`
`Paper No. ___
`Filed: May 21, 2021
`
`
`
`
`
`
`
`
`
`
`UNITED STATES PATENT AND TRADEMARK OFFICE
`_____________________________
`
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`
`_____________________________
`
`AMAZON.COM, INC.; AMAZON.COM LLC; AMAZON WEB
`SERVICES, INC.; A2Z DEVELOPMENT CENTER, INC. D/B/A LAB126;
`RAWLES LLC; AMZN MOBILE LLC; AMZN MOBILE 2 LLC;
`AMAZON.COM SERVICES, INC. F/K/A AMAZON FULFILLMENT
`SERVICES, INC.; and AMAZON.COM SERVICES LLC (formerly
`AMAZON DIGITAL SERVICES LLC),
`Petitioners,
`
`v.
`
`VB ASSETS, LLC,
`Patent Owner.
`_____________________________
`
`Case IPR2020-01367
`Patent No. 8,073,681
`_____________________________
`
`
`PATENT OWNER’S RESPONSE
`PURSUANT TO 37 C.F.R. § 42.120
`
`
`
`B.
`
`INTRODUCTION ........................................................................................... 1
`I.
`PERSON OF ORDINARY SKILL IN THE ART .......................................... 3
`II.
`III. GROUND 1: CLAIMS 1-2, 9, 13-14, 21, 25-26, AND 33 ARE NOT
`RENDERED OBVIOUS BY KENNEWICK ................................................. 4
`A.
`The petition fails to establish that Kennewick renders obvious
`“the utterance includes one or more words that have different
`meanings in different contexts” ............................................................. 4
`The petition fails to establish that Kennewick renders obvious
`“short-term shared knowledge” ............................................................. 7
`1. Automated speech recognition (“ASR”), Natural Language
`Understanding (“NLU”), and User Modeling (“UM”) .................. 8
`2. Kennewick’s tagged recognized words are not short-term
`shared knowledge ......................................................................... 12
`IV. GROUNDS 2 THROUGH 6: CLAIMS 3-8, 10-12, 15-20, 22-24, 27-
`32, AND 34-36 ARE NOT OBVIOUS ......................................................... 18
`CONCLUSION .............................................................................................. 19
`
`V.
`
`i
`
`
`
`I.
`
`INTRODUCTION
`Amazon.com, Inc., et al. (“Petitioners”) filed a petition for inter partes
`
`review of claims 1-42 of U.S. Patent No. 8,073,681 (the “’681 patent,” EX1001).
`
`The Board issued its decision instituting trial (“Decision,” Paper 7) on February 4,
`
`2021, finding that Petitioners had established a reasonable likelihood of success on
`
`only claims 37, 39, and 41. VB Assets, LLC (“Patent Owner”) hereby requests
`
`that the Board now issue a final written decision rejecting all grounds of challenge
`
`that are not moot and to confirm that claims 1-36 are not unpatentable.1
`
`As already recognized by the Board, the petition’s arguments with respect to
`
`the three remaining independent claims 1, 13, and 25 are flawed at least because
`
`the petition fails to identify prior art disclosure of interpreting one or more words
`
`that have “different meanings in different contexts.” Petitioners rely on two
`
`disclosures in the Kennewick application. In the first, Kennewick teaches that the
`
`word “temperature” within different dialogs can indicate different contexts—e.g.
`
`the context of weather or temperature. Kennewick never says there is an
`
`
`1 Patent Owner disclaimed claims 37-42 by filing a Statutory Disclaimer under
`
`37 C.F.R. § 1.321(a) on May 17, 2021, a copy of which is submitted as Exhibit
`
`2005. This response addresses the grounds of unpatentability not rendered moot
`
`by Patent Owner’s disclaimer.
`
`1
`
`
`
`ambiguity in the word “temperature” within a single utterance. Moreover, as the
`
`Board already recognized, the word “temperature” has the same meaning whether
`
`referring to body temperature or outdoor temperature—the “measurement of
`
`warmth or coolness.” Decision, 14. Similarly, as the Board also understood in the
`
`institution decision, Petitioners’ reliance on Kennewick’s discussion of the term
`
`“flight one hundred and twenty t[w]o” is misplaced because that term is only
`
`disclosed in a single context. Id., 14-15. This shortcoming is fatal to Petitioners’
`
`challenge to all claims remaining in the ’681 patent.
`
`The Board need look no further than the deficiency already recognized in the
`
`institution decision to reject Petitioners’ challenge at this stage as well.
`
`Nevertheless, the petition suffers from an additional significant flaw. Petitioners
`
`argue that Kennewick’s tagged recognized words constitute short-term shared
`
`knowledge. Yet this collection of recognized words tagged with the user’s identity
`
`has not been interpreted by the system. This is fatal to Petitioners’ case because
`
`the ’681 patent discloses short-term shared knowledge is accumulated using
`
`recognized words as input and then interpreting those words to extract shared
`
`expectations and assumptions in order to determine a user’s intent. The limitations
`
`of mere recognized words preclude them from being short-term shared knowledge
`
`since they impart no meaning on the utterances—limitations that Petitioners’ own
`
`expert, Dr. Smyth, acknowledged in both his declaration and at his deposition.
`
`2
`
`
`
`EX1002, ¶42 (Recognized words alone are “not sufficient for an algorithm to
`
`understand the user’s goals and intent.”); EX2003, 88:16-24 (“Q: Does
`
`Kennewick’s dialog history keep track of the user’s goals and preferences? A: In
`
`the sense that the dialog history is a tagged set of recognized words and phrases
`
`that the user spoke earlier in the conversation, then, no.”).
`
`Accordingly, the Board should reject the grounds of challenge not mooted
`
`by Patent Owner’s disclaimer and, consistent with the preliminary finding in the
`
`institution decision, conclude that claims 1-36 have not been shown to be
`
`unpatentable.
`
`II.
`
`PERSON OF ORDINARY SKILL IN THE ART
`Petitioners define the level of a person of ordinary skill in the art (“POSA”)
`
`at the time of the invention of the ’681 patent as someone who “would have at least
`
`a Bachelor-level degree in computer science, computer engineering, electrical
`
`engineering, or a related field in computing technology, and two years of
`
`experience with automatic speech recognition and natural language understanding,
`
`or equivalent education, research experience, or knowledge.” Pet., 4-5; EX1002,
`
`¶30. Patent Owner and its expert, Dr. Gershman, have also applied the same
`
`definition this response. EX2001, ¶¶19-20.
`
`3
`
`
`
`III. GROUND 1: CLAIMS 1-2, 9, 13-14, 21, 25-26, AND 33 ARE NOT
`RENDERED OBVIOUS BY KENNEWICK
`The petition fails to establish that claims 1-2, 9, 13-14, 21, 25-26, and 33 are
`
`rendered obvious by Kennewick alone or in combination with the knowledge of a
`
`POSA. EX2001, ¶¶21-54. First, as the Board has already recognized, the petition
`
`fails to establish that Kennewick discloses an utterance containing one or more
`
`words that have different meanings in different contexts. EX2001, ¶¶22-31.
`
`Second, the petition fails to establish that Kennewick discloses short-term shared
`
`knowledge. EX2001, ¶¶32-54.
`
`A. The petition fails to establish that Kennewick renders obvious
`“the utterance includes one or more words that have different
`meanings in different contexts”
`The petition fails to show that Kennewick discloses limitation 1.a. of the
`
`’681 patent, which recites “receiving an utterance at a voice input device during a
`
`current conversation with a user, wherein the utterance includes one or more words
`
`that have different meanings in different contexts.” EX2001, ¶22. This limitation
`
`recites a specific type of ambiguity where “one or more words … have different
`
`meanings in different contexts.” EX2001, ¶23. However, as noted in the
`
`institution decision, the petition cites to no disclosure in Kennewick describing this
`
`specific type of ambiguity. Id.; Decision, 14. Instead, Petitioners cite a
`
`“temperature” example that does not have two different meanings, and a “flight”
`
`4
`
`
`
`example that relates to determining which of two different words that sound the
`
`same is the correct word in a single context. EX2001, ¶¶24-30.
`
`The petition first relies on Kennewick’s teachings that an utterance
`
`containing “temperature” can “impl[y] a context value of weather for the question”
`
`and that “within a different dialog, the keyword ‘temperature’ can imply a context
`
`for a measurement.” Pet., 16 (quoting EX1003, ¶160); EX2001, ¶24. At the
`
`outset, this disclosure in Kennewick is not talking about a single utterance
`
`containing one or more words that have different meanings in different contexts.
`
`EX2001, ¶25. Instead, Kennewick explicitly says that the keyword “temperature”
`
`can imply a different context within different dialogs. EX1003, ¶160. Kennewick
`
`never says it recognizes different meanings for the word “temperature,” and even
`
`when it says “temperature” can indicate different contexts, it explicitly says that is
`
`for “different dialog[s]”—that is, expressly not a single utterance. Id.; EX2001,
`
`¶25. Thus, the petition’s assertion that Kennewick teaches that “‘temperature’
`
`could be interpreted as an outdoor temperature or as a body temperature” is not
`
`true. Pet., 16; EX2001, ¶25.
`
`Moreover, the petition still fails to establish two different meanings of
`
`“temperature.” EX2001, ¶¶26-27. The petition asserts that “‘temperature’ could
`
`be interpreted as an outdoor temperature or as a body temperature depending on
`
`the context.” Pet., 16; see also EX1002, ¶69. But “outdoor temperature” and
`
`5
`
`
`
`“body temperature” do not involve different meanings of “temperature.” EX2001,
`
`¶26. The additional qualifiers “outdoor” and “body” do not change the meaning of
`
`“temperature,” that is, the degree of hotness or coldness. EX2001, ¶27. Instead,
`
`they simply indicate the object whose temperature is being measured. Id. As the
`
`Board correctly concluded in the institution decision, “the word ‘temperature’ has
`
`the same meaning in both of these examples” because “the word ‘temperature’
`
`refers to a measurement of warmth or coolness.” Decision, 14.
`
`The second example Petitioners rely on relates to the question “what about
`
`flight one hundred and twenty too.” Pet., 16–17 (citing EX1003, ¶163); EX2001,
`
`¶28. Kennewick explains: “The parser and domain agent use flight information in
`
`the database and network information along with context to determine the most
`
`plausible interpretation among; flight 100 and flight 20 also, flight 100 and flight
`
`22, flight 122, etc.” EX1003, ¶163. The petition asserts that this teaches one
`
`context “in which ‘flight 122’ exists” and a different context “where the system
`
`just listed several flight options including ‘flight 100’ and ‘flight 20.” Pet., 16-17.
`
`However, Kennewick does not say that there are multiple contexts. EX2001, ¶29.
`
`At most, Kennewick indicates interpreting “flight one hundred and twenty too” in a
`
`single flight information context. Id. As the Board correctly noted in finding no
`
`reasonable likelihood of unpatentability, “Dr. Smyth does not provide any support
`
`6
`
`
`
`for his assertion that these hypothetical situations would constitute different
`
`contexts.” Decision, 15; EX2001, ¶29.
`
`Notably, this passage of Kennewick describes formulating a question or
`
`command after the context has already been determined and the appropriate agent
`
`has been invoked. EX2001, ¶30. The preceding paragraph indicates “the context
`
`for the question or command has been determined” (EX1003, ¶162 (emphasis
`
`added)) and the following disclosure relates to formulating a question or command
`
`for processing by the agent (EX1003, ¶¶162-63). EX2001, ¶30. In other words,
`
`Kennewick teaches that it disambiguates the phrase “flight one hundred and twenty
`
`too” after a single context has already been established. Thus, to the extent there
`
`are multiple meanings of the phrase “flight one hundred and twenty too,” they are
`
`multiple meanings within the same context. Id. Accordingly, Petitioners have
`
`failed to show that “flight one hundred and twenty too” has different meanings in
`
`different contexts. Id.
`
`Thus, the petition fails to establish that Kennewick discloses this element or
`
`renders it obvious. EX2001, ¶31.
`
`B.
`
`The petition fails to establish that Kennewick renders obvious
`“short-term shared knowledge”
`While, as noted in the institution decision, Petitioners’ failure to demonstrate
`
`Kennewick discloses the “different meanings in different contexts” limitation is
`
`enough on its own to reject the ground of challenge, the petition also falls short in
`
`7
`
`
`
`another respect. Petitioners’ theory that Kennewick’s tagged recognized words are
`
`“short-term shared knowledge [that] includes knowledge about the utterance
`
`received during the current conversation” is flawed because both experts agree that
`
`mere recognized words alone are not sufficient for a system to understand a user’s
`
`preferences, goals, or intent. EX2001, ¶¶32-54.
`
`In order to understand the flaws in the petition’s case, some background is
`
`provided below. EX2001, ¶33. As acknowledged in Dr. Smyth’s declaration,
`
`natural language conversational systems typically have multiple, distinct
`
`components, including an automatic speech recognition (ASR) component, a
`
`natural language understanding (NLU) component, and a user modeling (UM)
`
`component. EX1002, ¶¶39-42; see also EX2001, ¶¶34-40. Because Kennewick
`
`discloses that its tagged recognized words are generated by an ASR component,
`
`and because the ’681 patent discloses that short-term shared knowledge is
`
`generated by an NLU component, it is important to understand the differences
`
`between ASR and NLU. EX2001, ¶33. With that understanding in mind, it
`
`becomes evident that Kennewick’s tagged recognized words fail to meet the
`
`recited short-term shared knowledge. Id.
`
`1.
`
`Automated speech recognition (“ASR”), Natural Language
`Understanding (“NLU”), and User Modeling (“UM”)
`To understand the flaws with the petition’s mapping of the claim elements, it
`
`is important to understand the key components of natural language conversational
`
`8
`
`
`
`systems at the time of the ’681 patent. EX2001, ¶34. As explained below, and as
`
`both experts agree, these components include automatic speech recognition
`
`(“ASR”) (which converts a user’s utterance into recognized words), natural
`
`language understanding (“NLU”) (which interprets the recognized words), and
`
`user modeling (“UM”) (which uses the interpreted words to create representations
`
`of knowledge about the user). Id.
`
`Petitioners’ expert Dr. Smyth devotes several pages to describing how
`
`automated speech recognition (ASR) and natural language understanding (NLU)
`
`are distinct components. EX1002, ¶¶39-42; EX2001, ¶¶34-38. According to Dr.
`
`Smyth, “ASR involves recognizing words from spoken audio for each user
`
`utterance” whereas “NLU is the process of determining the meaning of the
`
`recognized words in the utterance.” EX1002, ¶39. Since “recognized words alone
`
`are typically not sufficient for an algorithm to understand the user’s goals and
`
`intent,” it is up to the NLU component to “interpret the meaning of the recognized
`
`words.” Id., ¶42; see also EX2003, 18:11-13 (“If a speech recognition engine
`
`produces recognized words, then NLU can be used to try to understand what those
`
`words mean.”). As Dr. Smyth testified, ASR can “recognize the word Portland”
`
`but is not capable of determining the meaning of the word—it is “the role of the
`
`NLU component” to derive meaning. EX2003, 32:7-14; see also 34:9-12 (“ASR
`
`broadly refers to technologies that take acoustic waveforms and map them to
`
`9
`
`
`
`words, and NLU broadly refers to technologies that try to infer meaning from those
`
`words.”).
`
`Dr. Smyth also testifies that the NLU component extracts knowledge out of
`
`the interpreted words. EX2001, ¶37. For example, Dr. Smyth notes that the NLU
`
`component interprets the phrase “I would like to fly to New York” to extract such
`
`knowledge like the request being one “to find a flight” and “that the flight is for the
`
`individual who is speaking.” EX1002, ¶42. Furthermore, Dr. Smyth notes that the
`
`NLU component “would need to understand that ‘New York’ is the name of a
`
`location, and that it is not the name of specific airport but is a city with multiple
`
`airports associated with it.” Id. (emphasis added). Dr. Smyth makes clear that it is
`
`“NLU technologies” which “handle such situations,” and not the ASR component.
`
`Id.; EX2001, ¶37.
`
`The distinction between ASR and NLU further extends to the fact that they
`
`are often used separately. EX2001, ¶¶35, 38. For example, ASR components can
`
`be used on their own for speech dictation applications such as assisting disabled
`
`persons in transcribing messages. Id., ¶38. Similarly, NLU systems are used to
`
`interpret written text, in which case an ASR component is not necessary. Id.; see
`
`also EX2003, 19:3-4 (“It’s fair to say NLU can be used outside the scope of speech
`
`recognition.”); 26:12-15 (“[I]f you’re focusing on an NLU research project, it
`
`doesn’t make sense to design an ASR system from scratch.”).
`
`10
`
`
`
`Another component that Petitioners’ expert Dr. Smyth discusses is the User
`
`Modeling (UM) component, which allows a system to have “access to specific
`
`knowledge about a user (such as their travel preferences)” by UM methods that
`
`“build an internal representation of a user.” EX1002, ¶44; see also EX2003,
`
`40:23-24 (“[A] user model is … some representation of information or knowledge
`
`about a user.”); EX2001, ¶39. As Dr. Smyth acknowledges, the user model can be
`
`built from previous user utterances such as “a user’s travel preferences from words
`
`in past conversations that a user has had with the system, or inferences from words
`
`in earlier utterances in the current conversation.” EX1002, ¶44. In order to
`
`represent knowledge about a user extracted from an utterance, it would require
`
`extracting meaning from the user’s utterances as performed by the NLU
`
`component, not the ASR component. EX2001, ¶39; see EX1002, ¶42
`
`(“[R]ecognized words alone are typically not sufficient for an algorithm to
`
`understand the user’s goals and intent.”). Once the knowledge about the user’s
`
`travel preferences is extracted by the NLU, the UM component can use it to build a
`
`user model. EX2001, ¶39.
`
`As discussed below, Dr. Smyth admits that Kennewick’s dialog history,
`
`which consists of tagged recognized words, is created by an ASR component.
`
`However, the ’681 patent’s description of short-term shared knowledge
`
`necessitates interpretation of the words, which would be performed by an NLU
`
`11
`
`
`
`component using the recognized words as input. EX2001, ¶40. This distinction
`
`proves fatal to the petition’s theory.
`
`2. Kennewick’s tagged recognized words are not short-term
`shared knowledge
`The petition fails to establish that Kennewick discloses or renders obvious
`
`“short-term shared knowledge” because Kennewick’s tagged recognized words are
`
`not interpreted and therefore do not constitute short-term shared knowledge.
`
`EX2001, ¶¶41-54.
`
`The petition begins its analysis of this limitation by discussing Kennewick’s
`
`disclosure of “recognize[d] words and phrases” of a user’s utterance that are
`
`tagged “with [the user’s] identity in all further processing.” Pet., 17 (quoting
`
`EX1003, ¶155); EX2001, ¶42. Specifically, Kennewick discloses that “speech
`
`recognition engine 120 may recognize words and phrases” and then adds the user
`
`identity tags. EX1003, ¶155. The petition asserts that “Kennewick’s system of
`
`tagging recognized words and phrases of each utterance with a user identity allows
`
`the system to build a dialog history of words and phrases uttered by that user
`
`during the conversation.” Pet., 18. The petition thus asserts that this dialog history
`
`containing the tagged recognized words is the claimed accumulation of “short-term
`
`shared knowledge about the current conversation.” Id.
`
`As an initial matter, the petition fails to explain how tagged recognized
`
`words would constitute any sort of knowledge about the utterance received.
`
`12
`
`
`
`EX2001, ¶43. Indeed, the petition does not have any discussion as to what
`
`“knowledge about the utterance” the Petitioners believe are reflected in
`
`Kennewick’s tagged recognized words. Id. Nor does Dr. Smyth, whose analysis is
`
`identical to the petition’s. See EX1002, ¶¶71-73; EX2001, ¶43.
`
`Indeed, tagged recognized words do not constitute short-term shared
`
`knowledge about the received utterance as described and claimed in the ’681
`
`patent and as understood by ordinary artisans. EX2001, ¶¶44-46. Instead, the ’681
`
`patent explains that short-term shared knowledge is built by the Session Input
`
`Accumulator using recognized words as an input, whereupon assumptions and
`
`expectations are extracted from the utterances to build short-term knowledge. Id.
`
`In other words, the Session Input Accumulator is part of an NLU component, not
`
`an ASR component, and therefore mere recognized words do not constitute short-
`
`term shared knowledge. Id. While the ’681 patent does acknowledge that short-
`
`term shared knowledge can include “conversation history,” as explained below,
`
`unlike Kennewick’s tagged recognized words, the ’681 patent’s “conversation
`
`history” is a history of interpreted utterances. Id.
`
`Specifically, the ’681 patent explains that “shared knowledge,” which
`
`includes “both short-term and long-term knowledge,” “may enable a user and a
`
`voice user interface to share assumptions and expectations … that facilitate a
`
`cooperative conversation between human users and a system.” EX1001, 4:51-53,
`
`13
`
`
`
`5:25-30; EX2001, ¶46. Shared knowledge is built by the Session Input
`
`Accumulator. EX2001, ¶48; EX1001, 13:44, 15:39-40 (“Intelligent Hypothesis
`
`Builder 310 may use short-term shared knowledge from the Session Input
`
`Accumulator.”). The Session Input Accumulator accumulates a number of inputs
`
`upon which the short-term shared knowledge is built, including “recognition text
`
`for each utterance, a recorded speech file for each utterance, a list-item selection
`
`history, a graphical user interface manipulation history, or other input data.”
`
`EX1001, 13:47-49. The Session Input Accumulator uses the extracted knowledge
`
`to “build shared knowledge models.” Id., 13:44; EX2001, ¶48. The Session Input
`
`Accumulator then passes the shared knowledge to the “Intelligent Hypothesis
`
`Builder 310” which “may use short-term knowledge to generate one or more
`
`implicit hypotheses of a user's intent when an utterance may be missing required
`
`qualifiers or other information needed to complete a request or task.” EX1001,
`
`15:65-16:2 (emphasis added); EX2001, ¶48. These hypotheses about a user’s
`
`intent are graded and used by the Adaptive Response Builder 315 to generate an
`
`adapted natural language response. EX2001, ¶48; EX1001, 16:60-17:20. To
`
`summarize, the Session Input Accumulator receives input such as recognized
`
`words of an utterance, interprets those words to build shared knowledge, and
`
`passes such knowledge to the Intelligent Hypothesis Builder for building
`
`hypotheses about user intent that is used to form adapted responses. EX2001, ¶48.
`
`14
`
`
`
`The fact that recognized words are used as an input for building shared
`
`knowledge is supported by the ’681 patent’s explanation that shared knowledge
`
`consists of “assumptions and expectations such as topic knowledge, conversation
`
`history, word usage, jargon, tone (e.g., formal, humorous, terse, etc.), or other
`
`assumptions and/or expectations that facilitate interaction at a Human-to-Machine
`
`interface.” EX1001, 14:26-30 (emphasis added); EX2001, ¶¶44-46. In other
`
`words, shared knowledge allows participants in a conversation to determine each
`
`other’s intent and provide meaningful responses. EX2001, ¶¶46-47. For example,
`
`if a speaker says, “I had a burger last night, and it was so good, I left a big tip,”
`
`there is at least shared knowledge between the two speakers that the presence of a
`
`“tip” indicates that the burger was consumed at a sit-down restaurant (as opposed
`
`to a fast-food establishment or a home-made meal). Id. Thus, the
`
`assumption/expectation of a sit-down restaurant could result in the other speaker
`
`asking, “Which restaurant was it?” or “How was the service?” Id. However,
`
`without such shared, tacit knowledge about restaurants, the dialog would be
`
`incomprehensible. Id. The shared knowledge is an integral part of the
`
`conversation. Id.
`
`Thus, because shared knowledge consists of shared assumptions and
`
`expectations, and because the building of shared knowledge is performed by an
`
`NLU component that uses recognized words as input, Kennewick’s tagged
`
`15
`
`
`
`recognized words are not shared knowledge. EX2001, ¶49. Dr. Smyth’s own
`
`testimony confirms that mere tagged recognized words do not constitute short-term
`
`shared knowledge about a conversation. Id. As he explains it, recognized words
`
`are “not sufficient for an algorithm to understand the user’s goals and intent.”
`
`EX1002, ¶42; see also EX2003, 68:16-24 (“Q: Does Kennewick’s dialog history
`
`keep track of the user’s goals and preferences? A: In the sense that the dialog
`
`history is a tagged set of recognized words and phrases that the user spoke earlier
`
`in the conversation, then, no.”); cf. EX1001, 14:25-30 (“[A] user and a voice user
`
`interface may share assumptions and expectations such as topic knowledge …”).
`
`Since Kennewick’s tagged recognized words do not provide the system with
`
`understanding of the user’s goals, intent, or preferences, they do not provide the
`
`system with short-term shared knowledge about the conversation (i.e. assumptions
`
`and expectations about the conversation). EX2001, ¶49. Indeed, Kennewick’s
`
`tagged recognized words do not reflect any meaning of the user’s utterance. Id.;
`
`EX2003, 34:9-12 (“ASR broadly refers to technologies that take acoustic
`
`waveforms and map them to words, and NLU broadly refers to technologies that
`
`try to infer meaning from those words.”).2
`
`
`2 Kennewick’s tagged recognized words are not strictly the recognized words
`
`from the ASR component, as the recognized words are also tagged with the user
`
`
`
`16
`
`
`
`The petition argues that “Kennewick’s description of building dialog history
`
`by tagging dialog utterances is consistent with the ’681 patent’s disclosure of
`
`‘short-term knowledge’ including a ‘history of the conversation’ built ‘as the
`
`conversation progresses.’” Pet., 18 (quoting EX1001, 15:65-16:7, 15:38-42);
`
`EX2001, ¶¶51-52. However, the ’681 patent’s “conversation history,” which is
`
`built by an NLU component, is clearly not the same as Kennewick’s tagged
`
`recognized words, which is built by an ASR component. EX2001, ¶52.
`
`Kennewick’s tagged recognized words are built by its speech recognition engine,
`
`an ASR component. EX1003, ¶155; EX2003, 55:3-6 (“[A]s described by
`
`Kennewick, speech recognition engine has the characteristics of an ASR
`
`component in its conversational interface system.”). But as the petition
`
`acknowledges, the ’681 patent teaches that it is the “Session Input Accumulator”
`
`
`identity. EX1003, ¶155; EX2001, ¶50. However, the tags do not change the fact
`
`that the words are not interpreted and therefore the system has not obtained any
`
`knowledge about, e.g., a user’s goals, preferences, intents, or assumptions from the
`
`utterances. EX2001, ¶50; see also EX2003, 88:17-24 (“Q: Does Kennewick’s
`
`dialogue history keep track of the user’s goals and preferences? A: In the sense that
`
`the dialogue history is a tagged set of recognized words and phrases that the user
`
`spoke earlier in the conversation, then, no.”).
`
`17
`
`
`
`which “build[s] a history of the conversation.” EX1001, 16:6-7. Yet as explained
`
`above, the Session Input Accumulator receives as input “recognition text for each
`
`utterance” in order to build the conversation history. Id., 13:47-51; EX2001, ¶52.
`
`Moreover, the ’681 patent makes clear that “shared knowledge 305” includes
`
`“share[d] assumptions and expectations such as … conversation history.”
`
`EX1001, 14:24-27 (emphasis added). In other words, just like a typical NLU
`
`component, the Session Input Accumulator receives recognized text as input,
`
`interprets the text, and builds shared assumptions and expectations (i.e. shared
`
`knowledge), which includes the conversation history. EX2001, ¶52. Mere
`
`recognized words generated by an ASR component are insufficient to constitute
`
`shared assumptions and expectations. EX2001, ¶¶52-53; EX1002, ¶42; see also
`
`EX2003, 88:17-24. Thus, the petition is wrong to equate the building of
`
`Kennewick’s tagged recognized words by an ASR component with the building of
`
`the ’681 patent’s conversation history by an NLU component. EX2001, ¶¶52-53.
`
`For these reasons, Kennewick’s tagged recognized words are not “short-term
`
`shared knowledge.” EX2001, ¶54. Accordingly, the petition fails to establish that
`
`this limitation is disclosed or rendered obvious by Kennewick.
`
`IV. GROUNDS 2 THROUGH 6: CLAIMS 3-8, 10-12, 15-20, 22-24, 27-32,
`AND 34-36 ARE NOT OBVIOUS
`The claims (that were not disclaimed) challenged in Grounds 2 through 6 of
`
`the petition are all dependent of claims 1, 13, and 25. The petition’s analysis for
`
`18
`
`
`
`all the challenged claims in Grounds 2 through 6 rely on its analysis of claims 1,
`
`13, and 25 in Ground 1. See, e.g., Pet., 32 (“As described in Section IX.A,
`
`Kennewick renders obvious independent claims 1, 13, and 25 … This section
`
`identifies additional disclosures of Huang that render these dependent claims
`
`obvious.”). Thus, these grounds of challenge fail for the same reasons discussed
`
`above. EX2001, ¶55.
`
`V. CONCLUSION
`For at least the reasons set forth above, as well as those noted in the Board’s
`
`institution decision, Petitioners have failed to meet their burden and the challenged
`
`claims should be found not unpatentable.
`
`
`
`
`
`Date: May 21, 2021
`
`
`
`
`
`Respectfully submitted,
`
`
`
`/ Matthew A. Argenti /
`Matthew A. Argenti, Lead Counsel
`Reg. No. 61,836
`
`
`
`19
`
`
`
`CERTIFICATE OF COMPLIANCE
`
`Pursuant to §42.24(d), the undersigned certifies that this paper contains no
`
`more than 14,000 words, not including the portions of the paper exempted by
`
`§42.24(b). According to the word-processing system used to prepare this paper, the
`
`paper contains 4,096 words.
`
`
`
`
`
`Date: May 21, 2021
`
`
`
`
`
`Respectfully submitted,
`
`
`
`/ Matthew A. Argenti /
`Matthew A. Argenti, Lead Counsel
`Reg. No. 61,836
`
`
`
`20
`
`
`
`LIST OF EXHIBITS
`
`Exhibit No.
`
`Description
`
`2001
`
`2002
`
`2003
`
`2004
`
`2005
`
`Declaration of Anatole Gershman
`
`Curriculum Vitae of Anatole Gershman
`
`Transcript of May 4, 2021 deposition of Padhraic Smyth, Ph.D.
`
`Intentionally Left Blank
`
`Disclaimer in Patent Under 37 C.F.R. 1.32(a), filed May 17,
`2021 for U.S. Patent No. 8,073,681
`
`
`
`
`
`
`
`21
`
`
`
`CERTIFICATE OF SERVICE
`
`I certify that the foregoing Patent Owner’s Response Pursuant to 37 C.F.R. §
`
`42.120 and Exhibits 2001-2005 were served on this 21st day of May, 2021, on the
`
`Petitioners at the correspondence address of the Petitioners as follows:
`
`J. David Hadden
`Saina Shamilov
`Brian Hoffman
`Johnson K. Kuncheria
`Allen Wang
`FENWICK & WEST LLP
`VBAssets-IPR@fenwick.com
`
`
`
`
`
`
`Date: May 21, 2021
`
`
`
`
`
`
`Respectfully submitted,
`
`
`
`/ Matthew A. Argenti /
` Matthew A. Argenti, Lead Counsel
` Reg. No. 61,836
`
`
`
`22
`
`
Accessing this document will incur an additional charge of $.
After purchase, you can access this document again without charge.
Accept $ Charge
Still Working On It
This document is taking longer than usual to download. This can happen if we need to contact the court directly to obtain the document and their servers are running slowly.
Give it another minute or two to complete, and then try the refresh button.
A few More Minutes ... Still Working
It can take up to 5 minutes for us to download a document if the court servers are running slowly.
Thank you for your continued patience.
This document could not be displayed.
We could not find this document within its docket. Please go back to the docket page and check the link. If that does not work, go back to the docket and refresh it to pull the newest information.
Your account does not support viewing this document.
You need a Paid Account to view this document. Click here to change your account type.
Your account does not support viewing this document.
Set your membership
status to view this document.
With a Docket Alarm membership, you'll
get a whole lot more, including:
- Up-to-date information for this case.
- Email alerts whenever there is an update.
- Full text search for other cases.
- Get email alerts whenever a new case matches your search.
One Moment Please
The filing “” is large (MB) and is being downloaded.
Please refresh this page in a few minutes to see if the filing has been downloaded. The filing will also be emailed to you when the download completes.
Your document is on its way!
If you do not receive the document in five minutes, contact support at support@docketalarm.com.
Sealed Document
We are unable to display this document, it may be under a court ordered seal.
If you have proper credentials to access the file, you may proceed directly to the court's system using your government issued username and password.
Access Government Site
