U.S. patent application number 16/200411 was filed with the patent office on 2020-02-06 for generating additional training data for a natural language understanding engine.
The applicant listed for this patent is botbotbotbot Inc.. Invention is credited to Yi Ma, Antoine Raux.
Application Number | 20200043469 16/200411 |
Document ID | / |
Family ID | 64315482 |
Filed Date | 2020-02-06 |
![](/patent/app/20200043469/US20200043469A1-20200206-D00000.png)
![](/patent/app/20200043469/US20200043469A1-20200206-D00001.png)
![](/patent/app/20200043469/US20200043469A1-20200206-D00002.png)
![](/patent/app/20200043469/US20200043469A1-20200206-D00003.png)
![](/patent/app/20200043469/US20200043469A1-20200206-D00004.png)
United States Patent
Application |
20200043469 |
Kind Code |
A1 |
Raux; Antoine ; et
al. |
February 6, 2020 |
GENERATING ADDITIONAL TRAINING DATA FOR A NATURAL LANGUAGE
UNDERSTANDING ENGINE
Abstract
Methods, systems, and apparatus, including computer programs
encoded on computer storage media, for generating additional
training data for a natural language understanding engine. One of
the methods includes: obtaining data identifying (i) a first input
conversational turn and (ii) a first annotation, determining that
the first annotation accurately characterized the first input
conversational turn, determining that the natural language
understanding engine is likely to generate inaccurate annotations
of other conversational turns that are similar to the first input
conversational turn, in response to the determining, obtaining one
or more first paraphrases of the first input conversational turn;
and generating, for each of the one or more first paraphrases, a
respective first training example that identifies the first
annotation as the correct annotation for the first paraphrase; and
training the natural language understanding engine on at least the
first training examples.
Inventors: |
Raux; Antoine; (Palo Alto,
CA) ; Ma; Yi; (Santa Clara, CA) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
botbotbotbot Inc. |
Palo Alto |
CA |
US |
|
|
Family ID: |
64315482 |
Appl. No.: |
16/200411 |
Filed: |
November 26, 2018 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
16051362 |
Jul 31, 2018 |
10140977 |
|
|
16200411 |
|
|
|
|
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G06N 7/005 20130101;
G10L 15/063 20130101; G06K 9/6256 20130101; G10L 15/01 20130101;
G10L 2015/225 20130101; G06F 40/20 20200101; G10L 15/22 20130101;
G06N 20/00 20190101; G06N 3/02 20130101; G06F 40/35 20200101; G06F
40/51 20200101; G06N 3/0445 20130101; G06N 3/084 20130101 |
International
Class: |
G10L 15/06 20060101
G10L015/06; G06F 17/27 20060101 G06F017/27; G06K 9/62 20060101
G06K009/62; G10L 15/01 20060101 G10L015/01; G06F 17/28 20060101
G06F017/28; G06N 3/02 20060101 G06N003/02; G06N 20/00 20060101
G06N020/00 |
Claims
1. A system comprising one or more computers and one or more
storage devices storing instructions that when executed by the one
or more computers cause the one or more computers to perform
operations comprising: obtaining, during operation of a
computer-implemented dialogue system comprising a natural language
understanding engine, data identifying (i) a first input
conversational turn that was provided as input to the natural
language understanding engine during a dialogue between a user and
the computer-implemented dialogue system and (ii) a first
annotation of the first input conversational turn generated by the
natural language understanding engine, wherein the natural language
understanding engine has been trained on a first set of training
data comprising a plurality of training conversational turns;
determining that the first annotation did not accurately
characterize the first input conversational turn; in response to
determining that the first annotation did not accurately
characterize the first input conversational turn: determining a
correct annotation for the first conversational turn; obtaining one
or more first paraphrases of the first input conversational turn;
and generating, for each of the one or more first paraphrases, a
respective first training example that identifies the correct
annotation for the first conversational turn as the correct
annotation for the first paraphrase; and training the natural
language understanding engine on at least the first training
examples.
2. The system of claim 1, the operations further comprising
obtaining a confidence score generated by the natural language
understanding engine that represents a confidence that the first
annotation is an accurate characterization of the first input
conversational turn, and wherein determining that the first
annotation did not accurately characterize the first input
conversational turn comprises determining that the confidence score
fails to exceed a threshold.
3. The system of claim 1, wherein determining that the first
annotation did not accurately characterize the first input
conversational turn comprises: processing the first input
conversational turn and the first annotation using a post-hoc
annotation machine learning model that is configured to generate as
output a quality score that represents a likelihood that the first
annotation is an accurate characterization of the first input
conversational turn; and determining that the quality score fails
to exceed a threshold.
4. The system of claim 3, wherein the post-hoc annotation machine
learning model is configured to receive as input (i) the first
input conversational turn, (ii) the first annotation, (iii) one or
more conversational turns occurring before the first input
conversational turn in the dialogue, and (iv) one or more
conversational turns occurring after the first input conversational
turn in the dialogue.
5. The system of claim 1, wherein determining that the first
annotation does not accurately characterize the first input
conversational turn comprises: determining that conversational
turns occurring after the first input conversational turn in the
dialogue indicate that the first annotation did not accurately
characterize the first input conversational turn.
6. The system of claim 5, wherein determining that conversational
turns occurring after the first input conversational turn in the
dialogue indicate that the first annotation did not accurately
characterize the first input conversational turn comprises:
determining (i) that a task was completed as a result of the
dialogue and (ii) that one or more slot values assigned in the
first annotation were changed or removed by the user in one or more
of the conversational turns occurring after the first input
conversational turn in the dialogue.
7. A method comprising: obtaining, during operation of a
computer-implemented dialogue system comprising a natural language
understanding engine, data identifying (i) a first input
conversational turn that was provided as input to the natural
language understanding engine during a dialogue between a user and
the computer-implemented dialogue system and (ii) a first
annotation of the first input conversational turn generated by the
natural language understanding engine, wherein the natural language
understanding engine has been trained on a first set of training
data comprising a plurality of training conversational turns;
determining that the first annotation did not accurately
characterize the first input conversational turn; in response to
determining that the first annotation did not accurately
characterize the first input conversational turn: determining a
correct annotation for the first conversational turn; obtaining one
or more first paraphrases of the first input conversational turn;
and generating, for each of the one or more first paraphrases, a
respective first training example that identifies the correct
annotation for the first conversational turn as the correct
annotation for the first paraphrase; and training the natural
language understanding engine on at least the first training
examples.
8. The method of claim 7, further comprising obtaining a confidence
score generated by the natural language understanding engine that
represents a confidence that the first annotation is an accurate
characterization of the first input conversational turn, and
wherein determining that the first annotation does not accurately
characterize the first input conversational turn comprises
determining that the confidence score fails to exceed a
threshold.
9. The method of claim 7, wherein determining that the first
annotation accurately characterized the first input conversational
turn comprises: processing the first input conversational turn and
the first annotation using a post-hoc annotation machine learning
model that is configured to generate as output a quality score that
represents a likelihood that the first annotation is an accurate
characterization of the first input conversational turn; and
determining that the quality score fails to exceed a threshold.
10. The method of claim 9, wherein the post-hoc annotation machine
learning model is configured to receive as input (i) the first
input conversational turn, (ii) the first annotation, (iii) one or
more conversational turns occurring before the first input
conversational turn in the dialogue, and (iv) one or more
conversational turns occurring after the first input conversational
turn in the dialogue.
11. The method of claim 7, wherein determining that the first
annotation does not accurately characterize the first input
conversational turn comprises: determining that conversational
turns occurring after the first input conversational turn in the
dialogue indicate that the first annotation does not accurately
characterize the first input conversational turn.
12. The method of claim 11, wherein determining that conversational
turns occurring after the first input conversational turn in the
dialogue indicate that the first annotation does not accurately
characterize the first input conversational turn comprises:
determining (i) that a task was completed as a result of the
dialogue and (ii) that one or more slot values assigned in the
first annotation were changed or removed by the user in one or more
of the conversational turns occurring after the first input
conversational turn in the dialogue.
13. One or more non-transitory computer-readable storage media
storing instructions that when executed by one or more computers
cause the one or more computers to perform operations comprising:
obtaining, during operation of a computer-implemented dialogue
system comprising a natural language understanding engine, data
identifying (i) a first input conversational turn that was provided
as input to the natural language understanding engine during a
dialogue between a user and the computer-implemented dialogue
system and (ii) a first annotation of the first input
conversational turn generated by the natural language understanding
engine, wherein the natural language understanding engine has been
trained on a first set of training data comprising a plurality of
training conversational turns; determining that the first
annotation does not accurately characterize the first input
conversational turn; in response to determining that the first
annotation did not accurately characterize the first input
conversational turn: determining a correct annotation for the first
conversational turn; obtaining one or more first paraphrases of the
first input conversational turn; and generating, for each of the
one or more first paraphrases, a respective first training example
that identifies the correct annotation for the first conversational
turn as the correct annotation for the first paraphrase; and
training the natural language understanding engine on at least the
first training examples.
14. The computer-readable storage media of claim 13, further
comprising obtaining a confidence score generated by the natural
language understanding engine that represents a confidence that the
first annotation is an accurate characterization of the first input
conversational turn, and wherein determining that the first
annotation does not accurately characterize the first input
conversational turn comprises determining that the confidence score
fails to exceed a threshold.
15. The computer-readable storage media of claim 13, wherein
determining that the first annotation accurately characterized the
first input conversational turn comprises: processing the first
input conversational turn and the first annotation using a post-hoc
annotation machine learning model that is configured to generate as
output a quality score that represents a likelihood that the first
annotation is an accurate characterization of the first input
conversational turn; and determining that the quality score fails
to exceed a threshold.
16. The computer-readable storage media of claim 15, wherein the
post-hoc annotation machine learning model is configured to receive
as input (i) the first input conversational turn, (ii) the first
annotation, (iii) one or more conversational turns occurring before
the first input conversational turn in the dialogue, and (iv) one
or more conversational turns occurring after the first input
conversational turn in the dialogue.
17. The computer-readable storage media of claim 13, wherein
determining that the first annotation does not accurately
characterize the first input conversational turn comprises:
determining that conversational turns occurring after the first
input conversational turn in the dialogue indicate that the first
annotation does not accurately characterize the first input
conversational turn.
18. The computer-readable storage media of claim 17, wherein
determining that conversational turns occurring after the first
input conversational turn in the dialogue indicate that the first
annotation does not accurately characterize the first input
conversational turn comprises: determining (i) that a task was
completed as a result of the dialogue and (ii) that one or more
slot values assigned in the first annotation were changed or
removed by the user in one or more of the conversational turns
occurring after the first input conversational turn in the
dialogue.
Description
BACKGROUND
[0001] This specification relates to dialogue systems.
[0002] A dialogue system is a computer system that has
conversations with users by generating system outputs in response
to user conversational turns.
[0003] For example, during a given conversation, a dialogue system
can receive a speech input from the user that represents a
conversational turn, convert the speech input to text, and then
operate on the text to generate a speech output that is a response
to the speech input received from the user.
[0004] Dialogue systems can be used to converse with users to
accomplish any of a variety of tasks. For example, a dialogue
system can be used to allow users to select options from a menu,
e.g., food items from a restaurant menu. As another example, a
dialogue system can be used to allow users to make a reservation,
e.g., a restaurant reservation or a travel reservation.
[0005] Some dialogue systems use neural networks as part of
generating a system output from a user input.
[0006] Neural networks are machine learning models that employ one
or more layers of nonlinear units to predict an output for a
received input. Some neural networks include one or more hidden
layers in addition to an output layer. The output of each hidden
layer is used as input to the next layer in the network, i.e., the
next hidden layer or the output layer. Each layer of the network
generates an output from a received input in accordance with
current values of a respective set of parameters.
[0007] Some neural networks are recurrent neural networks. A
recurrent neural network is a neural network that receives an input
sequence and generates an output sequence from the input sequence.
In particular, a recurrent neural network can use some or all of
the internal state of the network from a previous time step in
computing an output at a current time step.
[0008] An example of a recurrent neural network is a Long
Short-Term Memory (LSTM) neural network that includes one or more
LSTM memory blocks. Each LSTM memory block can include one or more
cells that each include an input gate, a forget gate, and an output
gate that allow the cell to store previous states for the cell,
e.g., for use in generating a current activation or to be provided
to other components of the LSTM neural network.
SUMMARY
[0009] This specification describes how a dialogue system
implemented as computer programs on one or more computers in one or
more locations can generate additional training data that improves
the performance of a natural language understanding (NLU) engine
that is included in the dialogue system.
[0010] In general, one innovative aspect of the subject matter
described in this specification can be embodied in methods that
include the actions of obtaining, during operation of a
computer-implemented dialogue system comprising a natural language
understanding engine, data identifying (i) a first input
conversational turn that was provided as input to the natural
language understanding engine during a dialogue between a user and
the computer-implemented dialogue system and (ii) a first
annotation of the first input conversational turn generated by the
natural language understanding engine, wherein the natural language
understanding engine has been trained on a first set of training
data comprising a plurality of training conversational turns;
determining that the first annotation accurately characterized the
first input conversational turn; determining, based on the training
conversational turns in the first set of training data, that the
natural language understanding engine is likely to generate
inaccurate annotations of other conversational turns that are
similar to the first input conversational turn; in response to
determining that (i) the first annotation accurately characterized
the first input conversational turn but (ii) the natural language
understanding engine is likely to generate inaccurate annotations
of other conversational turns that are similar to the first input
conversational turn: obtaining one or more first paraphrases of the
first input conversational turn; and generating, for each of the
one or more first paraphrases, a respective first training example
that identifies the first annotation as the correct annotation for
the first paraphrase; and training the natural language
understanding engine on at least the first training examples.
[0011] Other embodiments of this aspect include corresponding
computer systems, apparatus, and computer programs recorded on one
or more computer storage devices, each configured to perform the
actions of the methods.
[0012] The foregoing and other embodiments can each optionally
include one or more of the following features, alone or in
combination. In particular, one embodiment includes all the
following features in combination.
[0013] Determining, based on the training conversational turns in
the first set of training data, that the natural language
understanding engine is likely to generate inaccurate annotations
of other conversational turns that are similar to the first input
conversational turn can include: identifying training
conversational turns in the training data that are similar to the
first input conversational turn according to a similarity
measure.
[0014] Identifying training conversational turns in the training
data that are similar to the first input conversational turn
according to a similarity measure can include: determining a
numeric embedding representation of the first input conversational
turn; determining respective numeric embedding representations of
each of the training conversational turns; and determining
respective distances between the numeric embedding representation
of the first input conversational turn and the numeric embedding
representations of the training conversational turns.
[0015] Determining, based on the training conversational turns in
the first set of training data, that the natural language
understanding engine is likely to generate inaccurate annotations
of other conversational turns that are similar to the first input
conversational turn can comprise: determining that there are less
than a threshold number of training conversational turns that are
similar to the first input conversational turn according to the
similarity measure.
[0016] Determining, based on the training conversational turns in
the first set of training data, that the natural language
understanding engine is likely to generate inaccurate annotations
of other conversational turns that are similar to the first input
conversational turn can comprise: determining that the natural
language understanding unit performs poorly on training
conversational turns that are similar to the first input
conversational turn according to the similarity measure.
[0017] The actions can further comprise obtaining a confidence
score generated by the natural language understanding engine that
represents a confidence that the first annotation is an accurate
characterization of the first input conversational turn and
determining that the first annotation accurately characterized the
first input conversational turn can comprise determining that the
confidence score exceeds a threshold score.
[0018] Determining that the first annotation accurately
characterized the first input conversational turn can comprise:
processing the first input conversational turn and the first
annotation using a post-hoc annotation machine learning model that
is configured to generate as output a quality score that represents
a likelihood that the first annotation is an accurate
characterization of the first input conversational turn; and
determining that the quality score exceeds a threshold score.
[0019] The post-hoc annotation machine learning model can be
configured to receive as input (i) the first input conversational
turn, (ii) the first annotation, (iii) one or more conversational
turns occurring before the first input conversational turn in the
dialogue, and (iv) one or more conversational turns occurring after
the first input conversational turn in the dialogue.
[0020] Determining that the first annotation accurately
characterized the first input conversational turn can comprise:
determining that conversational turns occurring after the first
input conversational turn in the dialogue indicate that the first
annotation accurately characterizes the first input conversational
turn.
[0021] Determining that conversational turns occurring after the
first input conversational turn in the dialogue indicate that the
first annotation accurately characterize the first input
conversational turn can comprise: determining (i) that a task was
completed as a result of the dialogue and (ii) that no slot values
assigned in the first annotation were changed or removed by the
user in any of the conversational turns occurring after the first
input conversational turn in the dialogue.
[0022] The actions can further comprise obtaining, during operation
of the computer-implemented dialogue system, data identifying (i) a
second input conversational turn that was provided as input to the
natural language understanding engine during the dialogue between
the user and the computer-implemented dialogue system and (ii) a
second annotation of the second input conversational turn generated
by the natural language understanding engine; determining that the
second annotation did not accurately characterize the second input
conversational turn; in response to determining that the second
annotation did not accurately characterize the second input
conversational turn: determining a correct annotation for the
second conversational turn; obtaining one or more second
paraphrases of the second input conversational turn; and
generating, for each of the one or more second paraphrases, a
respective second training example that identifies the correct
annotation for the second conversational turn as the correct
annotation for the second paraphrase; and training the natural
language understanding engine on at least the second training
examples.
[0023] The subject matter described in this specification can be
implemented in particular embodiments so as to realize one or more
of the following advantages.
[0024] The described systems can generate training data that is
used to improve the performance of the NLU engine and, therefore,
the performance of a computer-implemented dialogue system that
includes the NLU engine. That is, by generating additional training
data as described in this specification and training the NLU engine
on the generated additional training data, the computer-implemented
dialogue system is able to achieve increased performance on a
variety of tasks, i.e., is better able to interact with users to
assist the users in completing a variety of tasks, because the NLU
engine generates more accurate annotations of received user
inputs.
[0025] More specifically, the system can generate the additional
training data in a manner that results in high quality training
data being generated but minimizes the amount of human interaction,
time, and computational resources, e.g., memory and processing
power, required to generate the data.
[0026] In particular, generating training data for an NLU engine
requires accurate annotations to be obtained for the conversational
turns in the training data. That is, the conversational turns in
the training data must be labelled with accurate annotations in
order for the training data to be useful in improving the
performance of the NLU engine.
[0027] Obtaining such accurate annotations is conventionally a
time-intensive process that requires input from expert human
labelers, e.g., because generating an accurate annotation requires
accurately identifying a task-specific intent for the
conversational turn as well as accurately identifying the
intent-specific slots for the identified intent.
[0028] The described techniques, on the other hand, reduce the
amount of time, human involvement, and computational resources
necessary for generating high-quality training data. In particular,
the described techniques effectively generate additional training
data using paraphrases of a given identified conversational turn.
This allows the paraphrases to be associated with the same correct
annotation as the identified conversational turn, resulting in a
large number of high-quality additional training data being
generated without needing to expend computational resources or time
to obtain new correct annotations for the paraphrases.
[0029] Additionally, the additional data is generated to
specifically target areas where the NLU, as currently trained,
performs poorly. By generating high-quality training data that
targets these areas as described in this specification, the total
amount of training data required to train the NLU (and, therefore,
the computer-implemented dialogue system) to have a high
performance quality is reduced, thereby reducing the amount of time
and computational resources consumed by the training process. In
other words, using the described techniques, the system avoids
re-training the NLU on areas where the NLU already performs well,
reducing the amount of time and computational resources consumed by
the training process.
[0030] The details of one or more embodiments of the subject matter
described in this specification are set forth in the accompanying
drawings and the description below. Other features, aspects, and
advantages of the subject matter will become apparent from the
description, the drawings, and the claims.
BRIEF DESCRIPTION OF THE DRAWINGS
[0031] FIG. 1 shows an example dialogue system.
[0032] FIG. 2 is a flow diagram of an example process of generating
additional training data for a natural language understanding (NLU)
engine.
[0033] FIG. 3 is a flow diagram of an example process of generating
additional training data from an inaccurate annotation.
[0034] FIG. 4 is a flow diagram of an example process of generating
additional training data from an accurate annotation.
[0035] Like reference numbers and designations in the various
drawings indicate like elements.
DETAILED DESCRIPTION
[0036] FIG. 1 shows an example dialogue system 100. The dialogue
system 100 is an example of a system implemented as computer
programs on one or more computers in one or more locations, in
which the systems, components, and techniques described below are
implemented.
[0037] The dialogue system 100 is a system that engages in
conversations with users of the system.
[0038] Generally, each conversation is an attempt by the dialogue
system 100 to interact with a user to complete a task for the user.
Examples of tasks can include a food delivery task, a shopping
task, a restaurant reservation task, a travel reservation task, or
a route planning navigation task, among many others.
[0039] In some cases, the task is determined before the
conversation begins. For example, the dialogue system 100 may only
be configured to perform a single task. As another example, the
task to be completed by a given conversation may be defined by the
interface through which the user inputs are received during the
conversation. That is, each of multiple tasks that the system is
configured to complete is initiated by a user input submitted
through a corresponding interface.
[0040] In other cases, the dialogue system 100 determines the task
to be completed based on inputs submitted by the user during the
conversation. For example, the system may be configured to perform
multiple tasks and the dialogue system 100 can determine which task
to perform based on one or more initial user inputs during the
conversation.
[0041] During a given conversation with a user 102 of a user device
104, the dialogue system 100 receives a speech input 106
representing a conversational turn from the user device 104, e.g.,
over a data communication network, and, in response, generates a
speech output 172 representing a system response to the
conversational turn and provides the speech output 172 for playback
on the user device 104, e.g., also over the data communication
network.
[0042] To generate the speech output 172, a speech recognition
engine 110 in the dialogue system 100 receives the speech input 106
and converts the speech input 106 into a text input 112 that is a
transcription of the speech input (e.g., into the text "I'd like a
large pineapple pizza"). In other words, the speech recognition
converts the speech into a text representation of the
conversational turn submitted by the user. The speech recognition
engine 110 can convert the speech into text using any appropriate
conventional automatic speech recognition (ASR) technique, e.g., a
Hidden Markov Model (HMM) based recognition technique or a neural
network based recognition technique.
[0043] Once the text input 112 has been generated, a natural
language understanding (NLU) engine 120 converts the text input 112
into an annotation 122. Generally, the annotation 122 represents
the text input 112 in a way that captures the semantics of the text
input 112.
[0044] The annotation 122 can represent the semantics of the text
input 112 in any of a variety of ways.
[0045] For example, the annotation 122 can be a frame that
identifies (i) the intent of the text input 112 and (ii) values for
one or more slots corresponding to the intent. Each slot
corresponding to the intent is a slot whose value is required to be
filled in order for the intent to be satisfied. For example, when
the intent is to order a pizza from a restaurant menu, slots
required to be filled include one or more slots for toppings for
the pizza and a slot for the size of the pizza. As a specific
example, for the conversational turn "I'd like a large pineapple
pizza," the NLU engine 120 can generate an annotation 122 that
selects a "request item" intent from a set of intents for the
current task and assigns "large" as the value for the "size" slot
and "pineapple" as the value for the "topping1" slot while
indicating that the value for the "crust" slot is missing.
[0046] In some cases, the NLU engine 120 can also generate a
confidence score that indicates how confident the NLU is that the
conversational turn represented by the text input 122 has been
properly annotated by the NLU. In the example of FIG. 1, the NLU
engine 120 has assigned a confidence score of 0.9 to the annotation
122.
[0047] For example, the NLU engine 120 can be a recurrent neural
network (RNN)-based model, e.g., a Long Short-Term Memory
(LSTM)-based model, that receives as input the conversational turn
and outputs an annotation that (i) identifies the intent of the
conversational turn and (ii) an assignment of the various terms in
the conversational turns to slots corresponding to the intent. As
part of generating the annotation, the model also generates a
confidence score for the annotation. An example of such an
RNN-based model is described in Multi-Domain Joint Semantic Frame
Parsing using Bi-directional RNN-LSTM, Dilek Hakkani-Tur, et al,
available at
https://pdfs.semanticscholar.org/d644/ae996755c803e067899bdd5ea52498d7091-
d.pdf. Another example of such an RNN-based model is described in
Effective Spoken Language Labeling with Deep Recurrent Neural
Networks, Dinarelli, et al, available at
https://arxiv.org/pdf/1706.06896.pdf. Yet another example of such
an RNN-based model is described in SPOKEN LANGUAGE UNDERSTANDING
USING LONG SHORT-TERM MEMORY NEURAL NETWORKS, Yao, et al, available
at
https://groups.csail.mit.edu/sls/publications/2014/ZhangSLT2014.pdf.
[0048] The dialogue system 100 also includes a dialogue manager 130
that receives the annotation 122 and processes the annotation 122
to generate a semantic output 132. The semantic output 132 defines
the semantics of a response to the user input. For example, the
semantic output can include an output intent and values for one or
more slots corresponding to the output intent. In the example of
FIG. 1, the output 132 identifies the "prompt additional value"
output intent from a set of output intents for the task and
identifies "thin crust" as the value of the slot "value1" and "deep
dish" as the value of the slot "value2."
[0049] The dialogue system 100 also includes a natural language
generation engine 160 that receives the semantic output 132 and
generates a text output 166 that conveys the semantics defined by
the semantic output 132. For example, the natural language
generation engine 160 can generate text that reflects the output
intent and slot values included in the semantic output 132 (e.g.,
"Would you like thin crust or deep dish?").
[0050] Once the text output 166 has been generated, a
text-to-speech engine 170 converts the text output 166 into the
speech output 172, i.e., into speech that verbalizes the text
output 166. The text-to-speech engine 170 can use any appropriate
conventional text-to-speech technique to convert the text output
166 into the speech output 172, e.g., an HMM-based text-to-speech
technique or a neural network-based text-to-speech technique.
[0051] As described, the NLU engine 120 generally includes one or
more machine learning models, e.g., RNNs, and has been trained to
generate accurate annotations by a training engine 180.
[0052] In particular, the training engine 180 trains the NLU engine
120 on training data 190 that includes multiple training examples.
Each training example includes a training conversational turn and a
corresponding annotation that has been classified as the correct
annotation for the conversational turn.
[0053] The training engine 180 trains the NLU engine 120 on the
training data 190 by adjusting the weights of the one or more
machine learning models using an appropriate machine learning
training technique in order to cause the NLU engine 120 to generate
more accurate annotations for input conversational turns. For
example, when the NLU is an RNN-based model, the training engine
180 can train the NLU using backpropagation through time or
truncated backpropagation through time.
[0054] Once the training engine 180 has trained the NLU engine 120
to attain acceptable performance, the system 100 can begin
interacting with users, i.e., having conversations with users to
attempt to complete tasks.
[0055] During operation of the system, the training engine 180 can
generate additional training data for the NLU engine 120 and
periodically re-train the NLU engine 120 on the additional training
data to fine-tune the weights of the NLU engine 120. This can
further improve the accuracy of the NLU engine 120 and improve the
overall performance of the system.
[0056] In particular, the training engine 180 can generate the
additional training data by, during operation of the system,
identifying conversational turns that have been submitted by users
that are candidates for use in generating additional training data.
Once a conversational turn has been identified, the training engine
180 can obtain paraphrases of the conversational turn and use the
paraphrases as additional training conversational turns. In other
words, the training engine 180 can generate a respective training
example for each paraphrase and add the training examples to the
additional training data that will be used to continue to train the
NLU engine 120.
[0057] Generating additional training data will be described below
with reference to FIGS. 2-4.
[0058] In the example of FIG. 1, the speech recognition engine 110,
the natural language understanding engine 120, the dialogue manager
130, the natural language generation engine 160, and the
text-to-speech engine 170 are all described as separate components
of the dialogue system 100. In some other implementations, however,
the functionality of two or more of these systems may be
implemented as part of a single system. For example, a single
engine may directly process a speech input to generate a semantic
representation of the speech input.
[0059] Additionally, in the example of FIG. 1, the system input and
output are both illustrated as being speech. However, in other
examples, the modality of the system input, the modality of the
system output, or both may be different. For example, both the
input modality and the output modality may be text. As another
example, the input modality may be text while the output modality
is speech or vice versa.
[0060] FIG. 2 is a flow diagram of an example process 200 for
generating additional training data for the NLU engine. For
convenience, the process 200 will be described as being performed
by a system of one or more computers located in one or more
locations. For example, a dialogue system, e.g., the dialogue
system 100 of FIG. 1, appropriately programmed, can perform the
process 200.
[0061] During operation of the system, the system obtains a
conversational turn that was provided as input to the NLU engine
and an annotation that was generated by the NLU engine for the
conversational turn (step 202). In particular, the conversational
turn was received or generated by the system, i.e., from user
speech, during a conversation between a user and the system. If the
NLU engine also generates confidence scores, the system also
obtains the confidence score generated by the NLU engine for the
annotation.
[0062] The system determines that the conversational turn is a
candidate for paraphrasing to generate additional training data
(step 204).
[0063] In some implementations, the system determines that the
conversational turn is a candidate when the annotation did not
accurately characterize the conversational turn. Determining that
an annotation did not accurately characterize the conversational
turn is described below with reference to FIG. 3.
[0064] In some implementations, the system determines that the
conversational turn is a candidate when (i) the annotation
accurately characterized the conversational turn but (ii) the NLU
engine is likely to be inaccurate for conversational turns that are
similar to the current conversational turn. Making this
determination is described below with reference to FIG. 4.
[0065] In response to determining that the conversational turn is a
candidate for paraphrasing, the system performs steps 206 through
212 of the process 200 to generate additional training data. That
is, when the conversational turn is not a candidate, the system
does not generate any additional training data using the
conversational turn.
[0066] The system determines a final annotation for the
conversational turn (step 206). If it was determined that the
annotation accurately characterized the conversational turn, the
system can use the annotation generated by the NLU engine as the
final conversational turn. However, if it was determined that the
annotation did not accurately characterize the conversational turn,
the system determines a correct annotation for the conversational
turn and uses the correct annotation as the final annotation for
the conversational turn. Determining a correct annotation is
described below with reference to FIG. 3.
[0067] The system obtains one or more paraphrases of the
conversational turn (step 208).
[0068] A paraphrase of the current conversational turn is a
different conversational turn, i.e., a conversational turn that
uses different words from the current conversational turn, but that
should be interpreted the same way as the current conversational
turn by the system in the context of the current conversation. In
other words, each paraphrase is a conversational turn that (i) has
different words from the current conversational turn but (ii)
should be annotated the same way by the NLU engine in the context
of the current conversation.
[0069] In some implementations, the system generates the
paraphrases using a paraphrasing neural network that is configured
to receive as input a conversational turn and, optionally, data
characterizing the context for the current conversation, and to
output a paraphrase of the conversational turn. For example, the
paraphrasing neural network can be a sequence-to-sequence neural
network that includes an encoder configured to process the input to
generate an encoded representation of the input and a decoder
configured to process the encoded representations to generate the
paraphrase.
[0070] In some other implementations, the system generates the
paraphrases by submitting a request to a crowdsourcing platform,
e.g., through an application programming interface (API) provided
by the platform. The request can identify the conversational turn
and the context of conversation as of the time that the
conversational turn was submitted and request paraphrases that
should be assigned the same meaning as the current conversational
turn. In response, the system can obtain the paraphrases from the
crowdsourcing platform. An example of such a crowdsourcing platform
is the Amazon Mechanical Turk platform, hosted at
https://www.mturk.com/.
[0071] The system generates a respective training example for each
of the paraphrases (step 210). The training example for a given
paraphrase identifies the final annotation for the conversational
turn as the correct annotation for the paraphrase. That is, when
the annotation generated by the NLU engine for the conversational
turn was accurate, the training example identifies the annotation
generated for the conversational turn by the NLU engine as the
correct annotation for the paraphrase. When the annotation
generated by the NLU engine for the conversational turn was not
accurate, the training example identifies the correct annotation
determined by the system for the conversational turn as the correct
annotation for the paraphrase.
[0072] The system adds the training examples for the paraphrases to
the NLU engine training data (step 212).
[0073] Once certain criteria have been satisfied, the system trains
the NLU engine on the training examples that have been added to the
training data to fine-tune the current weights of the NLU engine.
For example, the system can train the NLU engine once a threshold
number of training examples have been added to the training data
since the last time the NLU engine was trained. As another example,
the system can train the NLU engine once a threshold amount of time
has elapsed since the last time the NLU engine was trained.
[0074] Thus, the system effectively generates additional training
examples on which the NLU engine can be trained to improve the
performance of the NLU engine by augmenting conversational
turn-annotation pairs encountered during operation of the system
with additional paraphrase-annotation pairs. The additional
paraphrase-annotation pairs include accurate annotations even
though they are generated with little to no human intervention and
in a computationally-efficient manner. Additionally, as described
below, additional paraphrase-annotation pairs are generated to
target areas where the NLU engine performs poorly or is likely to
perform poorly and therefore fewer additional training pairs are
required to improve the performance of the NLU engine, i.e.,
because further training of the NLU engine in areas where it
already performs well is avoided. This reduces the time and amount
of computational resources required to improve the performance of
the NLU engine.
[0075] FIG. 3 is a flow diagram of an example process 300 for
generating additional training data from an inaccurate annotation.
For convenience, the process 300 will be described as being
performed by a system of one or more computers located in one or
more locations. For example, a dialogue system, e.g., the dialogue
system 100 of FIG. 1, appropriately programmed, can perform the
process 300.
[0076] The system determines that the annotation did not accurately
characterize the conversational turn (step 302).
[0077] The system can apply a set of one or more criteria when
determining whether the annotation accurately characterized the
conversational turn. That is, when one or more of the criteria in
the set are satisfied, the system can determine that the annotation
did not accurately characterize the conversational turn.
[0078] For example, the set of criteria can include a criterion
that is based on the confidence score generated by the NLU engine
for the annotation. In particular, this criterion can be satisfied
when the score is below a threshold value.
[0079] As another example, the set of criteria can include one or
more criteria that are based on the progress of the conversation
after the annotation was generated, i.e., criteria that are
dependent on whether the conversation after the annotation was
generated indicates that the annotation for the current
conversational turn is inaccurate.
[0080] An example of such a criterion is a criterion that is
satisfied whenever a slot value assigned to a slot in the
annotation of the current conversational turn is changed or removed
by the user in a subsequent turn. Optionally, the criterion can
also require that the conversational resulted in a task
successfully being completed. For example, this criterion can be
satisfied if, during a conversation to order a pizza, (i) the
current turn is understood by the NLU as the user requesting a
large pizza, (ii) the user ends up placing an order at the end of
the conversation, and (iii) the placed order does not include a
large pizza. In this example, the fact that "large pizza" was not
included in the placed order is an indicator that the annotation
for the current conversational turn was incorrect.
[0081] As another example, the set of criteria can include a
criterion that is based on the output of a post-hoc annotation
machine learning model that has been trained to determine whether a
given annotation accurately characterizes a given conversational
input. In particular, the post-hoc annotation machine learning
model can be a recurrent neural network or a feedforward neural
network that is configured to receive (i) the current
conversational turn, (ii) the annotation assigned to the current
conversational turn, and (iii) the remainder of the conversation,
including conversational turns before the current conversational
turn and conversational turns after the current conversational turn
in the conversation. The output of the model can be an accuracy
score that represents the likelihood that the input current
conversational turn is accurately annotated by the input
annotation. The system can train the post-hoc annotation machine
learning model using an appropriate machine learning technique on
training data that includes a set of inputs that each include (i),
(ii), and (iii) above and, for each input in the set, a target
output that indicates whether or not the annotation in the input
accurately characterized the current conversational turn in the
input.
[0082] In response to determining that the annotation did not
accurately characterize the conversational turn, the system
determines that the conversational turn is a candidate for
paraphrasing and proceeds to perform steps 304 through 310 of the
process 300.
[0083] The system determines a correct annotation for the
conversational turn (step 304). That is, because it was determined
that the annotation did not accurately characterize the
conversational turn, the system determines a correct annotation for
the conversational turn and uses the correct annotation as the
final annotation for the conversational turn.
[0084] In particular, the system provides a request for a correct
annotation to the crowdsourcing platform or to a user device of an
expert user and obtains the correct annotation in response to the
request. The request for the correct annotation includes the
conversational turn and the context of the current conversation.
For example, the context can be the conversational turns and system
responses that have already been generated during the conversation
or data summarizing the state of the conversation prior to the
conversational turn being received.
[0085] The system obtains one or more paraphrases of the
conversational turn as described above with reference to step 208
(step 306).
[0086] The system generates a respective training example for each
of the paraphrases (step 308). The training example for a given
paraphrase identifies the correct annotation determined by the
system for the conversational turn as the correct annotation for
the paraphrase.
[0087] The system adds the training examples for the paraphrases to
the NLU engine training data (step 310).
[0088] FIG. 4 is a flow diagram of an example process 400 for
generating additional training data from an accurate annotation.
For convenience, the process 400 will be described as being
performed by a system of one or more computers located in one or
more locations. For example, a dialogue system, e.g., the dialogue
system 100 of FIG. 1, appropriately programmed, can perform the
process 400.
[0089] The system determines that the annotation accurately
characterizes the conversational turn (step 402).
[0090] As described above, the system can apply a set of one or
more criteria when determining whether the annotation accurately
characterized the conversational turn. That is, when one or more of
the criteria in the set are satisfied, the system can determine
that the annotation accurately characterized the conversational
turn.
[0091] For example, the set of criteria can include a criterion
that is based on the confidence score generated by the NLU engine
for the annotation. In particular, this criterion can be satisfied
when the score is above a threshold value. This threshold value can
be the same value or a different value from the one used when
determining that a given annotation was not accurate.
[0092] As another example, the set of criteria can include one or
more criteria that are based on the progress of the conversation
after the annotation was generated, i.e., criteria that are
dependent on whether the conversation after the annotation was
generated indicates that the annotation for the current
conversational turn is accurate.
[0093] An example of such a criterion is a criterion that is
satisfied whenever (i) the conversation results in a task
successfully being completed and (ii) none of the slot values
assigned to slots in the annotation of the current conversational
turn are changed or removed by the user in a subsequent turn. For
example, this criterion can be satisfied if, during a conversation
to order a pizza, (i) the current turn is understood by the NLU as
the user requesting a pizza having pineapple as one of the
toppings, (ii) the user ends up placing an order at the end of the
conversation, and (iii) the placed order includes a pizza with
toppings that include pineapple. In this example, the fact that
pineapple was included as a topping the placed order (and none of
the other slot values in the current annotation were later changed
by the user) is an indicator that the annotation for the current
conversational turn was correct.
[0094] As another example, the set of criteria can include a
criterion that is based on the output of the post-hoc annotation
machine learning model that has been trained to determine whether a
given annotation accurately characterizes a given conversational
input. In particular, this criterion can be satisfied when the
accuracy score generated by the model exceeds a threshold
value.
[0095] The system determines that the NLU engine is likely to be
inaccurate for conversational turns that are similar to the current
conversational turn (step 404).
[0096] To make this determination, the system identifies training
conversational turns that the NLU engine has already been trained
and that are similar to the current conversational turn according
to a specified similarity measure.
[0097] For example, the similarity measure can be a distance in an
embedding space. That is, the system can determine respective
numeric embedding representations of the current conversational
turn and the training conversational turns, with each numeric
embedding representation being a vector in the embedding space. In
this example, the system can determine that two conversational
turns are similar when the numeric embedding representations for
the two conversational turns are less than a threshold distance
apart in the embedding space.
[0098] The numeric embedding representations can be generated using
any appropriate technique that generates vectors that represent the
meaning of input sentences, i.e., input conversational turns, such
that two sentences that have similar or the same meaning will have
embedding vectors that are close together in the embedding space.
As an example, the embedding representations can be generated using
n-gram features of the n-grams in the conversational turn. An
example of such an embedding representation generation technique is
described in Unsupervised Learning of Sentence Embeddings using
Compositional n-Gram Features, Matteo Pagliardini, et al, available
at https://arxiv.org/pdf/1703.02507.pdf. As another example, the
embedding representations can be generated based on word vectors
that capture the semantics of the words in the conversational turn.
An example of such an embedding representation generation technique
is described in Exploiting Sentence and Context Representations in
Deep Neural Models for Spoken Language Understanding, Lina M.
Rojas-Barahona, et al, available at
http://mi.eng.cam.ac.uk/.about.sjy/papers/rgms16.pdf.
[0099] In some implementations, the system determines that the NLU
engine is likely to be inaccurate for conversational turns that are
similar to the current conversational turn when there are less than
a threshold number of training conversational turns that are
similar to the conversational turn according to the similarity
measure. That is, the NLU engine is likely to be inaccurate when
the NLU engine has not already been trained on very many
conversational turns that are similar to the current conversational
turn.
[0100] In some implementations, the system determines that the NLU
engine is likely to be inaccurate for conversational turns that are
similar to the current conversational turn when the NLU engine
performs poorly on training conversational turns that are similar
to the current conversational turn according to the similarity
measure.
[0101] That is, the system can maintain data that identifies, for
each training conversational turn, whether the annotation generated
by the NLU engine for the conversational turn was inaccurate, i.e.,
according to the determination described above with reference to
FIG. 3. The system can then determine that the NLU engine performs
poorly on training conversational turns that are similar to the
current conversational turn when the NLU engine has inaccurately
annotated more than a threshold proportion of the similar training
conversational turns.
[0102] In response to determining that (i) the annotation
accurately characterized the current conversational turn but also
(ii) the NLU engine is likely to be inaccurate for conversational
turns that are similar to the current conversational turn, the
system determines that the current turn is a candidate and performs
steps 406 through 410 of the process 400.
[0103] The system obtains one or more paraphrases of the
conversational turn as described above with reference to step 208
(step 406).
[0104] The system generates a respective training example for each
of the paraphrases (step 408). Because the NLU engine accurately
annotated the conversational turn, the training example for a given
paraphrase identifies the annotation generated for the
conversational turn by the NLU engine as the correct annotation for
the paraphrase. Thus, although the training examples are likely to
be helpful in improving the performance of the NLU engine, no
additional human or computational resources are required to
determine the correct annotation for the paraphrases even if a
large number of paraphrases were obtained in step 406.
[0105] The system adds the training examples for the paraphrases to
the NLU engine training data (step 410).
[0106] The above description describes that the annotations and
outputs generated by the system are represented using intent-slot
combinations. However, one of ordinary skill in the art would
appreciate that many other representations of the semantics of a
user input and a system output are possible. For example, the
techniques described above could be used with the semantics of
inputs and outputs represented as logical forms (first order logic,
lambda calculus, and so on), graph-based representations, and other
representation techniques.
[0107] This specification uses the term "configured" in connection
with systems and computer program components. For a system of one
or more computers to be configured to perform particular operations
or actions means that the system has installed on it software,
firmware, hardware, or a combination of them that in operation
cause the system to perform the operations or actions. For one or
more computer programs to be configured to perform particular
operations or actions means that the one or more programs include
instructions that, when executed by data processing apparatus,
cause the apparatus to perform the operations or actions.
[0108] Embodiments of the subject matter and the functional
operations described in this specification can be implemented in
digital electronic circuitry, in tangibly-embodied computer
software or firmware, in computer hardware, including the
structures disclosed in this specification and their structural
equivalents, or in combinations of one or more of them. Embodiments
of the subject matter described in this specification can be
implemented as one or more computer programs, i.e., one or more
modules of computer program instructions encoded on a tangible non
transitory storage medium for execution by, or to control the
operation of, data processing apparatus. The computer storage
medium can be a machine-readable storage device, a machine-readable
storage substrate, a random or serial access memory device, or a
combination of one or more of them. Alternatively or in addition,
the program instructions can be encoded on an artificially
generated propagated signal, e.g., a machine-generated electrical,
optical, or electromagnetic signal, that is generated to encode
information for transmission to suitable receiver apparatus for
execution by a data processing apparatus.
[0109] The term "data processing apparatus" refers to data
processing hardware and encompasses all kinds of apparatus,
devices, and machines for processing data, including by way of
example a programmable processor, a computer, or multiple
processors or computers. The apparatus can also be, or further
include, special purpose logic circuitry, e.g., an FPGA (field
programmable gate array) or an ASIC (application specific
integrated circuit). The apparatus can optionally include, in
addition to hardware, code that creates an execution environment
for computer programs, e.g., code that constitutes processor
firmware, a protocol stack, a database management system, an
operating system, or a combination of one or more of them.
[0110] A computer program, which may also be referred to or
described as a program, software, a software application, an app, a
module, a software module, a script, or code, can be written in any
form of programming language, including compiled or interpreted
languages, or declarative or procedural languages; and it can be
deployed in any form, including as a stand alone program or as a
module, component, subroutine, or other unit suitable for use in a
computing environment. A program may, but need not, correspond to a
file in a file system. A program can be stored in a portion of a
file that holds other programs or data, e.g., one or more scripts
stored in a markup language document, in a single file dedicated to
the program in question, or in multiple coordinated files, e.g.,
files that store one or more modules, sub programs, or portions of
code. A computer program can be deployed to be executed on one
computer or on multiple computers that are located at one site or
distributed across multiple sites and interconnected by a data
communication network.
[0111] In this specification, the term "database" is used broadly
to refer to any collection of data: the data does not need to be
structured in any particular way, or structured at all, and it can
be stored on storage devices in one or more locations. Thus, for
example, the index database can include multiple collections of
data, each of which may be organized and accessed differently.
[0112] Similarly, in this specification the term "engine" is used
broadly to refer to a software-based system, subsystem, or process
that is programmed to perform one or more specific functions.
Generally, an engine will be implemented as one or more software
modules or components, installed on one or more computers in one or
more locations. In some cases, one or more computers will be
dedicated to a particular engine; in other cases, multiple engines
can be installed and running on the same computer or computers.
[0113] The processes and logic flows described in this
specification can be performed by one or more programmable
computers executing one or more computer programs to perform
functions by operating on input data and generating output. The
processes and logic flows can also be performed by special purpose
logic circuitry, e.g., an FPGA or an ASIC, or by a combination of
special purpose logic circuitry and one or more programmed
computers.
[0114] Computers suitable for the execution of a computer program
can be based on general or special purpose microprocessors or both,
or any other kind of central processing unit. Generally, a central
processing unit will receive instructions and data from a read only
memory or a random access memory or both. The essential elements of
a computer are a central processing unit for performing or
executing instructions and one or more memory devices for storing
instructions and data. The central processing unit and the memory
can be supplemented by, or incorporated in, special purpose logic
circuitry. Generally, a computer will also include, or be
operatively coupled to receive data from or transfer data to, or
both, one or more mass storage devices for storing data, e.g.,
magnetic, magneto optical disks, or optical disks. However, a
computer need not have such devices. Moreover, a computer can be
embedded in another device, e.g., a mobile telephone, a personal
digital assistant (PDA), a mobile audio or video player, a game
console, a Global Positioning System (GPS) receiver, or a portable
storage device, e.g., a universal serial bus (USB) flash drive, to
name just a few.
[0115] Computer readable media suitable for storing computer
program instructions and data include all forms of non volatile
memory, media and memory devices, including by way of example
semiconductor memory devices, e.g., EPROM, EEPROM, and flash memory
devices; magnetic disks, e.g., internal hard disks or removable
disks; magneto optical disks; and CD ROM and DVD-ROM disks.
[0116] To provide for interaction with a user, embodiments of the
subject matter described in this specification can be implemented
on a computer having a display device, e.g., a CRT (cathode ray
tube) or LCD (liquid crystal display) monitor, for displaying
information to the user and a keyboard and a pointing device, e.g.,
a mouse or a trackball, by which the user can provide input to the
computer. Other kinds of devices can be used to provide for
interaction with a user as well; for example, feedback provided to
the user can be any form of sensory feedback, e.g., visual
feedback, auditory feedback, or tactile feedback; and input from
the user can be received in any form, including acoustic, speech,
or tactile input. In addition, a computer can interact with a user
by sending documents to and receiving documents from a device that
is used by the user; for example, by sending web pages to a web
browser on a user's device in response to requests received from
the web browser. Also, a computer can interact with a user by
sending text messages or other forms of message to a personal
device, e.g., a smartphone that is running a messaging application,
and receiving responsive messages from the user in return.
[0117] Data processing apparatus for implementing machine learning
models can also include, for example, special-purpose hardware
accelerator units for processing common and compute-intensive parts
of machine learning training or production, i.e., inference,
workloads.
[0118] Machine learning models can be implemented and deployed
using a machine learning framework, e.g., a TensorFlow framework, a
Microsoft Cognitive Toolkit framework, an Apache Singa framework,
or an Apache MXNet framework.
[0119] Embodiments of the subject matter described in this
specification can be implemented in a computing system that
includes a back end component, e.g., as a data server, or that
includes a middleware component, e.g., an application server, or
that includes a front end component, e.g., a client computer having
a graphical user interface, a web browser, or an app through which
a user can interact with an implementation of the subject matter
described in this specification, or any combination of one or more
such back end, middleware, or front end components. The components
of the system can be interconnected by any form or medium of
digital data communication, e.g., a communication network. Examples
of communication networks include a local area network (LAN) and a
wide area network (WAN), e.g., the Internet.
[0120] The computing system can include clients and servers. A
client and server are generally remote from each other and
typically interact through a communication network. The
relationship of client and server arises by virtue of computer
programs running on the respective computers and having a
client-server relationship to each other. In some embodiments, a
server transmits data, e.g., an HTML page, to a user device, e.g.,
for purposes of displaying data to and receiving user input from a
user interacting with the device, which acts as a client. Data
generated at the user device, e.g., a result of the user
interaction, can be received at the server from the device.
[0121] While this specification contains many specific
implementation details, these should not be construed as
limitations on the scope of any invention or on the scope of what
may be claimed, but rather as descriptions of features that may be
specific to particular embodiments of particular inventions.
Certain features that are described in this specification in the
context of separate embodiments can also be implemented in
combination in a single embodiment. Conversely, various features
that are described in the context of a single embodiment can also
be implemented in multiple embodiments separately or in any
suitable subcombination. Moreover, although features may be
described above as acting in certain combinations and even
initially be claimed as such, one or more features from a claimed
combination can in some cases be excised from the combination, and
the claimed combination may be directed to a subcombination or
variation of a subcombination.
[0122] Similarly, while operations are depicted in the drawings and
recited in the claims in a particular order, this should not be
understood as requiring that such operations be performed in the
particular order shown or in sequential order, or that all
illustrated operations be performed, to achieve desirable results.
In certain circumstances, multitasking and parallel processing may
be advantageous. Moreover, the separation of various system modules
and components in the embodiments described above should not be
understood as requiring such separation in all embodiments, and it
should be understood that the described program components and
systems can generally be integrated together in a single software
product or packaged into multiple software products.
[0123] Particular embodiments of the subject matter have been
described. Other embodiments are within the scope of the following
claims. For example, the actions recited in the claims can be
performed in a different order and still achieve desirable results.
As one example, the processes depicted in the accompanying figures
do not necessarily require the particular order shown, or
sequential order, to achieve desirable results. In some cases,
multitasking and parallel processing may be advantageous.
* * * * *
References