A MathSpeak Settings Tutorial has been added to the Examples section. Some of the settings described have not yet been implemented to the online Nemeth book.

Rule 15 of the Nemeth book has been posted and all the examples are considered correct according to the current specification and the default MathSpeak settings. Dr. Nemeth has given a lot of feedback for Rule 14 and a lot of corrections have been made, but it is still being considered as being under heavy construction.

The current approximate stats of the C++ Translator for the following rules are:

For Explicivity off and Semantic Interpretation set to the intermediate value:

Rule 11 - 100% (of 5 expressions)
Rule 12 - 95% (of about 40 expressions)
Rule 13 - 98% (of about 150 expressions)
Rule 14 - 80% (of about 70 expressions)
Rule 15 - 100% (of 17 expressions)
Rule 18 - 100% (of about 40 expressions)

A quick tutorial to MathSpeak has been added to the Examples section.

The rules associated with the cancellation indicators has been updated, including the addition of a rule for expressions cancelled out with another expression (such as 25 being replaced by 5).

The current approximate stats of the C++ Translator for the following rules are:

For Explicivity off and Semantic Interpretation off:

Rule 11 - 100% (of 5 expressions)
Rule 12 - 100% (of about 40 expressions)
Rule 13 - 99% (of about 150 expressions)
Rule 18 - 100% (of about 40 expressions)

For Explicivity off and Semantic Interpretation set to the intermediate value:

Rule 11 - 100% (of 5 expressions)
Rule 12 - 95% (of about 40 expressions)
Rule 13 - 98% (of about 150 expressions)
Rule 18 - 100% (of about 40 expressions)

Dr. Nemeth has been sent the current suggestions for Rule 14 regarding modifiers.

The estimated correctness of Rule 18 has been brought past the threshold of 95%, the correctness was previously around 60%.

The current approximate stats of the C++ Translator for the following rules are:

For Explicivity off and Semantic Interpretation off:

Rule 12 - 100% (of about 40 expressions)
Rule 13 - 99% (of about 150 expressions)
Rule 18 - 100% (of about 40 expressions)

For Explicivity off and Semantic Interpretation set to the intermediate value:

Rule 12 - 95% (of about 40 expressions)
Rule 13 - 98% (of about 150 expressions)
Rule 18 - 100% (of about 40 expressions)

The following rules have been posted to the new Nemeth site:

The accuracy of these rules and the stability of their associated MathSpeak rules, are not yet to the level of Rule 12 and Rule 13.

Near term goals include:

• Increase the accuracy of these rules and create estimates of their correctness.
• Implement the rules associated with the Explicivity settings and the high Semantic Interpretation setting.
• Finish working on a quick tutorial for MathSpeak.

The Braille in the new MathSpeak Nemeth site can now be displayed in different formats. The formats that can be chosen are currently ASCII Braille (the default), Unicode Braille, descriptions of the dot position, and numbers using the NUMBRL system. The gh Braille fonts currently only support ASCII Braille. Fonts that support the unicode encoding of Braille can be found here.

A new website has been created for the examples of the Nemeth Book created with the latest MathML-to-MathSpeak Translator. The URL for this website is:

http://www.gh-mathspeak.com/examples/NemethBook/

The Nemeth Book website is being relocated so that the two different versions can coexist while the new website is catching up to the existing website and to make room for future MathSpeak examples projects, such as for the MathSpeak specification and introductions to either MathSpeak or Nemeth Braille.

Around 60 people tentatively said they would be able to take the MathSpeak test. We expect around 40 to 50 to actually take the test.

The following items were accomplished during a meeting that lasted over 5 hours:

• The various elements of the test was put into booklet form.
• The script for the prerecorded audio was modified so that the proctor does not have to say anything during the actual test.
• The above script was recorded.
• The recorded script was rearranged and combined with the Text-To-Speech generated audio to be a single audio recording.

The C++ Translator is going to allow finer control over Semantic Interpretation and Explicivity. Both Semantic Interpretation and Explicivity will allow an off value, an intermediate value, and a full or all-the-way value. The default values for the website will also change.

In the effort to integrate the changes of the C++ Translator into the website, the new Nemeth Book will be created at a different location. This URL will be made public at a future date.

The current approximate stats of the C++ Translator for the following rules are:

For Explicivity off and Semantic Interpretation off:

Rule 12 - 100% (of about 40 expressions)
Rule 13 - 99% (of about 150 expressions)

For Explicivity off and Semantic Interpretation set to the intermediate value:

Rule 12 - 95% (of about 40 expressions)
Rule 13 - 85% (of about 150 expressions)

These statistics apply to only the visual output, the audio input is harder to test and less effort has been put in it thus far, so it is unknown how well the audio output fares.

After spending Tuesday to generate all the needed files, there was a meeting at Purdue on Wednesday that lasted over 4 hours. Numerous errors where found throughout the test (which have since been corrected). Also, it was decided that the audio would be revised so that the proctor does not have to say anything after starting the recording. The exact details of these changes were not decided. It was decided that the pauses of the MathSpeak examples needed to be hand modified to be similar to that of the regular spoken math examples (this has since been done). It was decided that the "o-over" reserved word for complex fractions had to be modified. We are meeting on Monday to decide on the changes to the recorded script and to re-record this script.

The C++ version of the Translator has been improved in regards to Rule 13 of the Nemeth Book and likewise the percent correct has increased from about 74% to 100%. Unfortunately, the changes made to Rule 13 decreased the correctness of Rule 12 to 90%. The current approximate stats of the C++ Translator for the following rules are:

Rule 12 - 90% (of about 40 expressions)
Rule 13 - 100% (of about 150 expressions)
Rule 14 - 80% (of about 70 expressions)

This is with both verbosities, explicivity off, and semantic interpretation off. Future changes to the rules could affect the correctness of the output.

A new mechanism for checking the correctness of the C++ translator output has been implemented into a local web-based system. It displays the correct output, displays the translator output, allows the correct output to be modified directly, and displays the percent that match at the bottom. The translator output is displayed in red if it doesn't match the correct output, otherwise it is displayed in green. It also allows the correct output to be hid, so that it is easy to focus on the remaining rules to be implemented.

There was a meeting today from 2 pm to 6 pm and many things were accomplished, including:

• The improvised instructions that were created at the last minute on Monday were revised. These instructions will be recorded on Monday October 31, 2005.
• It was decided to add pauses in the text-to-speech generated speech for the Mathematical expressions during the test and then run the pilot test to ensure that the question we had addressed the ambiguity issue even with the pauses. This is to address the possible criticism that adding pauses would eliminate the ambiguity. The pilot test will be given on Wednesday November 2, 2005.
• The experimental design was revised to correct for a flaw in the old design.
• The scripts for the proctor to read were reviewed and revised.

Rule 12 has been updated according to specification changes to fractions. The output for the Semantics Off and Explicivity Off has been created with the C++ MathSpeak Translator. There are some mistakes with the examples when displayed with either Semantics On or Explicivity Off that will be fixed when the new translator has these rules implemented.

A new mechanism for indicating the complexity of fractions and radicals was proposed to Dr. Lyle Lloyd's research team. They were insistently opposed to the idea. We all agreed that the instructions introducing spoken math and MathSpeak needed to be revised once more and that we would meet on Monday to record it for the final time. The study at Purdue should be able to be ran in about two or three weeks.

The scripts for the instructions of the regular spoken math and MathSpeak have been revised, although not yet finalized. Also the initial script describing the math tokens has been created and Sam and Mick are meeting today to finalize it. Later on today, we are going to record the scripts so that we can review them and then we are going to meet again Monday to re-record anything that needs to be.

The C++ version of the Translator has been improved in regards to Rule 13 of the Nemeth Book and likewise the percent correct has increased from about 60% to about 74%. The current approximate stats of the C++ Translator for the following rules are:

Rule 12 - 90% (of about 40 expressions)
Rule 13 - 74% (of about 150 expressions)
Rule 14 - 80% (of about 70 expressions)

This is with both verbosities, explicivity off, and semantic interpretation off. Future changes to the rules could affect the correctness of the output.

The C++ version of the Translator has been improved in regards to Rule 14 of the Nemeth Book and likewise the percent correct has increased from about 50% to about 80%. However, the specification that applies to Rule 14 is less certain than that which applies to many other chapters and future changes are expected. The current approximate stats of the C++ Translator for the following rules are:

Rule 12 - 90% (about 40 expressions)
Rule 13 - 60% (about 150 expressions)
Rule 14 - 80% (about 70 expressions)

This is with both verbosities, explicivity off, and semantic interpretation off. Future changes to the rules could affect the correctness of the output.

The part of the translator that deals with conversion of the MathSpeak internal representation into the various output formats - voice markup, html, and plain text - was overhauled to increase the amount of code that was shared between the different types of output. This will make it a lot easier to keep the outputs consistent.

A mechanism has been created to check the correctness of MathSpeak. Whenever the correct MathSpeak is known, it can be created for this checking mechanism, which it will compare with the current results. This is important to be able to catch changes in code that might introduce errors in other examples and to keep track of progress.

Dr. Nemeth has been sent a new CD regarding the Alternative formats of fractions, because the CD he received previously could not be read. He stated that he was okay with a modified form of the alternative fraction suggestion. Likewise, one of the alternative formats for saying fractions will most likely be implemented into MathSpeak. The exact format has yet to be decided.

Significant progress has been made with the new MathML-to-MathSpeak C++ Translator for Rule 13. The current approximate stats of the C++ Translator for the following rules are:

Rule 12 - 90% (about 40 expressions)
Rule 13 - 60% (about 150 expressions)
Rule 14 - 50% (about 70 expressions)

This is with both verbosities, explicivity off, and semantic interpretation off. Future changes to the rules could affect the correctness of the output.

These results generally apply to audio rendering, but the correctness of the audio is not nearly as cut and dry as the written result. It is expected that the audio will continue to improve with Cepstral's hard work and with tweaking of the translator-produced prosodic pauses.

The C++ translator has implemented some of the rules for translating modifiers and stacked expressions. It can translate about half of the examples in Rule 14 correctly, according to the current MathSpeak specification and with Semantic Interpretation and Explicivity off.

Due to some difficulty understanding complex nested fractions, alternative approaches to saying fractions have been created with the help of Jeff Dittel. Dr. Nemeth has been sent audio of various math equations with equations using two alternative methods of saying fractions. Also sent to him were documents that specify how MathSpeak will handle literal text and inquiring more on his suggestions to semantically interpret vertical lines.

Al Lovati from the ISB has been given multiple possible dates and times to have a teleconference and for gh to go down to ISB. He currently waiting for others at ISB to respond to these times.

On the Purdue part of the MathSpeak study, there will be a meeting on Wednesday October the 5th to discuss the results of the preliminary run of the MathSpeak study.

Significant work has been put forward to insert breaks dynamically depending on what complexity of constructs, although it is far from perfect. Also, the reserved words lexicon mechanism has been extended to allow modification of the strength of the breaks before and after without touching the C++ code.

All of the required audio was generated for a test run of the MathSpeak study on Wednesday. This includes the instructions and the equations for the MathSpeak and Control portions of the test. The audio for the MathSpeak version needs to be modified, especially with adding more breaks within the equations and speeding up the speech.

The website was updated to make it more accessible, particularly in the Nemeth example pages.

A meeting for the MathSpeak study was held at gh. We have changed the design of the study to be much more statistically viable. We've finalized the equations that we are using for the final study at Purdue. The multiple-choice test sheet will be created, along with the instructions for the regular spoken math. The instructions will be recorded and the speech of the equations will be generated using TTS at the beginning of the week. A test run of the study will be given on Wednesday September 28.

The C++ translator has been able to produce the correct html/text for the MathSpeak for 93% of the equations in Rule 12 of the Nemeth book in the brief verbosity, explicitness off, and semantics off. The translator has implemented a rule for continued fractions, which has not been implemented in the XSLT translator. It currently implements neither the MathML units marked up with csymbol nor indicators for uppercase when spelling out (the latter of which isn't correctly implemented in XSLT, according to recent changes in MathSpeak).

Upon correspondence with Dr. Nemeth, the comparison operators will not have an implied baseline.

With the help of Cepstral, new rules for the placement of breaks in speech were set in place. These rules were implemented in the XSLT translator and new SSML examples were given to them.

The rule for implied numeric subscripts was implemented as well as a switch to easily turn this mechanism on or off.

The rules regarding fractions have been written down and have been linked from the specification page. This section has not been formatted due to the transitional nature of material.

The mechanism to speak superscript / subscript levels has been integrated into the mechanism that handles reserved words. The base process of deciding when to speak the script levels has been created. It only speaks the script levels in front of something that is spoken, which depends on the lexicon being used. For example, "invisible times" may be spoken in more explicit lexicons, but typically it isn't. Thus, sometimes it will speak it's script level and sometimes it will not. If the translator didn't check for this, it might give an extra script level indicator in some obscure cases; this is one area where the restructuring of the Internal Representation when moving from XSLT to C++ has paid off. The rules for implied baselines, numeric subscripts, adjacent script levels from different elements, and staggered script levels have yet to be implemented.

Cepstral has been given more feedback on their latest voice. This will continue for at least the rest of the week.

The MathSpeak Translator (previously called "Parser") is now able to translate a matrix as marked up in presentational MathML (a table "fenced" in with either parentheses or brackets) as a matrix and is able to output a matrix in several different versions, including interpretations that use ordinal numbers to indicate row and column indices and/or that indicate the grouping symbols used in presentational markup (for example, "begin-2-by-2-matrix enclosed by left-parenthesis and right-parenthesis … end-matrix").

The translator also successfully marks up the super/subscript level of every element in the MathSpeak Internal Representation, however, it does not yet do the more complicated task of deciding when to speak this super/subscript level.

Cepstral has implemented an interface to track feedback on specific examples, which has the functionality of generating examples from different versions of the William voice.

We have come up with 20 families of equations that when spoken in regular spoken math can be interpreted in at least 4 different ways. Sam commented that equations may be too difficult for our intended audience.

The initial stage of converting the presentational MathML is about 80% completed, with mostly uncommon presentational MathML elements remaining. The internal representation of the math has respectively been solidifying and has been documented. The infrastructure used to store rules for MathSpeak reserved words in XML, to read in this data, and to implement it, has largely been completed. The infrastructure used to store and implement conversion of Unicode characters into spoken characters (such as saying "plus") has largely been created. The main meat of the code that modifies and augments the internal representation of the math has not yet begun. This includes the code that determines when the superscript or subscript level is spoken or not.

We've received an update from Cepstral on the William voice and Dr. Nemeth's feedback on chapters 12, 13, 15, 17-20 in the Nemeth book. Some of Dr. Nemeth's feedback on audio rendering was still applicable and this feedback is being conveyed to Cepstral.

A meeting for the MathSpeak study was held at gh. Feedback from the Purdue's institutional review board (IRB) was recently returned and only a few minor modifications are needed. We decided to recreate the equations that would be used for the Purdue study. We decided that each verbal rendering without MathSpeak must have at least 4 interpretations and that only one of these interpretations would appear in the test.

Some minor changes where made to the website. These include adding a page about Jonathan Williford to the MathSpeak™ team section, adding a link to the 2005 NISO Specification in the reference links page, and removing the link to the discussion forum board.

A lot of progress has been made in the design and documentation of the the C++ MathML-to-MathSpeak Parser.

Dr. Nemeth has responded to the proposal to add special rules for matrices, determinants, and binomial coefficients. We will have special rules for saying these elements in both Semantic Interpretation on and off (previously this only applied to Semantic Interpretation on). The way to say matrices and determinants has essentially been decided and although there is some disagreement as to how to say binomial coefficients.

The development of the C++ MathML-to-MathSpeak Parser has begun. The reasons for creating the C++ version are speed and the ability to put it in the ghPlayer™. It will implement an Internal Representation that is vastly different than that of the XSLT2 parser, overcoming the limitations that were encountered with the old design. Unfortunately, this version will be incompatible with non-Windows systems and servers.

Currently, the initial phase has been implemented for a small subset of MathML. A mechanism for defining and reading XML data of Reserved Words has been implemented, although it currently only supports static reserved words and has not been thoroughly tested.

Rule 23 of the Nemeth book has been posted to the website for review.

Rule 9 of the Nemeth book has been posted to the website for review.

Rule 6 and 8 of the Nemeth book has been posted to the website for review.

Rule 7 of the Nemeth book has been posted to the website for review.

A total of 3 people participated in the preliminary Math Study and exceeded expectations in showing the desired results.

Dr. Nemeth was sent the audio CD's for Rule 12, 13, 15, 17, 18, 19, and 20.

Rule 10 of the Nemeth book has been posted to the website for review. It has 1 problem regarding long division which has been omitted.

The following has been updated

• The Research Page has been updated and information regarding the variables has been added.
• The TTS page has been updated with audio examples of the progress of the tuning by Cepstral LLC.
• The preferences link has been made more obvious on the example pages.
• A link to Dr. Nemeth's original MathSpeak article has been added to the example pages.
• The Recent News and Events sidebar has been updated.

Rule 2 and 11 of the Nemeth book has been posted to the website for review. Rule 2 has 1 problem regarding long division which has been omitted.

Rule 20 of the Nemeth book has been posted to the website.

The IRB has been reviewed. MathSpeak instructions have been created.

The draft of the MathSpeak™ Specification has been posted to the specification page of the website.

The infrastructure on the website has been updated to generate audio on the fly. This will greatly reduce the amount of time required to update the chapters and will open up the possibility of allowing the user to select the rate of speech.

The MathSpeak parser now supports every Unicode Character! If we have not manually set a name for the character, the parser will lookup the unicode description as given by the Unicode Consortium.

The semantic rules for matrices, determinants, and binomial coefficients have been created and implemented.

The rules for modifiers and stacked expressions have been changed to make it easier to understand.

The rules have been updated according to Dr. Nemeth's comments on rules 15, 17 and 18.

Cepstral has recently updated the William voice with more tuning for MathSpeak.

Rule 19 of the Nemeth book has been posted to the website. This makes a total of 8 rules that have been posted.

All of the published rules were updated according to Nemeth's response to Rule 12 and 13 that was received on June 23, 2005.

The changes include how the stretched grouping symbols and layout problems are spoken. Also, dictionary words that don't conflict with MathSpeak reserved words or numbers are now spoken as words (except when explictness is on), otherwise the are spelled out. There were also several mistakes that were corrected.

Rule 3 of the Nemeth book has been posted to the website.

The examples for Phase 1 of the Math Ambiguity Study have been created.

The audio has also been generated and tested. Cepstral has been contacted about changes necessary to the William voice and said they would get back to us with a new voice by July 7, 2005. Some problems with the MathSpeak parser have been noted and will be promptly updated.

Rules 14, 15, 17, and 18 have been posted to the website and are ready for review.

Rules 12 and 13 have been updated with verbosity, semantic interpretation and explicitness variables. The examples can be displayed in any combination of these by choosing the drop down menus.

The MathML to MathSpeak parser has been implemented for Rule 12 and 13. Instead of hand making the MathSpeak, it is directly generated from MathML. This makes the rules more consistent and our work more resilient to modifications.

Upon feedback today, the links to Ogg Vorbis (.ogg) files on the example pages been removed from the site. This is to avoid confusion for most computer users who do not have ogg support in their audio clients like MusicMatch and Windows Media Player. Currently, all audio files that are available in both wav and mp3 formats will still be available in ogg format, just not through links on the pages. So, if there is an audio file for an example like "18-126-x-001v.mp3", then you will be able to get the wav file as "18-126-x-001v.wav" and also as "18-126-x-001v.ogg".

Additional changes include clearer instructions for downloading fonts and other minor changes.

Dr. Abraham Nemeth, a member of the MathSpeak team and inventor of Nemeth Braille, has given his first preliminary feedback for the website content. Revisions will follow shortly.

MathSpeak Website officially launched today, being announced to the public at the Ribbon cutting ceremony for the company's graduation from the Purdue Research Park incubator program.

MathSpeak Website Beta was posted today with examples shown for internal review. Once the gh MathSpeak team looks it over, we are ready to post it publicly.

MathSpeak Website Alpha was posted today with incomplete examples. This is solely for internal review and testing for browser compatibility and accessibility. When the changes are listed, we will incorporate them, then move into beta stage.