Personalized User Model LLP v. Google Inc.

Filing 584

REDACTED VERSION of 578 Declaration,, of Antonio Sistos in Support of Google's Opposition to PUM's Motion to Exclude Portions of Dr. Edward Fox's Non-Infringement Report [Volume II of II] (Exhibits 2-8) by Google Inc.. (Attachments: # 1 Exhibit 2-4, # 2 Exhibit 5-6, # 3 Exhibit 7 part 1, # 4 Exhibit 7 part 2, # 5 Exhibit 7 part 3, # 6 Exhibit 7 part 4, # 7 Exhibit 7 part 5, # 8 Exhibit 7 part 6, # 9 Exhibit 8 part 1, # 10 Exhibit 8 part 2, # 11 Exhibit 8 part 3, # 12 Exhibit 8 part 4)(Moore, David)

Download PDF
1 1 IN THE UNITED STATES DISTRICT COURT 2 FOR THE DISTRICT OF DELAWARE 3 4 PERSONALIZED USER ) 5 MODEL, LLP, ) 6 Plaintiff, 7 vs. 8 ) GOOGLE, INC., 9 ) CA number 09-525 (LPS) ) Defendant. 10 11 - ) - - - - VIDEOTAPED DEPOSITION OF JAIME CARBONELL 12 WASHINGTON, D.C. 13 NOVEMBER 27, 2012 14 The videotaped deposition of JAIME CARBONELL was 15 convened on Tuesday, November 27, 2012, 16 commencing at 10:05, at the law offices of SNR 17 Denton, located at 1301 K Street, Northwest, in 18 Washington, D.C., before Paula G. Satkin, 19 Registered Professional Reporter and Notary 20 Public. 21 - - - - - 22 23 24 25 Job No. CS1565706 Veritext Corporate Services 800-567-8658 973-410-4040 2 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 4 CONTENTS APPEARANCES 1 2 3 ON BEHALF OF THE PLAINTIFF: JENNIFER BENNETT, ATTORNEY AT LAW 4 5 SNR DENTON 6 1530 Page Mill Rd. 7 Suite 200 8 Palo Alto, CA 94304-1125 9 650.798.0300 10 11 ON BEHALF OF THE DEFENDANT: 12 DAVID PERLSON, ATTORNEY AT LAW 13 JOSH SOHN, ATTORNEY AT LAW 14 QUINN EMANUEL 15 50 California Street 16 22nd Floor 17 San Francisco, CA 94111 18 415.875.6600 19 and 20 MARC FRIEDMAN, ATTORNEY AT LAW 21 QUINN EMANUEL 22 51 Madison Avenue 23 22nd Floor 24 New York, NY 10010 25 212.849.7000 JAMIE CARBONELL EXAMINATION BY MR. PERLSON BY MS. BENNETT BY MR. PERLSON 7 289 294 3 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 ALSO PRESENT: T.J. O'TOOLE, Videographer 5 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 EXHIBITS CARBONELL EXHIBIT NO: PAGE NO: Ex. 1 - A Personal Evolvable Advisor 93 For WWW Knowledge-Based Systems M. Montebello Ex. 2 - '040 patent 111 Ex. 3 - '276 patent 125 Ex. 4 - Personal WebWatcher 137 Dunja Mladenic Ex. 5 - Syskill & Webert: Identifying 186 interesting sites Ex. 6 - Learning and Revising User 187 Profiles: The Identification of Interesting Web Sites Ex. 7 - Learning Probabilistic User 198 Models Ex. 8 - '032 patent 201 Ex. 9 - Order 224 Ex. 10 - Collecting User Access 254 Patterns for Building User Profiles and Collaborative Filtering Ex. 11 - Report of Michael I. 276 Jordan, Ph.D. 2 (Pages 2 to 5) Veritext Corporate Services 800-567-8658 973-410-4040 6 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 PROCEEDINGS (10:05 a.m.) THE VIDEOGRAPHER: On the record with disk number one of the video deposition of Dr. Jaime Carbonell taken by the Defendant in the matter of Personalized User Model LLP versus Google Inc. and Google Inc. versus Personalized User Model LLP, both cases being heard before the United States District Court for the District of Delaware, Civil Action Number 09-525 LPS. This deposition is being held at the law offices of SNR Denton, located at 1301 K Street, Northwest, in Washington, D.C., on November 27th, 2012, at approximately 10:05 a.m. My name is T.J. O'Toole. I am the certified legal video specialist. The court reporter is Paula Satkin. We are both here representing Veritext of New Jersey. Will counsel please introduce themselves and indicate which parties they represent. MS. BENNETT: Jennifer Bennett representing Plaintiff Personalized User Model and the witness, and with me today I have Marc 8 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 Q. And you understand that you are testifying under oath as if you were testifying before a jury; correct? A. Yes. Q. And I'm going try to be as clear as I can today and -- but if I ask a question that you do not understand, please let me know and I will do my best to make it more clear. Okay? A. Okay. Q. Now, Dr. Carbonell, you -- you co-authored a book called Machine Learning and Artificial Intelligence Approach; is that right? A. There were three of them in that series. Q. Okay. And when -- when was the first one? A. I believe it was 1983. Q. When was the second one? A. 1986. Q. And how about after that? A. There was one more. I don't recall the date. A year or two afterwards. Q. It was about 1990; does that sound right? 7 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 Friedman. 1 2 Quinn Emanuel representing Defendant Google. 3 MR. SOHN: And Josh Sohn of Quinn 4 Emanuel also representing the Defendant. 5 THE VIDEOGRAPHER: Thank you. 6 Will the court reporter please swear in the 7 witness. 8 Whereupon-9 JAIME CARBONELL 10 a witness, called for examination, having been 11 first duly sworn, was examined and testified as 12 follows: 13 EXAMINATION BY COUNSEL FOR THE DEFENDANT 14 BY MR. PERLSON: 15 Q. Good morning. Could you state and 16 spell your name for the record? 17 A. It's Jamie Carbonell, J-A-I-M-E, 18 C-A-R-B-O-N-E-L-L. 19 Q. And do you go by Dr. Carbonell or 20 Mr. Carbonell? 21 A. Dr. Carbonell. 22 Q. Okay. And you've been deposed 23 before; correct? 24 A. Yes. 25 MR. PERLSON: David Perlson from 9 A. It could be, or it could have been a little earlier. Q. You've been publishing in the machine learning field since then? A. Yes, I have. Q. When was the last time? A. This year. Q. What -- what did you publish this year? A. The latest paper is one at the Association of Computing Machinery on learnability of DNF, disjunctive normal form, expressions. Q. What's that? A. Disjunctive normal form is a -learning when you have different expressions of the target concept. So maybe an example is the clearest way to explain it. Q. Sure. A. If there is bank fraud, there are different ways of defrauding the bank. For example, by pretending to be a customer when you really aren't. By pretending you have a lot more in an account than you really do and withdrawing it. Insider transactions that are 3 (Pages 6 to 9) Veritext Corporate Services 800-567-8658 973-410-4040 22 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 be safe in here. Q. That's good to know. A. A Bayesian world, you can use information like that without data based on priors which can be updated if you have other observations. If you observe that a meteor has struck somewhere else and a second one has struck, then the probability that a third one will strike might be higher than it would have been had there been no other meteor strikes. In the frequentist case, you're not allowed to use the equivalent of a prior. You base it only on the data. And if there is no data, you basically cannot provide an estimate. Q. And -- but mathematically, is -is that probability expressed as a number between 0 and 1? MS. BENNETT: Objection. Form. THE WITNESS: In the frequentist approach, it is. BY MR. PERLSON: Q. I'm sorry. A. I was trying to answer your earlier question. Q. The -- okay. Go ahead. 24 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 of 1, that is -- and that does not mean that a number of 2 is twice as likely to be of interest to the user as 1, would that be a probability? MS. BENNETT: Objection. Form. THE WITNESS: Sorry. I did not quite grasp the -- the premise of your question. If you're talking about if you had a scale that went from 1 to 10, 1 was the lowest value and 10 was the highest value -BY MR. PERLSON: Q. Correct. A. -- 2 then would not represent twice as likely as 1 because if 1 is the low end of the scale, 1 means it's not going to happen. Q. Okay. Okay. Let's say it's between 0 and 10? A. Okay. Q. And if I assign something a number 1 and in order for that range to be a range of probabilities, wouldn't it be the case that a number of 2 would have to be twice as likely to be -- show the interest of a user in a document in order for it to be a probability? MS. BENNETT: Objection. Form. THE WITNESS: Okay. So, first of 23 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 A. It's not just frequentist and Bayesian. If you go a little broader, there are things interpreted like fuzzy logic and other forms of reasoning with degrees of belief, of belief propagation, that do not require the values of the interval to be between 0 and 1. Q. Okay. So fuzzy logic doesn't require a number between 0 and 1? A. Some types of fuzzy logic do not require that; others do. Q. Okay. A. Fuzzy logic is a broad term for introducing numbers into logic -- degree of belief into logic. Q. But in order for there to be a degree of belief, there has to be some sort of scale of the -- the degree of likelihood of interest? A. Yes, sir. That's right. Q. So if I had a -- if I had a -numbers that went from 1 to 10 and assigning something a number of 2 was not -- well, in order for something to be a probability, would -- let's say I have a -- I can assign numbers 1 through 10, and if I assign something a number 25 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 all, let's call it a likelihood rather than a probability. BY MR. PERLSON: Q. Okay. A. Technically speaking, it's hard to think about probability other than 0-to-1 interval. That's how the math works out. In probability theory, what you're saying is correct in the sense that it's a linear scale. If something has twice the probability value of another event so long as it's not 0, it means it's twice as likely to happen. If you use likelihoods, the typical interpretation is the same. So if a 0-to-10 scale, an event has a probability or likelihood of 1 and the second event has a likelihood of 2, the second event would be twice as likely to happen as the first. It is not required that the scale be linear, but by convention you assume linearity unless told otherwise. So anybody's scheme of likelihood is either linear or they inform you how to calculate it. Q. How to -- what do you mean, inform 7 (Pages 22 to 25) Veritext Corporate Services 800-567-8658 973-410-4040 26 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 you how to calculate it? A. If it's -- if the scale were not linear, it could be, for example, on a -- based on a sigmoid function or something else, then they would have to provide you that sigmoid function that says if something has a value of 2 and something else has a value of 1, here's how you estimate how much more likely the value of 2 is over the value of 1. So in the absence of providing a function, and I used sigmoid function as an example of one that is sometimes used, it would be exactly as you say. It would be linear. Q. And then let's say that I had a -a -- likelihood numbers of 1 through 4 where 1 was somewhat likely, 2 was very likely, 3 was extremely likely, and 4 was a certainty of likelihood. Would that be -- would those numbers 1 through 4 be probabilities? MS. BENNETT: Objection. Form. THE WITNESS: Okay. So, first of all, you didn't define the bottom end of the range -- 0 means unlikely or 0 means impossible? BY MR. PERLSON: Q. Let's say 0 means highly unlikely? 28 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 it's called an estimation. It's an approximate calculation of the value. Now, a model can have multiple parameters. It can have parameters that represent whether they like certain terms, certain concepts, whether they like certain sources of documents, whether they like certain topics within the documents, whether they like to see documents about the same area they've seen before and so forth. The collection of all these parameters together with a mathematical function that combines them is the model. And estimating the parameters is finding or estimating a value, approximating a value, for each one of these inputs to the model, as it were, one of these variables in the model. A parameter is like a variable. It has a value. And you're estimating the values. BY MR. PERLSON: Q. Is the parameter the value or the -- or the variable? A. It's used to mean both, and that is a cause of confusion, I'm afraid. I wish that my colleagues had been, let's say, more 27 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 A. So where's the point that it means -- so neither of the two ends are definitive, so that cannot be converted directly into a probability. Q. So you -A. A probability requires both end points to be nailed down, to be defined. The impossible versus the certain. Q. Now, the patent talks about estimating parameters. Are you familiar with that? A. Yes. Q. And what does it mean to estimate a parameter? MS. BENNETT: Objection. Form. THE WITNESS: It means to compute the value of that parameter based on the information available. That computation can be inexact. It can be an approximation because the amount of information available is finite. It's not all possible likes or dislikes by a user. It's a finite set of those documents they have already seen. Given that it's based on partial observations of how a person would react to a document rather than the totality, that's why 29 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 discriminating in using it to mean only one of the two. That would have avoided future -future confusion, but a parameter is used to mean the value and the parameter is also used to be the variable. Q. And is that -- is that how it's used in -- in the patents, too? MS. BENNETT: Objection. Form. THE WITNESS: The patent talks about estimating the parameters. It really talks about estimating the values of variables. BY MR. PERLSON: Q. Sorry. Were you done? A. Yeah. I'm done. Q. And the -- in order to -- to estimate the values of the variables, is that done by a calculation? MS. BENNETT: Objection. Form. THE WITNESS: It is done -everything is done by a calculation. So an estimation is a calculation based on the available data. BY MR. PERLSON: Q. And that's the -- that's -- the -you mentioned a Dr. Jordan earlier. Is Michael 8 (Pages 26 to 29) Veritext Corporate Services 800-567-8658 973-410-4040 258 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 THE VIDEOGRAPHER: Counsel passed me a note asking me how much time we had left, and I told her that we've been on the record 6 hours and 10 minutes. I have no idea how much time is left. MR. PERLSON: Okay. MS. BENNETT: 50 minutes. MR. PERLSON: That's all I need to know. BY MR. PERLSON: Q. The -- okay. So let's go to -okay. And so on page 59, you see it says "representation for web navigation"? A. Yes. Q. And then underneath it, it says, "The probability distribution of the pages to be accessed is based on collecting the visiting patterns of many users." A. Yes. Q. And what do you understand the probability distribution of the pages is that's referred to there? A. That is the probability of navigating from one web page to another web page by following a link between these pages. In 260 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 -- of hyperlinks in order to be in the same current state. So he trades off the order of the model in order to balance accuracy with -- and generality, and he mostly does an order M equals 2 model as he states on the second -- in the middle of the second column. He goes through an example that I have no need to repeat here. And so this is essentially a navigation process, and he shows in a finite state diagram in Figure 1 on the next page -- he does that illustration so he can calculate the probability that you will traverse a certain link, a certain hyperlink, from one page to another based on what others have traversed before. Q. And that probability is used to determine the variable TIJ; is that right? A. That probability may be used to initialize the variable TIJ. He has -- there's two parts to this -- to this paper, the part that we are talking about now and the entropy-based part which is just before it. In the entropy-based part that he -- excuse me. E-N-T-R-O-P-Y. In the 259 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 some cases some links are followed by many users. Other links may be followed by few users. Some links may be followed by no users. Q. And how is that information used in the -- in Wasfi? A. The main use that he puts to it is he builds what's called an order M model. So is there any problem with the recording? Q. No. A. We can just continue? Q. He said that there was 6 hours 10 minutes, and then he passed a note that said 5 minutes. We were just chuckling -- it seemed inconsistent, but nothing to do -- sorry. A. So an order -- an order M model decides how far back in the sequence of navigation you look to. So an order 1 model means that you look at the current page and where else you go next. An order 2 model is where you came from, the current page. An order 3 model, where you came from before the last one you came from and so on. The higher the order of the model, the more information you have, but then again, the less generalization because you must have traversed this particular sequence of 261 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 entropy-based part that comes before it, he defines TIJ in a different way as a negative log of probability, rather than the probability itself. That negative log of the probability is bounded from -- let's see. The probability is zero -- is bounded from zero to infinity. So it's not really a probability in -- in that particular definition of TIJ on page 59, column 1. MR. PERLSON: Okay. I think we need to take a break. THE VIDEOGRAPHER: This ends disk number 4 of the Carbonell deposition. The time is 6:01:58. Off the record. (A brief recess was taken.) THE VIDEOGRAPHER: On the record with disk number five of the testimony of Dr. Jamie Carbonell in the matter of Personalized User Model versus Google. The date is November 27th, 2012. The time is 6:11:21. BY MR. PERLSON: Q. So now, we were discussing the variable TIJ in Wasfi? A. Yes. Q. What is that? 66 (Pages 258 to 261) Veritext Corporate Services 800-567-8658 973-410-4040 262 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 A. TIJ is meant to be measure of importance or interestingness of the page -- of the Ith page the Jth user. Ith page to the Jth user. In fact, I believe Wasfi says so explicitly -- let me find it. Yes, column 1, page 59, just above the formula. However, that statement is not exactly consistent with his -with his formula. This is sometimes called stochastic entropy rather than the more traditional or more commonly used Shannon entropy. Shannon entropy is minus P log P, and that is bounded on both ends. This is unbounded, at infinity. Q. Which is unbounded? A. TIJ, the H of PR, which is the same thing. Q. So something that is unbounded cannot be a probability; is that right? A. That's right. It cannot be normalized into a probability. Q. The TIJ variable, that indicates how much weight a new page should get in a user's profile when that user accesses that page; right? A. That's right. 264 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 zero. And so this pathological case will not occur with nearly as much frequency as it would occur if it was just an individual user who had not traversed that link. Q. Can the weights of a user-specific learning machine be -- let me start over. Can the parameters of a user-specific learning machine be set based on formulas that take into account the activity of other users? MS. BENNETT: Objection. Form. THE WITNESS: You're talking about a learning machine for an individual user or a learning machine for all users? BY MR. PERLSON: Q. A learning machine for an individual user? A. A learning machine for an individual user can take account of behaviors of other users, but it must also take account of behavior by this specific user so at least some of the parameters must be estimated from data specific to this user, not necessarily all of them. Q. So a user-specific learning 263 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 Q. And then -A. So a page that would have zero probability in this case would have an infinite value. Q. Okay. But what -- I don't understand what the -A. The negative -- the logarithm of zero is minus infinity. And so if you take minus the minus infinity, it becomes positive infinity. Q. You agree that a probability can't be a negative number? A. That's right. Q. So -A. It also cannot be infinite. Q. Does the fact that the -- the TIJ is calculated based on a probability distribution of pages based on collecting the visiting patterns of many users affect your view of whether the -- the variable TIJ is a parameter of a learning machine or user model? A. No, not really. The -- the fact that there are many -- information is collected about many users means that it's less often the probability of a particular transition will be 265 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 machine must at least have some parameters that are specific to that user? A. Yes. Q. What if the parameters that are specific to that user -- well, let me give you -- let me give you an example of something. If a -- the system creates a parameter for the user interest in sports and it determines by the fact that users -- in reference to all users, that if you've clicked on sports pages five times, that that indicates that you should get a weight of .5 for the variable interest in sports. Would that -would that be a user-specific parameter? MS. BENNETT: Objection. Form. THE WITNESS: So how did you determine that it should have a weight of .5? BY MR. PERLSON: Q. Because you look to see it was assigned based on the activity of all users, that if in observing all the users, they saw that if a user clicked on sports pages five times, that an appropriate weight was .5. A. Okay. And then this specific user also clicked on it exactly five times? 67 (Pages 262 to 265) Veritext Corporate Services 800-567-8658 973-410-4040 298 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 THE WITNESS: No. Estimating the parameters is -- always means estimating the values or weights with those parameters. It is the case, as I mentioned in answers to both of you, that the field uses the word "parameter" more loosely sometimes to mean the variables of the -- and sometimes to mean the values. And Refuah does that as well because -- and the way that is consistent with the Court's construction is consistent with my report and is consistent with the claim language is "parameters" mean the values that are being estimated. BY MR. PERLSON: Q. Right. And if you look at 1E, it refers to estimating a probability PUD that an unseen document D is of interest to the user U. Then it goes on to say, "wherein the probability PUD is estimated by applying the identified properties of the document to the learning machine having the parameters defined buy the User Model." Do you see that? A. Yes. Q. So that requires that the learning machine must actually have the values of the variables that are defined by the user model; 300 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 ACKNOWLEDGMENT OF DEPONENT I do hereby acknowledge that I have read and examined the foregoing of the transcript of my deposition and that: (Check appropriate box): ( ) the same is a true, correct and complete transcription of the answers given by me to the questions therein recorded. ( ) except for the changes noted in the attached errata sheet, the same is a true, correct and complete transcription of the answers given by me to the questions therein recorded. _________________ __________________________ DATE SIGNATURE 299 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 right? MS. BENNETT: Objection. Form. THE WITNESS: That's right. MR. PERLSON: I don't have any further questions. MS. BENNETT: Okay. And we reserve the right to review the transcript and provide an errata. THE VIDEOGRAPHER: This ends disk number 5 and concludes the testimony of Dr. Jamie Carbonell in the matter of Personalized User Model versus Google. The date is November 27th, 2012. The time is 7:08:47. Off the record. MR. FRIEDMAN: Ms. Satkin, you did a stellar job. (Signature not waived.) (Whereupon, at 7:08 p.m., the deposition was concluded.) - - - - - 301 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 CERTIFICATE OF NOTARY PUBLIC I, Paula G. Satkin, the officer before whom the foregoing proceedings were taken, do hereby certify that the witness whose testimony appears in the foregoing proceeding was duly sworn by me; that the testimony of said witness was taken by me in stenotype and thereafter reduced to typewriting under my direction; that said proceedings is a true record of the testimony given by said witness; that I am neither counsel for, related to, nor employed by any of the parties to the action in which these proceedings were taken; and, further, that I am not a relative or employee of any attorney or counsel employed by the parties hereto, nor financially or otherwise interested in the outcome of the action. My commission expires November 14, 2015. 23 24 25 ________________________________ PAULA G. SATKIN Notary Public in and for the District of Columbia 76 (Pages 298 to 301) Veritext Corporate Services 800-567-8658 973-410-4040 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 ERRATA SHEET VERITEXT CORPORATE SERVICES 800-567-8658 ASSIGNMENT NO. CS1565706 CASE NAME: Personalized User Model v. Google DATE OF DEPOSITION: 11/27/2012 WITNESS' NAME: Jaime Carbonell PAGE/LINE(S)/ CHANGE REASON ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ____/_______/_________________/__________ ________________________ Jaime Carbonell 21 22 SUBSCRIBED AND SWORN TO BEFORE ME THIS______DAY OF_______________, 2012. 23 24 25 _______________________ NOTARY PUBLIC MY COMMISSION EXPIRES__________________ Veritext Legal Solutions 25B Vreeland Road - Suite 301 Florham Park, New Jersey 07932 Toll Free: 800-227-8440 Fax: 973-629-1287 _____________, 2012 To: JENNIFER BENNETT, Esq. Case Name: Personalized User Model v. Google Veritext Job Number: 1565706 Witness: Jaime Carbonell Deposition Date: 11/27/2012 Dear Ms. Bennett: Enclosed please find a deposition transcript. Please have the witness review the transcript and note any changes or corrections on the included errata sheet, indicating the page, line number, change, and the reason for the change. Have the witness’ signature at the bottom of the sheet notarized and forward errata sheet back to us at the address shown above. If the jurat is not returned within thirty days of your receipt of this letter, the reading and signing will be deemed waived. Sincerely, Production Department Encl. Cc: DAVID PERLSON, Esq. 77 (Pages 302 to 303) Veritext Corporate Services 800-567-8658 973-410-4040

Disclaimer: Justia Dockets & Filings provides public litigation records from the federal appellate and district courts. These filings and docket sheets should not be considered findings of fact or liability, nor do they necessarily reflect the view of Justia.


Why Is My Information Online?