Case 1:13-cv-00633-DEP Document 41-2 Filed 01/03/14 Page 2 of 17
`
`
`
`IN THE UNITED STATES DISTRICT COURT
`FOR THE NORTHERN DISTRICT OF NEW YORK
`ALBANY DIVISION
`
`
`
`
`
`
`Civil Action No. 1:13-cv-00633-DEP
`
`JURY TRIAL DEMANDED
`
`RENSSELAER POLYTECHNIC
`INSTITUTE AND
`DYNAMIC ADVANCES, LLC,
`
`
`Plaintiffs,
`
`
`APPLE INC.,
`
`
`
`v.
`
`
`
`
`Defendant.
`
`DECLARATION OF JAIME CARBONELL
`
`
`I, Dr. Jaime Carbonell, declare as follows:
`
`1.
`
`I have been retained as a technical expert on behalf of Rensselaer
`
`Polytechnic Institute and Dynamic Advances, LLC.
`
`2.
`
`The following Declaration is based on my personal knowledge and all
`
`facts and statements contained herein are true and accurate to the best of my
`
`knowledge, information, and belief.
`
`3.
`
`In preparing this Declaration, I have reviewed the following materials:
`
`U.S. Patent Number 7,177,798 (the “’798 Patent”) and its file wrapper; and Apple
`
`Inc.’s Petition for Inter Partes Review of U.S. Patent 7,177,798 (the “IPR Petition”)
`
`and all exhibits thereto.
`
`4.
`
`It is my professional opinion that the IPR Petition is based on
`
`fundamentally flawed arguments and inferences. Through the IPR Petition, Apple
`
`argues to the U.S. Patent and Trademark Office that the claims of the ’798 Patent are
`
`Apple Inc. v. Rensselaer Polytechnic Institute
` IPR2014-00320
` RPI Ex. 2005 PAGE 1
`
`

`

`Case 1:13-cv-00633-DEP Document 41-2 Filed 01/03/14 Page 3 of 17
`
`invalid in light of certain prior art references (exhibits to the IPR Petition). Apple’s
`
`invalidity arguments are premised on the prior art references disclosing every claim
`
`limitation of the claims of the ’798 Patents. But it is clear that each of the references –
`
`alone and in combination – fail to disclose certain key limitations common to all claims
`
`of the ’798 Patent. Thus, Apple’s invalidity arguments facially and necessarily fail.
`
`A.
`
`5.
`
`Qualifications
`
`My qualifications for forming the opinions set forth in this report are
`
`summarized here and explained in more detail in my curriculum vitae, which is attached
`
`as Exhibit A. Additionally, in the past 5 years I have been an expert witness on a
`
`number of legal cases, primarily involving intellectual property in the areas of search
`
`engines, information retrieval, software, data mining and text mining.
`
`6.
`
`I graduated from the Massachusetts Institute of Technology in 1975 with
`
`degrees in Physics and Mathematics. I went on to Yale University where I received a
`
`Masters Degree in Computer Science in 1976 and a Ph.D. in Computer Science in
`
`1979.
`
`7.
`
`In 1979, I became an Assistant Professor of Computer Science at
`
`Carnegie Mellon University. I was subsequently promoted to Associate Professor and
`
`then to Full Professor. Since 1995, I have been the Allen Newell Professor of Computer
`
`Science at Carnegie Mellon University. Since 1996, I have also been the Director of the
`
`Language Technologies Institute at Carnegie Mellon University. Last year, I was
`
`appointed “University Professor” at Carnegie Mellon University.
`
`8.
`
`I have published over 300 technical and scientific articles, primarily in
`
`peer-reviewed journals and conferences in multiple computational fields, including:
`
`computer science, computational linguistics, natural language processing, machine
`
`2
`
`Apple Inc. v. Rensselaer Polytechnic Institute
` IPR2014-00320
` RPI Ex. 2005 PAGE 2
`
`

`

`Case 1:13-cv-00633-DEP Document 41-2 Filed 01/03/14 Page 4 of 17
`
`learning, databases, data mining, modeling, information retrieval, search engines,
`
`computational biology, machine translation, mathematical and statistical foundations,
`
`and integrated systems applications. These reflect my active lines of research over the
`
`past 35 years.
`
`9.
`
`My research includes computational methods for analyzing text (typically
`
`called “natural language processing”) in order to organize it, retrieve it, summarize it,
`
`index it, parse it, and translate it. One of my papers describing my invention, Maximum
`
`Marginal Relevance for Retrieval and Summarization, is among the most highly cited
`
`(over 1,200 citations) in the Association for Computing Machinery’s Special Interest
`
`Group in Information Retrieval (ACM-SIGIR), the premier academic conference for
`
`search engines and related research. Among my other highly cited works is “Machine
`
`Learning: An Artificial Intelligence Approach” edited with Michalski and Mitchell (1,700
`
`citations). I have researched mathematical approaches to analyze text, including
`
`statistical machine learning approaches over textual corpus, hand-built
`
`linguistic/heuristic methods, and combinations thereof. If asked, I can further discuss
`
`these areas of research and accomplishments. I have been a researcher in Natural
`
`Language Processing since my first paper on the subject in 1979. I have since
`
`published extensively in the field, as documented in my CV. From 1986 to 1996, I was
`
`the director of CMU’s Center for Machine Translation, which later evolved into the
`
`Language Technologies Institute that I still direct.
`
`10.
`
`I teach courses and seminars in data mining, search engines, electronic
`
`commerce, machine learning, machine translation and aspects of computational biology
`
`at Carnegie Mellon University, mostly at the graduate level. I am also engaged in
`
`3
`
`Apple Inc. v. Rensselaer Polytechnic Institute
` IPR2014-00320
` RPI Ex. 2005 PAGE 3
`
`

`

`Case 1:13-cv-00633-DEP Document 41-2 Filed 01/03/14 Page 5 of 17
`
`designing distance-learning and learning-by-doing curricula, also at the graduate level.
`
`I advise Ph.D. and M.S. students in the above subject areas.
`
`B.
`
`Person of Ordinary Skill in the Art
`
`11.
`
`In my opinion, a person of ordinary skill in the art pertinent to the ’798
`
`Patent at the time its application was filed would have a bachelor of science degree in
`
`computer science or a bachelor of science degree in engineering, mathematics, or
`
`physics with computer science coursework, and either graduate level coursework or 2-3
`
`years of direct technical experience in database systems, and knowledge of natural
`
`language processing from courses or work experience.
`
`C.
`
`Anticipation and Obviousness
`
`12. Standard for Anticipation: I understand that to anticipate a claim, a
`
`reference must disclose each and every limitation of that claim, and that this should be
`
`assessed on a claim-by-claim basis. I also understand that anticipation can occur when
`
`an undisclosed limitation is literally missing, but is present because the prior art
`
`necessarily functions in accordance with, or includes, the undisclosed limitation.
`
`13.
`
`I understand that for a reference to anticipate a patent claim, that
`
`reference must also enable one of ordinary skill in the art to make and use the full scope
`
`of the claimed invention without undue experimentation.
`
`14. Standard for Obviousness: I understand that a combination of prior art
`
`references may render a claim obvious if, at the time of the invention, a person of
`
`ordinary skill in the art would have selected and combined those prior art elements in
`
`the normal course of research and development to yield the claimed invention. I
`
`understand in making the obviousness inquiry, one should consider the so-called
`
`Graham factors: the scope and content of the prior art; the differences between the
`
`4
`
`Apple Inc. v. Rensselaer Polytechnic Institute
` IPR2014-00320
` RPI Ex. 2005 PAGE 4
`
`

`

`Case 1:13-cv-00633-DEP Document 41-2 Filed 01/03/14 Page 6 of 17
`
`claimed inventions and the prior art; the level of ordinary skill in the art; and certain
`
`secondary considerations. I further understand the obviousness analysis is to be
`
`performed on a claim-by-claim basis. I understand that a person of ordinary skill in the
`
`art is a person of ordinary creativity, not an automaton. When there is a design need or
`
`market pressure to solve a problem and there are a finite number of identified,
`
`predictable solutions, a person of ordinary skill has good reason to pursue the known
`
`options within his or her technical grasp.
`
`15.
`
`It is also my understanding that obviousness requires more than a mere
`
`showing that the prior art includes separate references covering each separate
`
`limitation in a claim under examination. I understand obviousness requires the
`
`additional showing that a person of ordinary skill at the time of the invention would have
`
`been motivated to select and combine those prior art elements in the normal course of
`
`research and development to yield the claimed invention. I also understand a fact-
`
`finder should be aware of the distortion caused by hindsight bias and must be cautious
`
`of arguments reliant upon ex post reasoning.
`
`D.
`
`Natural Language Processing and Databases
`
`16. Natural Language Processing (NLP) is a very large well-established
`
`discipline, almost since the start of computer science itself, and I have been an active
`
`researcher and teacher in NLP for over 30 years. NLP encompasses machine
`
`translation, text/web search, text mining, natural language database query, and more.
`
`Despite its long tradition, innovation was strong and steady in NLP in May 2001 (when
`
`the application for the ’798 Patent was filed) and continues at present. For instance,
`
`scientific conferences, each with hundreds of participants and dozens of presentations
`
`of new results abound: Association for Computational Linguistics (ACL), Empirical
`
`5
`
`Apple Inc. v. Rensselaer Polytechnic Institute
` IPR2014-00320
` RPI Ex. 2005 PAGE 5
`
`

`

`Case 1:13-cv-00633-DEP Document 41-2 Filed 01/03/14 Page 7 of 17
`
`Methods in Natural Language Processing (EMNLP), and International Conference on
`
`Computational Linguistics (COLING), to name a few. Patents in NLP continue to be
`
`issued in quantity; for instance, the USPTO lists 120 patents issued since January 1,
`
`2013 with “natural language” in the abstract. The ’798 Patent itself discloses a large
`
`number of issued patents in the area as well as pertinent scientific publications.
`
`17. NLP interfaces for databases was an active sub-area of NLP since the
`
`early days, continued to be so through the filing of the application for the ’798 Patent,
`
`and continues to this date, with steady innovation as evidenced by over 30 issued
`
`patents in this area in the past 2 years. The message should be clear: NLP for
`
`database query is both a fairly substantial area and an actively investigated one with
`
`steady innovation. There are many approaches to NLP in general, and to NLP for
`
`database query in particular. The simple presence of past work in NLP interfaces to
`
`databases is by no means evidence of lack of patentability; instead, the specifics of the
`
`patent claims in question, namely those of the ’798 Patent, must be analyzed in detail
`
`with respect to each alleged prior art.
`
`E.
`
`The ’798 Patent
`
`18.
`
`The ’798 Patent is directed squarely at natural language interfaces to
`
`databases (NLP for databases or NLP-DB for short). Relational databases were
`
`invented by Ted Codd (who received the Turing Award for his invention, the highest
`
`honor in Computer Science), and since then they became the dominant paradigm for
`
`databases, with new variants such as object-oriented and distributed databases. The
`
`standard means of querying databases is via SQL (Structured Query Language), which
`
`is in essence a special-purpose programming language. However many non-
`
`programmers, or programmers who do not know SQL, need to access databases
`
`6
`
`Apple Inc. v. Rensselaer Polytechnic Institute
` IPR2014-00320
` RPI Ex. 2005 PAGE 6
`
`

`

`Case 1:13-cv-00633-DEP Document 41-2 Filed 01/03/14 Page 8 of 17
`
`routinely, and therein lies the challenge – how to do so without specialized knowledge
`
`for programming SQL. One approach is to hire SQL programmers to act as
`
`intermediaries between the user and the database, but that is slow and expensive.
`
`Another approach is via a graphical user interface to SQL, but this limits query
`
`expressivity to predefined types of queries. The preferred approach relies on natural
`
`language interfaces, and that is the approach taken by the invention of the ’798 Patent.
`
`19.
`
`The ’798 Patent distinguishes itself from prior art by addressing head-on
`
`the issue of linguistic coverage and query disambiguation through direct use of the
`
`database, metadata database, and reference dictionary (essentially a specification of
`
`the metadata database). The problem of “linguistic closure” – that is, coping with all
`
`possible linguistic variants of queries and elements therein is stated up front (’798
`
`Patent col.2 ll.23-31): “A problem realized with many conventional natural language
`
`system designs is that these designs require exceedingly large collections of linguistic
`
`terms that users use, but still might not be able to assure successful closure of users’
`
`queries. Because of design complexity and keyword data-base size, most systems are
`
`not practical to implement. A better approach to processing natural language inputs is
`
`therefore needed.”
`
`20.
`
`The ’798 Patent goes on to critique as insufficient the prior approaches to
`
`NLP-DB, including implicitly those cited in Apple’s alleged prior art, for instance (’798
`
`Patent col.5 ll.43-50): “Most previous research efforts on NLI [Natural Language
`
`Interfaces] have set out to find ways for the computer to understand the user’s
`
`articulation, following the more established tradition of Artificial Intelligence (AI). But
`
`their results have stopped significantly short of being truly natural. They have all
`
`endeavored to devise particular controls and limitations on the naturalness of input and
`
`7
`
`Apple Inc. v. Rensselaer Polytechnic Institute
` IPR2014-00320
` RPI Ex. 2005 PAGE 7
`
`

`

`Case 1:13-cv-00633-DEP Document 41-2 Filed 01/03/14 Page 9 of 17
`
`use these artifacts to assist interpreting queries into some standard database query
`
`languages.”
`
`21.
`
`Then the ’798 Patent provides the key insight underlying the invention
`
`(’798 Patent col.6 ll.14-21): “One could argue that users are bound to refer, either
`
`directly or indirectly, to these known database objects (types or semantic models,
`
`instances or values, and operators) in their natural queries. If they do not use these
`
`database objects directly, they still have to use other words and phrases (henceforth
`
`known as “keywords”) that correspond to these objects. Thus, the domain of
`
`interpretation is finite, compared to natural language processing in general.” And the
`
`specification further states (id. col.6 ll.24-30): “The critical success factor of the last
`
`approach [the semantic model approach] depends clearly on the semantic model-
`
`dictionary employed, which must be powerful enough at least to span the range of
`
`possible usage . . . .” The specification goes on to describe the invention.
`
`22. One key aspect of the invention is the set of ingredients used in the
`
`semantic model, as detailed in the ’798 Patent (id. col.8 ll.51-59): “[A]ccording to one
`
`aspect of the invention, there are four layers of enterprise metadata (resources of
`
`search) considered; i.e., cases, keywords, information models, and database values.
`
`According to one aspect of the invention they are integrated in an extensible metadata
`
`representation method so that every resource item references all other related
`
`resources for query interpretation. A repository of metadata may be implemented as,
`
`for example, a reference dictionary.” Further details follow. However, the above is the
`
`crux of the invention: instead of first performing opened-ended NLP – a daunting task
`
`not yet truly solved – use the database itself, together with the metadatabase semantic
`
`model, to restrict the possible words, word meanings, keywords, and semantic
`
`8
`
`Apple Inc. v. Rensselaer Polytechnic Institute
` IPR2014-00320
` RPI Ex. 2005 PAGE 8
`
`

`

`Case 1:13-cv-00633-DEP Document 41-2 Filed 01/03/14 Page 10 of 17
`
`interpretations to a manageable finite set. In other words, tailor the interface to the
`
`database and the metadata that describes it, rather than a two step process of general
`
`NLP followed by NLP-to-database-query mapping, since the first of these two traditional
`
`steps is much too difficult and error-prone. That is the essence of the invention, though,
`
`of course, the details provided in the specification are also material.
`
`23. Claim 1 of the ’798 Patent recites:
`
`A method for processing a natural language input provided by a user, the method
`
`comprising:
`
`a. providing a natural language query input by the user;
`
`b. performing, based on the input, without augmentation, a search of one or
`
`more language-based databases including at least one metadata
`
`database comprising at least one of a group of information types
`
`comprising:
`
`i. case information;
`
`ii. keywords;
`
`iii. information models; and
`
`iv. database values;
`
`c. providing, through a user interface, a result of the search to the user;
`
`d. identifying, for the one or more language-based databases, a finite
`
`number of database objects; and
`
`e. determining a plurality of combinations of the finite number of database
`
`objects.
`
`24. Element (b) of claim 1 should be interpreted as including a metadata
`
`database comprised of one or more groups of information types, each group containing
`
`9
`
`Apple Inc. v. Rensselaer Polytechnic Institute
` IPR2014-00320
` RPI Ex. 2005 PAGE 9
`
`

`

`Case 1:13-cv-00633-DEP Document 41-2 Filed 01/03/14 Page 11 of 17
`
`all four of: case information, keywords, information models, and database values. This
`
`reading is fully supported by the ’798 Patent: for example, the patent (col.8 ll.51-54)
`
`states: “there are four layers of enterprise metadata (resources of search) considered;
`
`i.e., cases, keywords, information models and database values, they are integrated in
`
`an extensible metadata representation method so that every resource[] item references
`
`all other related resources for query interpretation.” And Figure 2 contains all four
`
`information types in the “reference dictionary” (metadata) – as stated in the ’798 Patent
`
`(col.8 ll.57-59): “A repository of metadata may be implemented as, for example, a
`
`reference dictionary.” In other words, these four metadata elements are meant to
`
`function together, as a group, as claim 1 states. Claim 9 largely parallels claim 1 with
`
`respect to the group of information types, reciting “…at least one of a group of
`
`information comprising case information, keywords, information models, and database
`
`values.” Moreover a “group” cannot refer to a single element, and “database values”
`
`are not in themselves metadata (although they can be contained in a metadata
`
`database along with metadata). Hence the above reading is the one consistent with the
`
`specification, with the other claims, and with the normal meaning of “group” and of
`
`“metadata.”
`
`25. Element (b) of claim 1 of the ’798 Patent, and the corresponding elements
`
`of claims 4, 9, and 14 are absent from each of the prior art references Apple relies on in
`
`its IPR Petition. None of Apple’s references disclose case information as metadata, and
`
`certainly none disclose the combination of all four types of metadata for NLP of
`
`database queries.
`
`10
`
`Apple Inc. v. Rensselaer Polytechnic Institute
` IPR2014-00320
` RPI Ex. 2005 PAGE 10
`
`

`

`Case 1:13-cv-00633-DEP Document 41-2 Filed 01/03/14 Page 12 of 17
`
`F.
`
`Validity of the ’798 Patent
`
`26.
`
`In its IPR Petition, Apple argues that U.S. Patent 5,197,005 (inventors:
`
`Shwartz, et al., hereinafter “Shwartz”) anticipates claims 1-21 of the ’798 Patent.
`
`Shwartz does not anticipate the ’798 Patent claims at least because it fails to disclose
`
`the group of information types as claimed in the ’798 Patent.
`
`27. Shwartz discloses a database retrieval system with “A database
`
`independent, canonical internal meaning representation of a natural language query”
`
`(Shwartz, Abstract (emphasis added)). The natural language query goes through a
`
`pipeline of steps divorced from the database or its associated metadata, and only at the
`
`end is it converted into an SQL query (id. Fig 1). A “semantic network” (i.e., a
`
`knowledge base or “internal meaning representation”) separate from any specific
`
`database is illustrated in Fig 7. In fact, Appendix A of Shwartz provides explicitly the
`
`Internal Meaning Representation Grammar, which does not use or rely upon any
`
`database or the claimed metadata database or reference dictionary or case information.
`
`Essentially, Shwartz represents an excellent example of the traditional paradigm for
`
`NLP-DB wherein the NLP is preformed independent of the database, contrary to the
`
`teachings of the ’798 Patent, wherein data and metadata play central roles in the NLP.
`
`Simply, Shwartz discloses technology quite different from the invention of the ’798
`
`Patent.
`
`28.
`
`In its IPR Petition, Apple also argues that Jurgen M. Janas’s article, The
`
`Semantics-Based Natural Language Interface to Relational Databases, in Cooperative
`
`Interfaces to Information Systems (Bolc & Jarke Eds.) (hereinafter “Janas”) anticipates
`
`claims 1-11, 13, 15, 16, and 18-21 of the ’798 Patent. Janas does not anticipate the
`
`11
`
`Apple Inc. v. Rensselaer Polytechnic Institute
` IPR2014-00320
` RPI Ex. 2005 PAGE 11
`
`

`

`Case 1:13-cv-00633-DEP Document 41-2 Filed 01/03/14 Page 13 of 17
`
`’798 Patent claims at least because it fails to disclose the group of information types as
`
`claimed in the ’798 Patent.
`
`29.
`
`Janas discloses the NLP processing of queries into a formal query
`
`language (pp. 150-151), a simplified form of SQL or its equivalent, as a “translation
`
`process” (e.g., pp. 146, 154-156). Like Shwartz, as discussed above, and unlike the
`
`’798 Patent, Janas does not rely on the database or the claimed metadata database to
`
`take part in the processing of the natural language query. Moreover, Janas makes no
`
`reference to past query processing or past disambiguation of queries, or any reuse of
`
`this dynamically recorded prior computation, and therefore fails to disclose “case
`
`information.” Instead, much of Janas is directed towards coping with linguistic
`
`phenomena such as conjunctions and ellipsis – important in their own right, but not
`
`pertinent to the ’798 Patent’s invention.
`
`30.
`
`In its IPR Petition, Apple also argues that Beatrice Bouchou’s and Denis
`
`Maurel’s article, Using Transducers in Natural Language Database Query, in the 4th
`
`International Conference on Applications of Natural Language to Information Systems
`
`(hereinafter “Bouchou”) anticipates claims 1-6, 8-12, 14-16, and 20-21 of the ’798
`
`Patent. Bouchou does not anticipate the ’798 Patent claims at least because it fails to
`
`disclose the group of information types as claimed in the ’798 Patent.
`
`31. Bouchou proposes a very simple method for NLP of database queries,
`
`relying on finite-state machines called “transducers,” which essentially map one string
`
`into another (i.e., transducers map a word or sequence of words into another word or
`
`sequence of words). Transducers are less powerful than even the simplest grammars
`
`typically applied in NLP, namely context-free grammars. As such they have been used
`
`for simple tasks such as morphology, for instance generating the plural form of a word
`
`12
`
`Apple Inc. v. Rensselaer Polytechnic Institute
` IPR2014-00320
` RPI Ex. 2005 PAGE 12
`
`

`

`Case 1:13-cv-00633-DEP Document 41-2 Filed 01/03/14 Page 14 of 17
`
`from the singular by adding “s” or “es” at the end, adding “ed” at the end of a verb to
`
`change it to its past tense, or substituting a word with a synonymous one or equivalent
`
`one, the latter being illustrated in Bouchou (p. 9).
`
`32. Moreover, Bouchou relies on linguistic methods (Sec 1.3 on pp. 8-9)
`
`stating “the use of linguistic operators rather than of logical operators,” albeit the
`
`simplistic linguistic methods mentioned above, and does not rely on semantic (logical)
`
`methods. Bouchou in fact teaches away from semantic methods, whether they use a
`
`general knowledge base or use the more specific and reliable semantics from the
`
`metadata database claimed in the ’798 Patent. This is further emphasized elsewhere in
`
`Bouchou (Sec 3.3 on p. 12). Nowhere does Bouchou disclose “case information” or
`
`“information models” and thus fails to disclose the group of information types as claimed
`
`in the ’798 Patent.
`
`33.
`
`In its IPR Petition, Apple also argues that Sajul Dar, Gadi Entin, Shai
`
`Geva and Eran Palmon’s paper, DTL’s DataSpot: Database Exploration Using Plain
`
`Language, in the 24th VLDB Conference (hereinafter “Dar”) anticipates claims 9-10 and
`
`13-16 of the ’798 Patent. Dar does not anticipate the ’798 Patent claims at least
`
`because it fails to disclose the group of information types as claimed in the ’798 Patent.
`
`34. Dar describes “DTL’s DataSpot,” which is a “database publishing tool” and
`
`first converts databases into something Dar et al. call the “hyperbase” offline (abstract).
`
`“A hyperbase is a graph structure comprised of nodes, edges and node labels” (p. 646,
`
`col.1); it is no longer a database comprised of tables, rows, columns, entries. Then, the
`
`users may query the hyperbase “much in the same way as they use a Web search
`
`engine, such as Altavista” (p. 645, col.2); that is, by using key terms, such as “Mexico”
`
`or “Nancy’s sea food orders,” rather than natural language queries (p. 647, col.2). In
`
`13
`
`Apple Inc. v. Rensselaer Polytechnic Institute
` IPR2014-00320
` RPI Ex. 2005 PAGE 13
`
`

`

`Case 1:13-cv-00633-DEP Document 41-2 Filed 01/03/14 Page 15 of 17
`
`fact, Dar differentiates itself from systems containing natural language interfaces that
`
`translate queries into SQL (p. 649, col.1). Further, Dar completely fails to disclose case
`
`information, disclosing instead “thesaurus” and “heuristics” as powering its search for
`
`answers to queries. Simply, Dar does not disclose a method for natural language
`
`processing that includes reference to the group of information types claimed in the ’798
`
`Patent.
`
`35.
`
`In its IPR Petition, Apple also argues that U.S. Patent 6,584,464 (inventor:
`
`Warthen, hereinafter “Warthen”) anticipates claims 1-11 and 13-21 of the ’798 Patent.
`
`Warthen does not anticipate the ’798 Patent claims at least because it fails to disclose
`
`the group of information types as claimed in the ’798 Patent.
`
`36. Warthen describes the AskJeeves search engine. Unlike other search
`
`engines, AskJeeves stores questions and answers, or more generally question
`
`templates (questions in canonical form such as “what is the weather in Boston”?) and
`
`answers (e.g., a URL to the weather.com page with Boston as the location). Questions
`
`in non-canonical form (such as “what’s Boston’s weather?” Or “For Boston, tell me the
`
`weather”) go through a multi-step process for conversion into canonical form, and then
`
`the answer is looked up in the question-answer mapping table and returned to the user
`
`(Fig 1(a) and 1:27-56). Warthen simply teaches a question-answer mapping table – the
`
`equivalent to a very large FAQ (Frequently-Asked Questions) list. Nothing new can
`
`ever be answered unless there is a canned answer in the FAQ (the question-answer
`
`mapping table). Hence, Warthen discloses a fairly impoverished model, and does not
`
`have the power of either a true search engine (such as Google, Bing, Alta Vista or
`
`Lycos), nor does it have the power of a natural language interface to a real database,
`
`where answers can be calculated by retrieving and combining multiple elements from
`
`14
`
`Apple Inc. v. Rensselaer Polytechnic Institute
` IPR2014-00320
` RPI Ex. 2005 PAGE 14
`
`

`

`Case 1:13-cv-00633-DEP Document 41-2 Filed 01/03/14 Page 16 of 17
`
`one or multiple tables. Warthen discloses nothing more than a big lookup list, like a
`
`phone book, albeit allowing a degree of flexibility via attempting to canonicalize the
`
`input. It does not disclose case information, metadata database information models, or
`
`the group of information types claimed in the ’798 Patent.
`
`37.
`
`Finally, in its IPR Petition, Apple also argues that claims 11, 12, 14, and
`
`17 are obvious in light of David L. Waltz’s article, An English Language Question and
`
`Answering System for a Large Relational Database, (hereinafter “Waltz”), in various
`
`combinations with Janas, Warthen, Bouchou, and Dar.
`
`38. As with Apple’s other references, Waltz fails to disclose at least the group
`
`of information types claimed in the ’798 Patent. Indeed, Apple does not suggest that
`
`Waltz discloses such. Therefore the combinations of references including Waltz
`
`necessarily fail to disclose the group of information types as claimed in each of the
`
`claims of the ’798 Patent.
`
`G.
`
`Conclusion
`
`39.
`
`Taken alone or in combination, the references that Apple relies on in its
`
`IPR Petition clearly fail to disclose certain key limitations common to all of the ’798
`
`Patent claims, including the case information and the group of information types as
`
`claimed in the patent. Apple’s invalidity arguments in the IPR Petition, therefore, facially
`
`and necessarily fail.
`
`
`
`
`
`15
`
`Apple Inc. v. Rensselaer Polytechnic Institute
` IPR2014-00320
` RPI Ex. 2005 PAGE 15
`
`

`

`Case 1:13-cv-00633-DEP Document 41-2 Filed 01/03/14 Page 17 of 17
`
`I declare under penalty of perjury under the laws of the United States of America that
`
`the foregoing is true and correct.
`
`
`
`Executed on January 2, 2014,
`
`
`
`Dr. Jaime Carbonell
`
`16
`
`Apple Inc. v. Rensselaer Polytechnic Institute
` IPR2014-00320
` RPI Ex. 2005 PAGE 16
`
`