender of the noun. Such an *964error is not necessarily innocuous. For example, the Spanish term “derecha” means "right,” as in the opposite of "left,” whereas the term “derecho” means "right,” as in a human right or constitutional right. See Oxford Spanish Dictionary 256 (3d ed.2003).
. Romo-Chavez agreed that he "went to a place near Phoenix, Scottsdale,” and that he was "going to go to Dillard’s in Scottsdale.”