Machine Learning For Natural Language Processing

Lecture Notes Columbia E6998 Itebooks download

https://ebookbell.com/product/machine-learning-for-natural-
language-processing-lecture-notes-
columbia-e6998-itebooks-23836890

Explore and download more ebooks at ebookbell.com

Here are some recommended products that we believe you will be
interested in. You can click the link to download.

Deep Learning For Natural Language Processing Develop Deep Learning

Models For Natural Language In Python V11 Jason Brownlee

https://ebookbell.com/product/deep-learning-for-natural-language-
processing-develop-deep-learning-models-for-natural-language-in-
python-v11-jason-brownlee-11575184

Applied Natural Language Processing With Python Implementing Machine

Learning And Deep Learning Algorithms For Natural Language It Pro York
University Skillsoft Books York University Ii

https://ebookbell.com/product/applied-natural-language-processing-
with-python-implementing-machine-learning-and-deep-learning-
algorithms-for-natural-language-it-pro-york-university-skillsoft-
books-york-university-ii-22049616

Natural Language Processing And Machine Learning For Developers

Campesato

https://ebookbell.com/product/natural-language-processing-and-machine-
learning-for-developers-campesato-32802672

Natural Language Processing In Python Master Data Science And Machine

Learning For Spam Detection Sentiment Analysis Latent Semantic
Analysis And Article Spinning Machine Learning In Python
Lazyprogrammer
https://ebookbell.com/product/natural-language-processing-in-python-
master-data-science-and-machine-learning-for-spam-detection-sentiment-
analysis-latent-semantic-analysis-and-article-spinning-machine-
learning-in-python-lazyprogrammer-5481600
Natural Language Annotation For Machine Learning 1st Edition James
Pustejovsky Amber Stubbs

https://ebookbell.com/product/natural-language-annotation-for-machine-
learning-1st-edition-james-pustejovsky-amber-stubbs-22437086

Natural Language Annotation For Machine Learning A Guide To

Corpusbuilding For Applications Early Release James Pustejovsky

https://ebookbell.com/product/natural-language-annotation-for-machine-
learning-a-guide-to-corpusbuilding-for-applications-early-release-
james-pustejovsky-2561356

Natural Language Annotation For Machine Learning James Pustejovsky

Amber Stubbs Pustejovsky

https://ebookbell.com/product/natural-language-annotation-for-machine-
learning-james-pustejovsky-amber-stubbs-pustejovsky-23530826

Machine Learning For Ecology And Sustainable Natural Resource

Management 1st Edition Grant Humphries

https://ebookbell.com/product/machine-learning-for-ecology-and-
sustainable-natural-resource-management-1st-edition-grant-
humphries-7247946

Machine Learningbased Natural Scene Recognition For Mobile Robot

Localization In An Unknown Environment 1st Ed 2020 Xiaochun Wang

https://ebookbell.com/product/machine-learningbased-natural-scene-
recognition-for-mobile-robot-localization-in-an-unknown-
environment-1st-ed-2020-xiaochun-wang-10805550
 Tagging with Hidden Markov Models
Michael Collins

1 Tagging Problems
In many NLP problems, we would like to model pairs of sequences. Part-of-speech
(POS) tagging is perhaps the earliest, and most famous, example of this type of
problem. In POS tagging our goal is to build a model whose input is a sentence,
for example
the dog saw a cat
and whose output is a tag sequence, for example

DNVDN (1)

(here we use D for a determiner, N for noun, and V for verb). The tag sequence is
the same length as the input sentence, and therefore specifies a single tag for each
word in the sentence (in this example D for the, N for dog, V for saw, and so on).
We will use x1 . . . xn to denote the input to the tagging model: we will often
refer to this as a sentence. In the above example we have the length n = 5, and
x1 = the, x2 = dog, x3 = saw, x4 = the, x5 = cat. We will use y1 . . . yn to denote
the output of the tagging model: we will often refer to this as the state sequence or
tag sequence. In the above example we have y1 = D, y2 = N, y3 = V, and so on.
This type of problem, where the task is to map a sentence x1 . . . xn to a tag se-
quence y1 . . . yn , is often referred to as a sequence labeling problem, or a tagging
problem.
We will assume that we have a set of training examples, (x(i) , y (i) ) for i =
(i) (i)
1 . . . m, where each x(i) is a sentence x1 . . . xni , and each y (i) is a tag sequence
(i) (i) (i)
y1 . . . yni (we assume that the i’th example is of length ni ). Hence xj is the j’th
(i)
word in the i’th training example, and yj is the tag for that word. Our task is to
learn a function that maps sentences to tag sequences from these training examples.

1
2 Generative Models, and The Noisy Channel Model
Supervised problems in machine learning are defined as follows. We assume train-
ing examples (x(1) , y (1) ) . . . (x(m) , y (m) ), where each example consists of an input
x(i) paired with a label y (i) . We use X to refer to the set of possible inputs, and Y
to refer to the set of possible labels. Our task is to learn a function f : X → Y that
maps any input x to a label f (x).
Many problems in natural language processing are supervised learning prob-
lems. For example, in tagging problems each x(i) would be a sequence of words
(i) (i) (i) (i)
x1 . . . xni , and each y (i) would be a sequence of tags y1 . . . yni (we use ni to
refer to the length of the i’th training example). X would refer to the set of all
sequences x1 . . . xn , and Y would be the set of all tag sequences y1 . . . yn . Our
task would be to learn a function f : X → Y that maps sentences to tag sequences.
In machine translation, each input x would be a sentence in the source language
(e.g., Chinese), and each “label” would be a sentence in the target language (e.g.,
English). In speech recognition each input would be the recording of some ut-
terance (perhaps pre-processed using a Fourier transform, for example), and each
label is an entire sentence. Our task in all of these examples is to learn a function
from inputs x to labels y, using our training examples (x(i) , y (i) ) for i = 1 . . . n as
evidence.
One way to define the function f (x) is through a conditional model. In this
approach we define a model that defines the conditional probability

p(y|x)

for any x, y pair. The parameters of the model are estimated from the training
examples. Given a new test example x, the output from the model is

f (x) = arg max p(y|x)

y∈Y

Thus we simply take the most likely label y as the output from the model. If our
model p(y|x) is close to the true conditional distribution of labels given inputs, the
function f (x) will be close to optimal.
An alternative approach, which is often used in machine learning and natural
language processing, is to define a generative model. Rather than directly estimat-
ing the conditional distribution p(y|x), in generative models we instead model the
joint probability
p(x, y)
over (x, y) pairs. The parameters of the model p(x, y) are again estimated from the
training examples (x(i) , y (i) ) for i = 1 . . . n. In many cases we further decompose

2
the probability p(x, y) as follows:

p(x, y) = p(y)p(x|y) (2)

and then estimate the models for p(y) and p(x|y) separately. These two model
components have the following interpretations:

• p(y) is a prior probability distribution over labels y.

• p(x|y) is the probability of generating the input x, given that the underlying
label is y.

We will see that in many cases it is very convenient to decompose models in this
way; for example, the classical approach to speech recognition is based on this type
of decomposition.
Given a generative model, we can use Bayes rule to derive the conditional
probability p(y|x) for any (x, y) pair:

p(y)p(x|y)
p(y|x) =
p(x)
where X X
p(x) = p(x, y) = p(y)p(x|y)
y∈Y y∈Y

Thus the joint model is quite versatile, in that we can also derive the probabilities
p(x) and p(y|x).
We use Bayes rule directly in applying the joint model to a new test example.
Given an input x, the output of our model, f (x), can be derived as follows:

f (x) = arg max p(y|x)

y
p(y)p(x|y)
= arg max (3)
p(x)
y

= arg max p(y)p(x|y) (4)

y

Eq. 3 follows by Bayes rule. Eq. 4 follows because the denominator, p(x), does not
depend on y, and hence does not affect the arg max. This is convenient, because it
means that we do not need to calculate p(x), which can be an expensive operation.
Models that decompose a joint probability into into terms p(y) and p(x|y) are
often called noisy-channel models. Intuitively, when we see a test example x, we
assume that has been generated in two steps: first, a label y has been chosen with
probability p(y); second, the example x has been generated from the distribution

3
p(x|y). The model p(x|y) can be interpreted as a “channel” which takes a label y
as its input, and corrupts it to produce x as its output. Our task is to find the most
likely label y, given that we observe x.
In summary:

• Our task is to learn a function from inputs x to labels y = f (x). We assume

training examples (x(i) , y (i) ) for i = 1 . . . n.

• In the noisy channel approach, we use the training examples to estimate

models p(y) and p(x|y). These models define a joint (generative) model

p(x, y) = p(y)p(x|y)

• Given a new test example x, we predict the label

f (x) = arg max p(y)p(x|y)

y∈Y

Finding the output f (x) for an input x is often referred to as the decoding
problem.

3 Generative Tagging Models

We now see how generative models can be applied to the tagging problem. We
assume that we have a finite vocabulary V, for example V might be the set of
words seen in English, e.g., V = {the, dog, saw, cat, laughs, . . .}. We use K to
denote the set of possible tags; again, we assume that this set is finite. We then give
the following definition:

Definition 1 (Generative Tagging Models) Assume a finite set of words V, and

a finite set of tags K. Define S to be the set of all sequence/tag-sequence pairs
hx1 . . . xn , y1 . . . yn i such that n ≥ 0, xi ∈ V for i = 1 . . . n and yi ∈ K for
i = 1 . . . n. A generative tagging model is then a function p such that:

1. For any hx1 . . . xn , y1 . . . yn i ∈ S,

p(x1 . . . xn , y1 . . . yn ) ≥ 0

2. In addition,
X
p(x1 . . . xn , y1 . . . yn ) = 1
hx1 ...xn ,y1 ...yn i∈S

4
Hence p(x1 . . . xn , y1 . . . yn ) is a probability distribution over pairs of sequences
(i.e., a probability distribution over the set S).
Given a generative tagging model, the function from sentences x1 . . . xn to tag
sequences y1 . . . yn is defined as

f (x1 . . . xn ) = arg max p(x1 . . . xn , y1 . . . yn )

y1 ...yn

Thus for any input x1 . . . xn , we take the highest probability tag sequence as the
output from the model.

Having introduced generative tagging models, there are three critical questions:

• How we define a generative tagging model p(x1 . . . xn , y1 . . . yn )?

• How do we estimate the parameters of the model from training examples?

• How do we efficiently find

arg max p(x1 . . . xn , y1 . . . yn )

y1 ...yn

for any input x1 . . . xn ?

The next section describes how trigram hidden Markov models can be used to
answer these three questions.

4 Trigram Hidden Markov Models (Trigram HMMs)

In this section we describe an important type of generative tagging model, a trigram
hidden Markov model, describe how the parameters of the model can be estimated
from training examples, and describe how the most likely sequence of tags can be
found for any sentence.

4.1 Definition of Trigram HMMs

We now give a formal definition of trigram hidden Markov models (trigram HMMs).
The next section shows how this model form is derived, and gives some intuition
behind the model.

Definition 2 (Trigram Hidden Markov Model (Trigram HMM)) A trigram HMM

consists of a finite set V of possible words, and a finite set K of possible tags, to-
gether with the following parameters:

5
 • A parameter
q(s|u, v)
for any trigram (u, v, s) such that s ∈ K ∪ {STOP}, and u, v ∈ V ∪ {*}.
The value for q(s|u, v) can be interpreted as the probability of seeing the tag
s immediately after the bigram of tags (u, v).

• A parameter
e(x|s)
for any x ∈ V, s ∈ K. The value for e(x|s) can be interpreted as the
probability of seeing observation x paired with state s.

Define S to be the set of all sequence/tag-sequence pairs hx1 . . . xn , y1 . . . yn+1 i

such that n ≥ 0, xi ∈ V for i = 1 . . . n, yi ∈ K for i = 1 . . . n, and yn+1 = STOP.
We then define the probability for any hx1 . . . xn , y1 . . . yn+1 i ∈ S as
n+1
Y n
Y
p(x1 . . . xn , y1 . . . yn+1 ) = q(yi |yi−2 , yi−1 ) e(xi |yi )
i=1 i=1

where we have assumed that y0 = y−1 = *.

As one example, if we have n = 3, x1 . . . x3 equal to the sentence the dog

laughs, and y1 . . . y4 equal to the tag sequence D N V STOP, then

p(x1 . . . xn , y1 . . . yn+1 ) = q(D|∗, ∗) × q(N|∗, D) × q(V|D, N) × q(STOP|N, V)

×e(the|D) × e(dog|N) × e(laughs|V)

Note that this model form is a noisy-channel model. The quantity

q(D|∗, ∗) × q(N|∗, D) × q(V|D, N) × q(STOP|N, V)

is the prior probability of seeing the tag sequence D N V STOP, where we have
used a second-order Markov model (a trigram model), very similar to the language
models we derived in the previous lecture. The quantity

e(the|D) × e(dog|N) × e(laughs|V)

can be interpreted as the conditional probability p(the dog laughs|D N V STOP):

that is, the conditional probability p(x|y) where x is the sentence the dog laughs,
and y is the tag sequence D N V STOP.

6
4.2 Independence Assumptions in Trigram HMMs
We now describe how the form for trigram HMMs can be derived: in particular, we
describe the independence assumptions that are made in the model. Consider a pair
of sequences of random variables X1 . . . Xn , and Y1 . . . Yn , where n is the length
of the sequences. We assume that each Xi can take any value in a finite set V of
words. For example, V might be a set of possible words in English, for example
V = {the, dog, saw, cat, laughs, . . .}. Each Yi can take any value in a finite set K
of possible tags. For example, K might be the set of possible part-of-speech tags
for English, e.g. K = {D, N, V, . . .}.
The length n is itself a random variable—it can vary across different sentences—
but we will use a similar technique to the method used for modeling variable-length
Markov processes (see the previous lecture notes).
Our task will be to model the joint probability
P (X1 = x1 . . . Xn = xn , Y1 = y1 . . . Yn = yn )
for any observation sequence x1 . . . xn paired with a state sequence y1 . . . yn , where
each xi is a member of V, and each yi is a member of K.
We will find it convenient to define one additional random variable Yn+1 , which
always takes the value STOP. This will play a similar role to the STOP symbol seen
for variable-length Markov sequences, as described in the previous lecture notes.
The key idea in hidden Markov models is the following definition:
P (X1 = x1 . . . Xn = xn , Y1 = y1 . . . Yn+1 = yn+1 )
n+1
Y n
Y
= P (Yi = yi |Yi−2 = yi−2 , Yi−1 = yi−1 ) P (Xi = xi |Yi = yi ) (5)
i=1 i=1

where we have assumed that y0 = y−1 = *, where * is a special start symbol.

Note the similarity to our definition of trigram HMMs. In trigram HMMs we
have made the assumption that the joint probability factorizes as in Eq. 5, and in
addition we have assumed that for any i, for any values of yi−2 , yi−1 , yi ,
P (Yi = yi |Yi−2 = yi−2 , Yi−1 = yi−1 ) = q(yi |yi−2 , yi−1 )
and that for any value of i, for any values of xi and yi ,
P (Xi = xi |Yi = yi ) = e(xi |yi )
Eq. 5 can be derived as follows. First, we can write
P (X1 = x1 . . . Xn = xn , Y1 = y1 . . . Yn+1 = yn+1 )
= P (Y1 = y1 . . . Yn+1 = yn+1 )P (X1 = x1 . . . Xn = xn |Y1 = y1 . . . Yn+1 = yn+1 )
(6)

7
This step is exact, by the chain rule of probabilities. Thus we have decomposed
the joint probability into two terms: first, the probability of choosing tag sequence
y1 . . . yn+1 ; second, the probability of choosing the word sequence x1 . . . xn , con-
ditioned on the choice of tag sequence. Note that this is exactly the same type of
decomposition as seen in noisy channel models.
Now consider the probability of seeing the tag sequence y1 . . . yn+1 . We make
independence assumptions as follows: we assume that for any sequence y1 . . . yn+1 ,
n+1
Y
P (Y1 = y1 . . . Yn+1 = yn+1 ) = P (Yi = yi |Yi−2 = yi−2 , Yi−1 = yi−1 )
i=1
That is, we have assumed that the sequence Y1 . . . Yn+1 is a second-order Markov
sequence, where each state depends only on the previous two states in the sequence.
Next, consider the probability of the word sequence x1 . . . xn , conditioned on
the choice of tag sequence, y1 . . . yn+1 . We make the following assumption:
P (X1 = x1 . . . Xn = xn |Y1 = y1 . . . Yn+1 = yn+1 )
n
Y
= P (Xi = xi |X1 = x1 . . . Xi−1 = xi−1 , Y1 = y1 . . . Yn+1 = yn+1 )
i=1
Yn
= P (Xi = xi |Yi = yi ) (7)
i=1
The first step of this derivation is exact, by the chain rule. The second step involves
an independence assumption, namely that for i = 1 . . . n,
P (Xi = xi |X1 = x1 . . . Xi−1 = xi−1 , Y1 = y1 . . . Yn+1 = yn+1 ) = P (Xi = xi |Yi = yi )
Hence we have assumed that the value for the random variable Xi depends only on
the value of Yi . More formally, the value for Xi is conditionally independent of the
previous observations X1 . . . Xi−1 , and the other state values Y1 . . . Yi−1 , Yi+1 . . . Yn+1 ,
given the value of Yi .
One useful way of thinking of this model is to consider the following stochastic
process, which generates sequence pairs y1 . . . yn+1 , x1 . . . xn :
1. Initialize i = 1 and y0 = y−1 = *.
2. Generate yi from the distribution
q(yi |yi−2 , yi−1 )

3. If yi = STOP then return y1 . . . yi , x1 . . . xi−1 . Otherwise, generate xi from

the distribution
e(xi |yi ),
set i = i + 1, and return to step 2.

8
4.3 Estimating the Parameters of a Trigram HMM
We will assume that we have access to some training data. The training data con-
sists of a set of examples where each example is a sentence x1 . . . xn paired with a
tag sequence y1 . . . yn . Given this data, how do we estimate the parameters of the
model? We will see that there is a simple and very intuitive answer to this question.
Define c(u, v, s) to be the number of times the sequence of three states (u, v, s)
is seen in training data: for example, c(V, D, N) would be the number of times the
sequence of three tags V, D, N is seen in the training corpus. Similarly, define
c(u, v) to be the number of times the tag bigram (u, v) is seen. Define c(s) to be
the number of times that the state s is seen in the corpus. Finally, define c(s ; x)
to be the number of times state s is seen paired sith observation x in the corpus: for
example, c(N ; dog) would be the number of times the word dog is seen paired
with the tag N.
Given these definitions, the maximum-likelihood estimates are
c(u, v, s)
q(s|u, v) =
c(u, v)

and
c(s ; x)
e(x|s) =
c(s)
For example, we would have the estimates

c(V, D, N)
q(N|V, D) =
c(V, D)

and
c(N ; dog)
e(dog|N) =
c(N)
Thus estimating the parameters of the model is simple: we just read off counts
from the training corpus, and then compute the maximum-likelihood estimates as
described above.

4.4 Decoding with HMMs: the Viterbi Algorithm

We now turn to the problem of finding the most likely tag sequence for an input
sentence x1 . . . xn . This is the problem of finding

arg max p(x1 . . . xn , y1 . . . yn+1 )

y1 ...yn+1

9
where the arg max is taken over all sequences y1 . . . yn+1 such that yi ∈ K for
i = 1 . . . n, and yn+1 = STOP. We assume that p again takes the form
n+1
Y n
Y
p(x1 . . . xn , y1 . . . yn+1 ) = q(yi |yi−2 , yi−1 ) e(xi |yi ) (8)
i=1 i=1

Recall that we have assumed in this definition that y0 = y−1 = *, and yn+1 =
STOP.
The naive, brute force method would be to simply enumerate all possible tag
sequences y1 . . . yn+1 , score them under the function p, and take the highest scor-
ing sequence. For example, given the input sentence

the dog barks

and assuming that the set of possible tags is K = {D, N, V}, we would consider all
possible tag sequences:

D D D STOP
D D N STOP
D D V STOP
D N D STOP
D N N STOP
D N V STOP
...

and so on. There are 33 = 27 possible sequences in this case.

For longer sentences, however, this method will be hopelessly inefficient. For
an input sentence of length n, there are |K|n possible tag sequences. The expo-
nential growth with respect to the length n means that for any reasonable length
sentence, brute-force search will not be tractable.

4.4.1 The Basic Algorithm

Instead, we will see that we can efficiently find the highest probability tag se-
quence, using a dynamic programming algorithm that is often called the Viterbi
algorithm. The input to the algorithm is a sentence x1 . . . xn . Given this sentence,
for any k ∈ {1 . . . n}, for any sequence y1 . . . yk such that yi ∈ K for i = 1 . . . k
we define the function
k
Y k
Y
r(y1 . . . yk ) = q(yi |yi−2 , yi−1 ) e(xi |yi ) (9)
i=1 i=1

10
Other documents randomly have
different content
“It is the lady I met on the boat between Dover and Calais.
Her necklet had been stolen, and she was naturally in tears.
We travelled together from Calais to Paris,” he explained.
“She is a very intelligent English society woman, and I
asked her to call.”

The French secretary, who had been engaged at the

Ministry in Fez for some years, bowed as his new master
spoke.

In a few moments Freda Crisp, elegantly-dressed, swept

into the luxurious room.

“Ah! So here you are!” she cried in French, which she spoke
extremely well. “I promised I would call. Do you know, the
French police are so much cleverer than the English! They
have already arrested the thief and returned my necklet to
me!”

His Excellency, after inviting his guest to be seated,

expressed pleasure at the news, and then the secretary
rose discreetly and left.

“I hope you are enjoying Paris,” Freda said in her low

musical voice, which always charmed her dupes. “Now that
the autumn is coming on everyone is returning from
Deauville. I am giving a little party to-night at the Ritz. I
wonder if you would honour me with your presence? I have
a friend, an Englishman, who wishes very particularly to
make Your Excellency’s acquaintance.”

The old Minister expressed himself as being delighted,

whereupon she suggested that he should dine with her and
her English friend at the hotel. The old Moor with his
Eastern admiration of feminine beauty found her charming,
and at eight o’clock that night when he entered the hotel,
his striking figure in the ample white burnous (upon which
was the glittering star of the Order of the Tower and the
Sword), and turban, caused all heads to be turned in his
direction.

“This is my friend, Mr Arthur Porter,” Freda said. “Will Your

Excellency allow me to present him?”

The old Moor took Porter’s hand and, with an expression of

pleasure, the trio sat down to dinner at a corner table in the
great restaurant.

The Moorish Minister spent a most enjoyable evening, for

though he touched no wine, he was after dinner introduced
to several very elegant and charming women, both English
and French, for in a certain circle in Paris Freda was well
known. Porter took good care to ingratiate himself with the
patriarchal-looking old fellow, declaring that he knew
Morocco, was delighted with the life there, and intended in
a few weeks’ time to visit Fez again. The truth, however,
was that he had never been there in his life and had no
intention of ever going. Freda had followed the old Minister
from London and had managed to become acquainted with
him with the sole object of introducing Arthur Porter, alias
Bertram Harrison. To them both the death of Sir Charles
was known, and Porter guessed that Mr Sandys’ financial
position would be greatly affected. He had seen Sir Charles
at several gaming-tables, and knew that he had been a
reckless gambler. So cleverly did the pair play their cards
that Mohammed ben Mussa invited Porter to call and see
him next day—which he did.

As the two men sat together smoking cigarettes, Porter

suddenly said in French:

“I heard the other day that the ancient emerald mines in

the Wad Sus are about to be worked again.”
“That is so. I granted the concession in London only a few
days ago.”

“Ah! How very unfortunate!” remarked his visitor. “I have a

big financial backing, and could have exploited those mines
with huge profits to all of us. Of course, I do not know how
much gratification Your Excellency has received for the
concession, but my friends would, I believe, have paid Your
Excellency fifty thousand francs down and one-quarter of
the profits of the undertaking.” The old Moor pursed his lips
and pricked up his ears. From Barclay he had received
nothing on account, and only one-eighth share. Porter could
see that the old fellow was filled with regret and chagrin
that he had granted the concession with such little gain to
himself.

“His Majesty the Sultan demands a share in the profits,” old

Mohammed remarked. “He has been allotted an eighth
share—similar to myself.”

“I could have arranged a quarter share for you and an

eighth for His Majesty,” said the crafty Englishman quickly.
“But I suppose it is unfortunately too late, now that you
have given the concession into another quarter.”

Mohammed ben Mussa remained silent, slowly stroking his

long beard with his brown claw-like hand.

The Englishman’s offer was extremely tempting. He was

reflecting.

At last he said very slowly:

“Perhaps if seventy-five thousand francs were offered me

and the shares you suggest, I might find some way out,”
and he smiled craftily.
“Well,” said Porter with affected hesitation, “I’m inclined to
think that my friends would pay that sum—and at once if
they received an unassailable concession. I mean a
concession given to Mr Rex Rutherford under your hand and
seal as Minister which would cancel the previous one.”
Porter knew well the one power in Oriental countries was
that of backsheesh, and wrote down the name Rex
Rutherford.

“I will consider it,” said the old man. “There is no hurry till
to-morrow. I may find it necessary to telegraph to Fez. I—I
have to think it over, M’sieur Porter.”

“Of course. Then I will come here to-morrow—shall we say

at eleven? And you will afterwards lunch with me at Voisin’s
—eh?”

“It is agreed,” said the representative of the Moorish Sultan,

and then, after another cigarette, Porter rose and left,
walking back to the Place Vendôme to tell Freda the result
of his morning’s negotiations.

Next day, at noon when the tall Englishman entered

Mohammed’s room he saw by the expression on the old
man’s face that he had triumphed.

“I have been reflecting,” His Excellency said when his visitor

was seated, “and I have prepared a copy of the concession
which I gave in London, with the name and terms altered as
we discussed yesterday, and with the payment of seventy-
five thousand francs to me direct at latest to-morrow as
being the consideration. You see, it is all in order—a
concession in perpetuity granted to your nominee, Mr
Rutherford, and sealed by my Ministerial seal, which I hold
from His Majesty, and signed by myself. Please examine it.”
Arthur Porter took the document, which was almost a
replica of that handed to Barclay in London. The date was,
however, different, as well as the terms.

“Yes,” he said, after carefully reading the French translation.

“It all seems in order. It rescinds the previous concession
granted in London.”

“Most certainly. No one will have any authority to enter the

Wad Sus except yourself and those you appoint.”

With satisfaction Porter drew from his inner pocket an

envelope containing seventy-five one-thousand-franc notes,
which he counted out upon the table one by one.

The old Moor’s thin yellow fingers handled them gleefully,

and placing them together he drew them beneath his ample
burnous, saying quite coolly:

“I trust, Monsieur Porter, that you are satisfied.”

“Perfectly,” was the Englishman’s reply. “My friend will at

once form a syndicate and work the mines. Of course, we
may have trouble with that Mr Barclay in London.”

“He paid no consideration. Therefore you need not trouble

about him. The concession you have is the only valid one,
for it is dated after the one I gave in London. If they
attempt to enforce it we shall instantly prevent their
entering the district, or arrest them if they attempt to do
so.”

And the old man chuckled to himself at the easy manner in

which he had obtained seventy-five thousand francs.
 Chapter Twenty Three.
A Caller at the Rectory.

That morning Gordon Gray, dapper and well-dressed as

ever, had scanned the papers and read the report of the
inquiry into the death of Sir Charles Hornton. The coroner’s
jury had returned a verdict of “death through
misadventure,” it having been proved that Sir Charles had
mistaken a bottle of poison for a prescription for indigestion
which the local doctor had sent him on the previous day. In
fact, it was a not too rare way of hushing-up the suicide of a
well-known man. In many cases where persons of means
commit wilful suicide the twelve local tradesmen are lenient,
and declare it to be pure accident, or “misadventure”—
unless, of course, the suicide leaves a letter, in which case
the truth cannot be circumvented. For a suicide to leave a
letter is a criminal act towards his family.

Early in the afternoon the telephone-bell rang in the

pleasant sitting-room of the cosy West End chambers Gray
was occupying, and on taking off the receiver he heard
Freda speaking from Paris.

“All O.K.,” she said. “Guinness has got the concession and is
bringing it over this afternoon. He’ll be with you to-night.”

“When does the old Moor leave?” asked Gray.

“The day after to-morrow. He goes straight back to Tangier.”

“Right. Keep in touch with him till he’s safely away, then get
back here,” were the great crook’s orders.
Meanwhile events were following close upon each other in
those crowded autumn days.

Roddy, checkmated by his failure to find the girl Manners

who had written to his dead father from Bayeux, made, in
company with the shoe-repairer Nicole, a number of
inquiries of the commissary of police and in other quarters,
but in vain.

From the worthy pair he learnt how they had received the
young lady at St. Malo from an Englishman and a woman,
apparently his wife. From the description of the woman he
felt convinced that it was Freda Crisp. The girl, under the
influence of the same drug that had been administered to
him, had been smitten by temporary blindness, in addition
to her mind being deranged. Here was still more evidence of
the dastardly machinations of Gray and his unscrupulous
associates. It was now plain that the girl Manners had not
died, after all, but had lapsed into a kind of cataleptic state,
just as he had done.

The problem of her whereabouts, however, was an all-

important one. With her as witness against Gray and the
woman Crisp the unmasking of the malefactors would be an
easy matter. Besides, had not Mr Sandys told him that it
was most important to him that the young lady’s fate
should be ascertained?

What had been her fate? The description of the mysterious

man who called himself a doctor and who had recently
visited the poor girl conveyed nothing to Roddy. It seemed,
however, as though after she had written the letter to his
father she had suddenly disappeared. Had she left Bayeux
of her own accord, or had she been enticed away?

The police suspected foul play, and frankly told him so.
It was during those eager, anxious days in Bayeux that
Roddy, on glancing at Le Nouvelliste, the daily paper
published in Rennes, saw to his astonishment news of the
tragic death of Mr Sandys’ partner, and hastened to
telegraph his condolences. Hence it was with great surprise
that Elma and her father were aware that the young man
was in France, for the telegram simply bore the place of
origin as Bayeux.

Little did he dream of the clever devil’s work which Freda

and her associate Porter had accomplished with old
Mohammed ben Mussa, but remained in Normandy
following a slender clue, namely, a statement made by a
white-capped peasant woman hailing from the neighbouring
village of Le Molay-Littry, who declared that she had, on the
day of the young English mademoiselle’s disappearance,
seen her on the railway platform at Lison entering a train
for Cherbourg. She was alone. To Cherbourg Roddy
travelled, accompanied by a police-officer from Bayeux and
Monsieur Nicole, but though they made every inquiry, no
trace of her could be found. At the office of the
Southampton boats nobody recollected her taking a
passage on the day in question. Therefore, saddened and
disappointed, he was compelled to relinquish his search and
cross back to England.

While on board the boat he paced the deck much puzzled

how to act. He wondered how Elma was faring. Mr Sandys
was, no doubt, too full of his partner’s tragic end to attend
to any fresh business proposal. Therefore he decided not to
approach him at present with the concession, which was in
the vaults of the Safe Deposit Company.

On arrival at Victoria he, however, drove to Park Lane to

call, see Elma, and express to her father his regret at the
tragedy. The footman who opened the door answered that
neither his master nor Miss Elma was at home.

“Are they at Farncombe?” asked Roddy, much disappointed.

“No, sir. They are in town. But I do not think they will be
back till very late.”

Roddy, who was a shrewd observer, could tell that the man
had received orders to say “not at home.”

“Not at home” to him? Why? He stood upon the wide

doorstep filled with wonder and chagrin. He wanted to tell
Mr Sandys of the second disappearance of Edna Manners,
and most of all to see the girl he so fondly loved.

But she was “not at home.” What could be the reason of

such an attitude?

He took the last train home from Waterloo, and on arrival at

the Rectory—which he still occupied until the new
incumbent should require it—old Mrs Bentley came down to
let him in.

“Oh, sir,” she exclaimed, “I’m glad you’ve come back.

There’s been a young lady here this evening inquiring for
your poor father. I told her I expected you home every day,
and she’s coming again to-morrow evening at five o’clock.
After she went I saw her wandering about Welling Wood, as
though searching for something. She told me to say that
her name is Miss Manners.”

Roddy stood staggered—too amazed to utter a word for the

moment. Edna Manners had returned, and to-morrow he
would know the truth.
Too puzzled and excited to sleep, he threw off his coat, and
entering his wireless-room took up his cigar-box receiver
with the newly invented and super-sensitive crystal
detector. Placing the ’phones over his ears he switched on
the little portable aerial wire which he used with it and
attached another wire to earth, whereupon he heard loud
and strong telephony—somebody in Rotterdam testing with
a station in London and speaking in Dutch. It proved
beyond all doubt that the new crystal was the most
sensitive type known, and that, for a portable set, was of
far greater utility than vacuum valves. The quality of the
telephony, indeed, astounded him.

He had been listening in for nearly an hour when suddenly

he heard the voice of a fellow-experimenter, a man named
Overton, in Liverpool, with whom he often exchanged tests.

At once he threw over his transmission switch, the

generator hummed with gathering speed, and taking up the
telephone, he said:

“Hulloa, 3.B.L.! Hulloa, 3.B.L.! Hulloa, Liverpool! This is

Homfray 3.X.Q. calling. Your signals are very good.
Modulation excellent 3.B.L. I am just back from France, and
will test with you to-morrow night at 22:00 G.M.T. Did you
get that 3.B.L., Liverpool? 3.X.Q. over.” And he threw over
the switch, the humming of the generator dying down.

In a few seconds came Overton’s familiar voice, saying:

“Hulloa 3.X.Q.! This is 3.B.L. answering! Thanks very much

for your report. I will call you to-morrow night at 22:00
G.M.T. Thanks again. Somebody was calling you half an
hour ago on one thousand metres. You did not get him.
Better try now. G.N.O.M. (Good-night, old man.) 3.B.L.
switching off.”
Roddy, interested as to who, in the wonderful modern world
of wireless where men and women only meet through the
ether, could have called him, raised his receiving
wavelength to a thousand metres and listened.

Beyond some “harmonics” there was nothing. Suddenly,

however, an unknown voice, so clear and high-pitched that
it startled him, said:

“Hulloa, 3.X.Q.! Hulloa, Farncombe! I have called you

several times to-night; the last time an hour ago. I’m
speaking for Mr Barclay. He did not know that you were
back. He is coming on urgent business to Guildford to-
morrow. Can you meet him at the station at eleven o’clock
in the morning. He has asked me to give you that message.
This is 3.T.M. at Kingston-on-Thames speaking. 3.T.M. over.”

Roddy was not surprised. He frequently—in contravention of

the Post-Office regulations, be it said—received such
relayed messages. He could be with Barclay at eleven and
meet Edna Manners at five.

So putting in his transmission switch, which caused the big

vacuum globes to light up and the generator to hum again,
he took up the microphone transmitter, and replied in a
sharp clear voice:

“Hulloa, 3.T.M.! This is 3.X.Q. answering. Thank you very

much for the message from Barclay—I will keep the
appointment to-morrow. 3.X.Q. switching off.”

Why did Barclay wish to see him so urgently? Perhaps the

urgency had not occurred until the post-office had closed,
hence he had been unable to send a telegram. And at the
Rectory there was no telephone, save that splendidly
equipped radio-phone.
Little did Roddy Homfray suspect that Mr Purcell Sandys
was faced with ruin, that Elma knew of the impending
disaster, and that there was a reason—a very clear and
distinct reason—why she and her father were neither of
them “at home” when he had called.

Black ruin had fallen upon the great financial house of

Sandys and Hornton, a fact of which, though Roddy was in
ignorance, Gordon Gray, alias Rex Rutherford, and his
accomplices were well aware, and were about to turn to
their own advantage.
 Chapter Twenty Four.
Rutherford Makes a Proposition.

On that evening when Roddy was told that neither Mr

Sandys nor Elma was at home both father and daughter
were, as a matter of fact, seated together in the library. Mr
Sandys had by that time been able to ascertain pretty
nearly the extent of his firm’s liabilities, and was in
complete despair.

Elma was kneeling beside her father with her arm lovingly
around his neck, nobly trying to comfort him.

She had confessed her affection for Roddy, and had spoken
of the young man’s high hopes and aspirations, and shown
her father a hasty letter she had received from him
announcing the fact that the concession for emerald mining
had actually been granted to him by the Moorish Minister,
Mohammed ben Mussa.

A new thought arose in Mr Sandys’ mind. If Roddy had

really been granted the concession for the mines known to
exist there—and he had made some searching inquiries
during the past week or so—then by dealing with it he
might, after all, be able to raise sufficient money to
discharge part of the immense liabilities of the firm, and
thus stem the tide which must otherwise rise in the course
of the next few days and overwhelm him.

Elma’s father spoke quite openly concerning the situation.

“In that case Roddy could marry me, dad,” she said. “And
further, even if he had no concession, I am poor enough
now to marry a poor man,” she added.
“Yes, my child,” was his reply. “If what young Homfray says
is true then he can be the saviour of our firm and of our
family. I confess I have taken a great liking to the young
fellow. I have liked him all along.”

Then Elma flung herself into her father’s arms and kissed
him again and again, with tears of joy. Strangely enough
her father’s ruin had brought about her own happiness.

It was at that moment when the footman entered, and said:

“Mr Homfray has called, sir, and I told him that you were
not at home, as you ordered.”

Elma looked at her father dismayed.

“Has he gone?” she gasped, her face falling.

“Yes, miss. He called about five minutes ago.”

And then the man bowed and retired, while the girl, turning
to her father, remarked:

“How very unfortunate, dad! I wanted to tell him the good

news. But now it must wait until to-morrow. Good-night,
dad. Cheer up now, won’t you, dearest? This is a black
cloud, but it will pass, as all clouds pass sooner or later, and
the sun shines out again.” And kissing him the girl ran off
joyously to her own room.

Roddy rose early, as was his wont, and went into his
wireless-room, as was his habit each morning to listen to
the transatlantic messages, and those from Moscow, Nantes
and the rest. His eye rested upon the sensitive little set in
the cigar-box, and it occurred to him to test it that day as a
portable set in the train and elsewhere.
His train arrived at Guildford from Haslemere soon after ten
o’clock, therefore he left the station, and climbing the old
disused coach-road known as the Mount, reached the long
range of hills called the Hog’s Back. There, upon the wide
grass-grown road which has not been used for nearly a
century, he threw up his aerial wire into a high elm and
placing in position his ground wire soldered to a long steel
skewer he put on the telephones, holding the box in his left
hand while he turned the condensers with his right.

At once he heard the voice of the radio-telephone operator

at Croydon, the shrewd, alert expert with the rolling r’s,
calling Le Bourget. Signals were excellent. He listened for
ten minutes or so and then, drawing down his temporary
aerial and withdrawing the skewer from the wet earth, put
the cigar-box into the pocket of his raincoat and descended
the hill to the station.

Upon the platform he awaited the incoming train from

Waterloo, and was determined to be at home at five o’clock
to meet Edna Manners. The train arrived but without
Barclay, so he strolled out into the yard to await the next.

In the meantime, however, another striking incident was

happening at Park Lane.

Old Hughes, summoned to the door, opened it to the

smiling, well-dressed Mr Rex Rutherford.

“Will you tell Mr Sandys I’m here. And apologise for my

early call. I have come on rather pressing business,” he said
briskly.

“Very well, sir,” replied Lord Farncombe’s old butler rather

stiffly, taking his hat and umbrella, and asking him into the
library.
A couple of minutes later the bearded old financier entered
with outstretched hand, and smiling.

“I really must apologise, Mr Sandys,” Rutherford said. “It’s

awfully early, I know, but between business men the hour,
early or late, doesn’t really count—does it? At least, we say
so in New York.”

“I agree,” said Mr Sandys with a smile, and then when both

were seated, Rutherford said:

“I’ve come to you, Mr Sandys, with a very important

proposition—one in which you will at once see big money—
the concession for some ancient emerald mines in Morocco.”

“Do you mean the Wad Sus mines?” asked Sandys, much
surprised.

“Yes. I have arranged with my friend, His Excellency

Mohammed ben Mussa, the Moorish Minister of the Interior,
for a concession in perpetuity over the whole region,
subject to a payment on results to His Majesty the Sultan.”

“I really don’t understand you,” exclaimed Elma’s father,

looking straight in his face. “A concession has already been
granted to a young man of my acquaintance, Mr Homfray.”

“Not of the same mines—ancient ones, from which one big

dark-coloured emerald has quite recently been taken? That
can’t be?”

“But it is.”

“Have you seen this concession given to your friend, Mr

Homfray? I don’t know who he is, but I fear it is not worth
the paper it is written upon, because here I have a
concession which revokes all previous ones, and which will
make it penal for anyone who attempts to trespass as a
prospector in any part of the Wad Sus region! Here it is!
Look for yourself,” he said, taking the sealed document from
his pocket and handing it to the astonished financier. “Of
course,” he added, “if the affair is too small for your
attention, Mr Sandys, I can easily negotiate it elsewhere.
But as we are friends, I thought I would let you have its
refusal.”

Purcell Sandys was utterly staggered. He knew French well,

and at a glance he convinced himself that the document
was genuine.

“And not only have we the concession, but here also is a

plan of the exact situation of the mines, together with a
statement from one of the Touareg tribesmen, Ben Chaib
Benuis, with its French translation. The man, a trusted
messenger of the Moorish Government, has quite recently
been upon the spot, and has brought back a very large and
valuable emerald which is in the possession of an ex-
Moorish official at Tangier, and can be seen any day.”

Mr Sandys scanned the French translation and sat back in

wonder.

It was quite evident that the concession granted to young

Homfray—if there had ever been one—was worthless, for
there was the sealed document dated only a few days
before which rescinded every other grant made by the
Moorish Government.

“I, of course, know nothing of your friend Mr Homfray,”

remarked Rutherford. “But I fear that if he attempts to
prospect in the Wad Sus he will be at once arrested. I alone
hold the only concession in that district,” and slowly picking
up both the formidable-looking documents, he carefully
refolded them and replaced them in his pocket.

“Well, Mr Rutherford,” said the pale, thoughtful old financier

at last. “I confess I am very much puzzled, and before
entering upon this affair as a matter of business I would
first like to look into young Homfray’s claims.”

“Very naturally,” laughed the easy-going Rutherford. “I

should do so myself in the circumstances. I fear, however,
that the young man, whoever he is, has somewhat misled
you. I’ll look in and see you to-morrow morning—about this
time—eh?” he added as he rose and left, while Mr Sandys
sat speechless and puzzled.

When Rutherford had gone he called Elma and told her of

his visit.

“What? That man here again?” cried the girl. “He can’t have
any valid concession. Roddy has it. He would never write a
lie to me!”

“My child, we can do nothing until we see and question

young Homfray.”

“You are right, dad. I’ll try at once to get hold of him. He is
probably at Farncombe. I’ll telephone to the Towers and tell
Bowyer to go to the Rectory at once.”

This she did, but half an hour later the reply came back.
The maid Bowyer had been to the Rectory, but Mr Homfray
was out and would not return till five o’clock. She had left a
message from Elma asking him to go to London at once.

At five o’clock Mrs Bentley at the Rectory opened the door

to Edna Manners, but Roddy had not returned. For an hour
she waited, idling most of the time in the garden. Then at
last she asked leave to write him a note, which she did in
the dead rector’s study, and then reluctantly left.

The evening passed until at half-past nine a man from the

Towers called to ask again for Roddy, but Mrs Bentley
repeated that her young master had gone out that morning
and had not yet returned. This report was later repeated to
Elma over the telephone from the Towers to Park Lane.

Meanwhile Mr Sandys telegraphed to the Minister

Mohammed ben Mussa in Tangier, asking for confirmation of
Mr Rutherford’s concession, and just before midnight came
a reply that the concession had been granted to Mr Rex
Rutherford.

Elma’s father showed her the reply. All Roddy’s assertions

were false! All her hopes were crushed. She burst into tears
and fled to her room.

Mr Sandys, left alone, faced the situation calmly. The only

way to stave off ruin would be to deal with Rutherford.

Meanwhile the master criminal was playing a clever double

game.

When he called next morning he asked to see Elma,

pleading that he had something very important to say to
her. When Hughes brought the message she was at first
reluctant to accede to his wish, but in a few moments she
steeled herself and walked to the morning-room into which
he had been shown.

As usual, he was smartly-groomed and the essence of

politeness. As he took her hand, he said:

“Miss Elma, I want to tell you that I sympathise very much

with your father in his great misfortune, the secret of which
I happen to know—though as yet the world suspects
nothing. But I fear it soon will, unless your father can come
forward with some big and lucrative scheme. I have it in my
power to help him with the mining concession in Morocco. I
will do so on one condition.”

“And what is that, Mr Rutherford?” she asked quite calmly.

He looked straight into her big, wide-open eyes and, after a

second’s pause, replied:

“That I may be permitted to pay my attentions to you—for I

confess that I love you.”

The girl’s cheeks coloured slightly and the expression in her

eyes altered.

“That cannot be,” she said. “I am already engaged.”

“To that young fellow Homfray, I believe?” he laughed. “Has

he not already misled you and your father into believing
that he is a rich man, inasmuch that he pretends to have
been granted some worthless concession also in Morocco?
Surely such a man is not suited to you as a husband, Miss
Elma? Could you ever trust him?”

“I will not have Mr Homfray’s character besmirched in my

presence, Mr Rutherford,” she said haughtily. “And if this is
the matter upon which you wished to speak with me I
should prefer that you said nothing further.”

“Elma! I love you!” he cried, with openly sensual

admiration.

The girl was horrified and revolted. She told him so, but he
treated with a conqueror’s contempt her frightened
attempts to evade him. She was to be his toy, his plaything
—or he would not lift a finger to save her father.

On her part she pleaded her love for Roddy, but he told her
brutally that the young fellow was a liar. Why had he not
produced the concession he alleged he had?

A last Elma, compelled to listen to his specious arguments,

almost gave up hope, but before leaving the room she
declared that she would starve rather than marry him. And
then she closed the door after her.

Ten minutes later Rutherford was shown into the library,

and in his most oleaginous manner greeted the ruined
financier.

“I have called to keep my appointment, Mr Sandys,” he

said. “But since I saw you circumstances have altered
somewhat, which makes it incumbent upon me to place the
concession elsewhere.”

“Why?” asked Sandys, his face falling. “Well, it is a private

matter. I—I really don’t care to discuss it, Mr Sandys.
Indeed, I think it is best for me to say that our negotiations
must conclude here, even though I regret it very deeply. It
is not my fault, but the—well, the barrier—lies in another
direction.”

“In what direction?” asked the grey-bearded man who had

been clutching at the straw offered him on the previous day.

“Well—if you ask Miss Elma, your daughter, she will

explain.”

“My daughter? What has she to do with our propositions?”

“I simply repeat my reply, Mr Sandys. I can’t say more. To
tell the truth, I don’t feel capable. I must go now. If you
want to see me later you know my telephone number.”

And taking his hat, he stalked out of the fine library, well
knowing himself to be the conqueror. To those who are
patient and painstaking the fruits of the world will arrive.
But there are exceptions, even though the devil controls his
own.

When Elma’s father sought her he found her in a paroxysm

of tears and tried to comfort her. She had thrown herself on
a couch at the foot of her bed and was sobbing out her
heart.

The ruined man told his daughter of Rutherford’s visit, and

asked her for the explanation which he had said that she
alone could give.

In a few halting sentences she related what had happened.

For some time the old man remained silent, standing at the
great window past which the motor-’buses were passing up
and down London’s street of the wealthy.

“Ah! my dear!” he sighed. “I am sorry that you have so

unfortunately fallen in love with young Homfray. At first I
liked him, I confess. But he seems to have sadly misled
you, and is now afraid to face the truth.”

“I agree, father. But I love him. There is some explanation,

I feel sure.”

“There can be none regarding the emerald concession.

Rutherford has it, as well as the plan showing the
whereabouts of the mine. I could float a big company to-
morrow, even upon the concession and the official plan
furnished by the Moorish Minister of the Interior. But he
has, alas! now withdrawn his offer.”

“Because I have refused him,” said Elma bitterly. “I love

Roddy. How could I possibly become that man’s wife?”

Her father drew a long breath and shrugged his shoulders.

He stood with his back towards her, looking idly out upon
the traffic in Park Lane and the Park beyond.

“Yes, darling,” he said at last. “But you must not sacrifice

yourself for me. It would be grossly unfair. I am ruined
through no fault of my own, I trust—ruined by a gambling
partner who cared for nothing save his obsession with
regard to games of chance. Let us say no more about it.
Rutherford may take his concession elsewhere, and I will
face the music. I have my comfort in my Yogi teaching—in
those two words ‘I am.’ I have done my best in life, and to
my knowledge have never injured anyone. I have tried to
act up to my Yogi teachers, with their magnificent
philosophy of the East. Therefore I will face disaster
unflinchingly.”

And seeing his daughter in tears, his further words were

choked by emotion. He merely patted her upon the shoulder
and, unable to bear the interview longer, withdrew.

For a fortnight past Rex Rutherford, like many crooks of his

calibre, had actually engaged a “Press agent”—one of those
parasites who fasten themselves upon the ambitious and
put forward lies and photographs to the Press at so many
guineas a time. The crook, in the financial Press, read of his
own wonderful financial operations in Paris and in New York,
reports which were calculated to raise him in the estimation
of the great house of Sandys and Hornton. The City had
read of Rex Rutherford day after day, and there were
rumours of a great scheme he had for a new electric tube
rail system for the outer suburbs of Paris, for which he was
negotiating with the French Government.

Purcell Sandys had read all this—a Press campaign which

had cost the master criminal a mere three hundred pounds.
With that sum he had established a reputation in the
financial papers. Editors of newspapers cannot always
exclude the “puff paragraphs” when they are cleverly
concealed by a master of that craft. And it often takes even
a shrewd sub-editor to detect the gentle art of self-
advertisement.

That afternoon the old financier walked alone through the

Park as far as Kensington Gardens and back. He knew that
the crash must come at latest in a day or two, and Sandys
and Hornton must suspend payment.

There was no way out.

 Chapter Twenty Five.
The Sacrifice.

For Elma the world held no future. Though surrounded by

every luxury in that magnificent Park Lane mansion, the
millionaire home that was the most notable in all London’s
modern houses, her only thoughts were of her father and of
her lover Roddy.

She hated that fat, beady-eyed but elegantly-dressed man

whom Mr Harrison had introduced to her father, and who
was now so openly making love to her. His words and his
manner were alike artificial. The feminine mind is always
astute, and she knew that whatever he said was mere
empty compliment. She saw upon his lips the sign of
sensuousness, a sign that no woman fails to note.
Sensuousness and real love are things apart, and every
woman can discriminate them. Men are deceivers. Women
may, on the other hand, allure, and be it said that the
vampire woman like Freda Crisp is ever with us.

In the life of London, of Paris, or of New York, the vampire

woman in society plays a part which is seldom suspected.

They are in a class by themselves, as was Freda Crisp. The

vampire woman is the popular term for a woman who lives
by preying upon others; men usually, but upon her own sex
if occasion demands.

Freda Crisp, though few of the characters in this human

drama of love and cupidity had suspected her, was a case in
point. She was a type that was interesting. As a girl of
eighteen everyone admired her for her charm of manner,
her conversational gifts and her bright intellect, which was
marred only by a rather too lively imagination, and a
tendency to romance so ingeniously that no one ever knew
if she told the truth or not.

Her career was abnormal, and yet not stranger than that of
some others in these post-war days.

At nineteen she had been to prison for swindling. Physically

she was wonderfully fascinating, but her chief characteristic
was an absence of all real affection and moral feeling. Even
as a girl she could profess passionate love for those from
whom she expected profit and gain; but misfortune and
death, even of those nearest her, would leave her quite
unmoved.

She was a perfect type of the modern adventuress. She

could act well, and at times would shed tears profusely if
she thought it the right thing at the moment.

As she grew older her unrestrained coquetry threw her into

the vicious adventurous circle of which Gordon Gray was the
master and moving spirit. She threw in her lot with him. On
board a transatlantic liner on which she went for a trip to
New York an officer fell a victim to her charms, and supplied
her with money that was not his. His defalcations were
discovered, and he committed suicide to escape disgrace.

That was the first unpleasant incident in her career after

meeting Gray. There were many afterwards. She was a
woman whose sole aim was to see and enjoy life. Without
heart and without feeling, active, not passive in her love-
making, she, like many another woman before her, aspired
to power and influence over men, and many an honourable
career was wrecked by her, and much gain had gone into
the joint pockets of Gordon Gray and herself.

Purcell Sandys had been ruined. She knew it, and laughed.
Welcome to our website – the perfect destination for book lovers and
knowledge seekers. We believe that every book holds a new world,
offering opportunities for learning, discovery, and personal growth.
That’s why we are dedicated to bringing you a diverse collection of
books, ranging from classic literature and specialized publications to
self-development guides and children's books.

More than just a book-buying platform, we strive to be a bridge

connecting you with timeless cultural and intellectual values. With an
elegant, user-friendly interface and a smart search system, you can
quickly find the books that best suit your interests. Additionally,
our special promotions and home delivery services help you save time
and fully enjoy the joy of reading.

Join us on a journey of knowledge exploration, passion nurturing, and

personal growth every day!

ebookbell.com